Details of Previous Robotics Colloquia

Go back to current colloquia page.

 
Spring 2017
03/31/2017
CSE 305
Richard Vaughan
Simon Fraser University
Simple, Robust Interaction Between Humans and Teams of Robots
Abstract: Sensing technology for robots has improved dramatically in the last few years, but we do not see robots around us yet. How should robots behave around people and each other to get things done? My group works on behavioural strategies for mobile robots that exploit the new sensing capabilities, and allows them to perform sophisticated, robust interactions with the world and other agents. I’ll show videos of a series of novel vision-mediated Human-Robot Interactions with teams of driving and flying robots. At their best, the robots work like those in sci-fi movies. Others need more work, but the robots are always autonomous, the humans are uninstrumented, the interactions surprisingly simple, and we often work outdoors over long distances.

Speaker’s Bio: Richard Vaughan directs the Autonomy Lab at Simon Fraser University. His research interests include long-term autonomous robots, multi-robot systems, behavioural ecology, human-robot interaction (HRI), and robotics software. He demonstrated the first robot to control animal behaviour in 1998, co-created the Player/Stage Project in 2000, and recently showed the first uninstrumented HRI with UAVs. He currently serves on the Administrative Committee of the IEEE Robotics and Automation Society, and the editorial board of the Autonomous Robots journal, and is the Program Chair for IROS 2017.
04/07/2017
CSE 305
Oussama Khatib
Stanford University
Ocean One: A Robotic Avatar for Oceanic Discovery
Abstract: The promise of oceanic discovery has intrigued scientists and explorers for centuries, whether to study underwater ecology and climate change, or to uncover natural resources and historic secrets buried deep at archaeological sites. The quest to explore the ocean requires skilled human access. Reaching these depth is imperative since factors such as pollution and deep-sea trawling increasingly threaten ecology and archaeological sites. These needs demand a system deploying human-level expertise at the depths, and yet remotely operated vehicles (ROVs) are inadequate for the task. A robotic avatar could go where humans cannot, while embodying human intelligence and intentions through immersive interfaces. To meet the challenge of dexterous operation at oceanic depths, in collaboration with KAUST’s Red Sea Research Center and MEKA Robotics, we developed Ocean One, a bimanual force-controlled humanoid robot that brings immediate and intuitive haptic interaction to oceanic environments. Teaming with the French Ministry of Culture’s Underwater Archaeology Research Department, we deployed Ocean One in an expedition in the Mediterranean to Louis XIV’s flagship Lune, lying off the coast of Toulon at ninety-one meters. In the spring of 2016, Ocean One became the first robotic avatar to embody a human’s presence at the seabed. This expedition demonstrated synergistic collaboration between a robot and a human operating over challenging manipulation tasks in an inhospitable environment. Tasks such as coral-reef monitoring, underwater pipeline maintenance, and offshore and marine operations will greatly benefit from such robot capabilities. Ocean One’s journey in the Mediterranean marks a new level of marine exploration: Much as past technological innovations have impacted society, Ocean One’s ability to distance humans physically from dangerous and unreachable work spaces while connecting their skills, intuition, and experience to the task promises to fundamentally alter remote work. We foresee that robotic avatars will search for and acquire materials in hazardous and inhospitable settings, support equipment at remote sites, build infrastructure for monitoring the environment, and perform disaster prevention and recovery operations— be it deep in oceans and mines, at mountain tops, or in space.

Speaker’s Bio: Oussama Khatib received his PhD from Sup’Aero, Toulouse, France, in 1980. He is Professor of Computer Science at Stanford University. His research focuses on methodologies and technologies in human-centered robotics including humanoid control architectures, human motion synthesis, interactive dynamic simulation, haptics, and human-friendly robot design. He is a Fellow of IEEE. He is Co-Editor of the Springer Tracts in Advanced Robotics (STAR) series and the Springer Handbook of Robotics, which received the PROSE Award for Excellence in Physical Sciences & Mathematics. Professor Khatib is the President of the International Foundation of Robotics Research (IFRR). He has been the recipient of numerous awards, including the IEEE RAS Pioneer Award in Robotics and Automation, the IEEE RAS George Saridis Leadership Award in Robotics and Automation, the IEEE RAS Distinguished Service Award, and the Japan Robot Association (JARA) Award in Research and Development.
04/14/2017
CSE 305
Debadeepta Dey
Microsoft Research
Learning via Interaction for Machine Perception and Control
Abstract: As autonomous robots of all shapes and sizes proliferate in the world and start working in increasing proximity to humans it is critical that they produce safe intelligent behavior while efficiently learning from limited interactions in such computationally constrained regimes. A reoccurring problem is considering a limited number of actions from a very large number of possible actions. Examples include grasp selection in robotic manipulation, where the robot arm must evaluate a sequence of grasps with the aim of finding one which is successful as early on in the sequence as possible, or trajectory selection for mobile ground robots, where the task is to select a sequence of trajectories from a much larger set of feasible trajectories for minimising expected cost of traversal. A learning algorithm must therefore be able to predict a budgeted number of decisions which optimises a utility function of interest. Traditionally machine learning has focused on producing a single best prediction. We build an efficient framework for making multiple predictions where the objective is to optimise any utility function which is (monotone) submodular over a sequence of predictions. For each of these cases we optimise for the content and order of the sequence. We demonstrate the efficacy of these methods on several real world robotics problems. Another closely related problem is the budgeted information gathering problem, where a robot with a fixed fuel budget is required to maximise the amount of information gathered from the world, appears in practice across a wide range of applications in autonomous exploration and inspection with mobile robots. We present an efficient algorithm that trains a policy on the target distribution to imitate a clairvoyant oracle - an oracle that has full information about the world and computes non-myopic solutions to maximise information gathered. Additionally, our analysis provides theoretical insight into how to efficiently leverage imitation learning in such settings. Our approach paves the way forward for efficiently applying data-driven methods to the domain of information gathering.

Speaker’s Bio: Debadeepta Dey is a researcher in the Adaptive Systems and Interaction (ASI) group at Microsoft Research, Redmond, USA. He received his doctorate in 2015 at the Robotics Institute, Carnegie Mellon University, Pittsburgh, USA, where he was advised by Prof. J. Andrew (Drew) Bagnell. He does fundamental as well as applied research in machine learning, control and computer vision motivated by robotics problems. He is especially interested in bridging the gap between perception and planning for autonomous ground and aerial vehicles. His interests include decision-making under uncertainty, reinforcement learning, and machine learning. From 2007 to 2010 he was a researcher at the Field Robotics Center, Robotics Institute, Carnegie Mellon University.
04/21/2017
CSE 303 (11 AM)
Eric Eaton
University of Pennsylvania
Efficient Lifelong Machine Learning: an Online Multi-Task Learning Perspective
Abstract: Lifelong learning is a key characteristic of human intelligence, largely responsible for the variety and complexity of our behavior. This process allows us to rapidly learn new skills by building upon and continually refining our learned knowledge over a lifetime of experience. Incorporating these abilities into machine learning algorithms remains a mostly unsolved problem, but one that is essential for the development of versatile autonomous systems. In this talk, I will present our recent progress in developing algorithms for lifelong machine learning for classification, regression, and reinforcement learning, including applications to optimal control for robotics. These algorithms approach the problem from an online multi-task learning perspective, acquiring knowledge incrementally over consecutive learning tasks, and then transferring that knowledge to rapidly learn to solve new tasks. Our approach is highly efficient, scaling to large numbers of tasks and amounts of data, and provides a variety of theoretical guarantees. I will also discuss our work toward autonomous cross-domain transfer between diverse tasks, and zero-shot transfer learning from task descriptions.

Speaker’s Bio: Eric Eaton is a non-tenure-track faculty member in the Department of Computer and Information Science at the University of Pennsylvania, and a member of the GRASP (General Robotics, Automation, Sensing, & Perception) lab. Prior to joining Penn, he was a visiting assistant professor at Bryn Mawr College, a senior research scientist at Lockheed Martin Advanced Technology Laboratories, and part-time faculty at Swarthmore College. His primary research interests lie in the fields of machine learning, artificial intelligence, and data mining with applications to robotics, environmental sustainability, and medicine.
04/21/2017
CSE 305
Katsu Ikeuchi
Microsoft Research
e-Heritage Project Part I: e-Intangible Heritage, from Dancing robots to Cyber Humanities
Abstract: Tangible heritage, such as temples and statues, is disappearing day-by- day due to human and natural disaster. In-tangible heritage, such as folk dances, local songs, and dialects, has the same story due to lack of inheritors and mixing cultures. We have been developing methods to preserve such tangible and in-tangible heritage in the digital form. This project, which we refer to as e-Heritage, aims not only record heritage, but also analyze those recorded data for better understanding as well as display those data in new forms for promotion and education. This talk, the first talk of e-Heritage Project, covers our effort for handling in- tangible heritage. We are developing a method to preserve folk dances by the performance of dancing robots. Here, we follow the paradigm, learning-from- observation, in which a robot learns how to perform a dance from observing a human dance performance. Due to the physical difference between a human and a robot, the robot cannot exactly mimic the human actions. Instead, the robot first extracts important actions of the dance, referred to key poses, and then symbolically describes them using Labanotation, which the dance community has been using for recording dances. Finally, this labanotation is mapped to each different robot hardware for reconstructing the original dance performance. The second part tries to answer the question, what is the merit to preserve folk dances by using robot performance by the answer that such symbolic representations for robot performance provide new understandings of those dances. In order to demonstrate this point, we focus on folk dances of native Taiwanese, which consists of 14 different tribes. We have converted those folk dances into Labanotation for robot performance. Further, by analyzing these Labanotations obtained, we can clarify the social relations among these 14 tribes.

Speaker’s Bio: Dr. Katsushi Ikeuchi is a Principal Researcher of Microsoft Research Asia, stationed at Microsoft Redmond campus. He received a Ph.D. degree in Information Engineering from the University of Tokyo in 1978. After working at Artificial Intelligence Lab of Massachusetts Institute of Technology as a pos-doc fellows for three years, Electrotechnical Lab of Japanese Government as a researcher for five years, Robotics Institute of Carnegie Mellon University as a faculty member for ten years, Institute of Industrial Science of the University of Tokyo as a faculty member for nineteen years, he joined Microsoft Research Asia in 2015. His research interest spans computer vision, robotics, and computer graphics. He has received several awards, including IEEE-PAMI Distinguished Researcher Award, the Okawa Prize from the Okawa foundation, and Si-Ju- Ho-Sho (the Medal of Honor with Purple ribbon) from the Emperor of Japan. He is a fellow of IEEE, IEICE, IPSJ, and RSJ.
04/28/2017
CSE 305
Henrik Christensen
UC San Diego
Object Based Mapping
Abstract: To build mobile systems that can operate autonomously it is necessary to endow them with a sense of location. One of the basic aspects of autonomy is the ability to not get lost. How can we build robots that acquire a model of the surrounding world and utilize these models to achieve their mission without getting lost along the way. Simultaneous Localization and Mapping (SLAM) is widely used to provide the mapping and localization compete to robots. The process has three facets: extraction of features from sensor data, association of features with prior detected structures and estimation of position/pose and update of the map to make it current. The estimation part of the process is today typically performed using graphical models to allow for efficient computations and enable flexible handling of ambiguous situations. Over time the feature extraction has matured from use of basic features such as lines and corners to utilization of significant structures such as man-made objects (building, chairs, tables, cars, ...) that are easily identifiable. The discriminative nature of major structures simplifies data-association and facilities more efficient loop-closing. In this presentation we will discuss our modular mapping framework - OmniMapper - and how it can be utilized across a range of different applications for efficient computing. We will discuss a number of different strategies for object detection and pose estimation and also provide examples of mapping across a number of different sensory modalities. Finally we will show a number of examples of use of the OmniMapper across in- and out-door settings using air and ground vehicles.

Speaker’s Bio: Dr. Henrik I. Christensen is a Professor in Dept. of Computer Science and Engineering, UC San Diego. He is also the director of the Institute for Contextual Robotics. Prior to UC San Diego he was the founding director of Institute for Robotics and Intelligent machines (IRIM) at Georgia Institute of Technology (2006-2016). Dr. Christensen does research on systems integration, human-robot interaction, mapping and robot vision. The research is performed within the Cognitive Robotics Laboratory. He has published more than 350 contributions across AI, robotics and vision. His research has a strong emphasis on "real problems with real solutions". A problem needs a theoretical model, implementation, evaluation, and translation to the real world. He is actively engaged in the setup and coordination of robotics research in the US (and worldwide). Dr. Christensen received the Engelberger Award 2011, the highest honor awarded by the robotics industry. He was also awarded the "Boeing Supplier of the Year 2011". Dr. Christensen is a fellow of American Association for Advancement of Science (AAAS) and Institute of Electrical and Electronic Engineers (IEEE). He received an honorary doctorate in engineering from Aalborg University 2014. He collaborates with institutions and industries across three continents. His research has been featured in major media such as CNN, NY Times, and BBC.
05/5/2017
CSE 305
Alberto Rodriguez
MIT
Reactive Robotic Manipulation
Abstract: The main goal of this talk is to motivate the need for feedback control and contact sensing in robotic grasping and manipulation. I'll start by briefing on recent work by team MIT-Princeton in the Amazon Robotics Challenge, and the lack of practical solutions that exploit feedback and contact sensing. Some of the key challenges to control contact interaction are hybridness, underactuation, and an effective use of tactile sensing. I’ll discuss these challenges in the context of the pusher-slider system, a classical simple problem where the purpose is to control the motion of an object sliding on a flat surface. I like to think of the pusher-slider problem as playing a role in robotic manipulation analogous to the inverted pendulum in classical control. It incorporates many of the challenges present in robotic manipulation tasks: noisy friction, instability, hybridness and underactuation. I will finish by discussing ongoing work and future directions in my group exploring strategies for real-time state estimation and control through frictional intermittent contact.

Speaker’s Bio: Alberto Rodriguez is the Walter Henry Gale (1929) Career Development Professor at the Mechanical Engineering Department at MIT. Alberto graduated in Mathematics ('05) and Telecommunication Engineering ('06) from the Universitat Politecnica de Catalunya (UPC) in Barcelona, and earned his PhD in Robotics (’13) from the Robotics Institute at Carnegie Mellon University. He spent a year in the Locomotion group at MIT, and joined the faculty at MIT in 2014, where he started the Manipulation and Mechanisms Lab (MCube). Alberto is the recipient of the Best Student Paper Awards at conferences RSS 2011 and ICRA 2013 and Best Paper finalist at IROS 2016. His main research interests are in robotic manipulation, mechanical design, and automation.
05/19/2017
CSE 305
Charlie Kemp
Georgia Tech
Mobile Manipulators for Intelligent Physical Assistance
Abstract: Since founding the Healthcare Robotics Lab at Georgia Tech 10 years ago, my research has focused on developing mobile manipulators for intelligent physical assistance. Mobile manipulators are mobile robots with the ability to physically manipulate their surroundings. They offer a number of distinct capabilities compared to other forms of robotic assistance, including being able to operate independently from the user, being appropriate for users with diverse needs, and being able to assist with a wide variety of tasks, such as object retrieval, hygiene, and feeding. We’ve worked with hundreds of representative end users - including older adults, nurses, and people with severe motor impairments - to better understand the challenges and opportunities associated with this technology. Among other points, I’ll provide evidence for the following assertions: 1) many people will be open to assistance from mobile manipulators; 2) assistive mobile manipulation at home is feasible for people with severe motor impairments using conventional interfaces; and 3) permitting contact and intelligently controlling forces increases the effectiveness of mobile manipulators. I’ll conclude with a brief overview of some of our most recent research.

Speaker’s Bio: Charles C. Kemp (Charlie) is an Associate Professor at the Georgia Institute of Technology in the Department of Biomedical Engineering with adjunct appointments in the School of Interactive Computing and the School of Electrical and Computer Engineering. He earned a doctorate in Electrical Engineering and Computer Science (2005), an MEng, and BS from MIT. In 2007, he joined the faculty at Georgia Tech where he directs the Healthcare Robotics Lab ( http://healthcare-robotics.com ). He is an active member of Georgia Tech’s Institute for Robotics & Intelligent Machines (IRIM) and its multidisciplinary Robotics Ph.D. program. He has received a 3M Non-tenured Faculty Award, the Georgia Tech Research Corporation Robotics Award, a Google Faculty Research Award, and an NSF CAREER award. He was a Hesburgh Award Teaching Fellow in 2017. His research has been covered extensively by the popular media, including the New York Times, Technology Review, ABC, and CNN.
05/19/2017
CSE 305
Karol Hausman
University of Southern California
Rethinking the Action Perception Loops
Abstract: While perception has traditionally served action in robotics, it has been argued for some time that intelligent action generation can benefit perception, and carefully coupling perception with action can improve the performance of both. In this talk, I will report on recent progress in model-based and learning-based approaches that address aspects of the problem of closing perception-action loops. The first part of my talk will focus on a model-based, active perception technique that optimizes trajectories for self-calibration. This method takes into account motion constraints and produces an optimal trajectory that yields fast convergence of estimates of the self-calibration states and other user-chosen states. In the second part of my talk, I will present a deep reinforcement learning framework that learns manipulation skills on a real robot in a reasonable amount of time. The method handles contact and discontinuities in dynamics by combining the efficiency of model-based techniques and the generality of model-free reinforcement learning techniques.

Speaker’s Bio: Karol Hausman is a Ph.D. student at the University of Southern California in the Robotic Embedded Systems Lab under the supervision of Prof. Gaurav S. Sukhatme. His research interests lie in the field of interactive perception, reinforcement learning, and state estimation in robotics. He received his B.E. and M.E. degrees in Mechatronics from the Warsaw University of Technology, Poland, in 2010 and 2012, respectively. In 2013 he graduated with an M.Sc. degree in Robotics, Cognition and Intelligence from the Technical University Munich. During his Ph.D., he interned with Bosch Research Center Palo Alto, NASA JPL, Qualcomm Research and he will join Google DeepMind for an internship this Summer.
05/26/2017
CSE 305
Silvia Ferrari
University of Southern California
Neuromorphic Planning and Control of Insect-scale Robots
Abstract: Recent developments in neural stimulation and recording technologies are providing scientists with the ability of recording and controlling the activity of individual neurons in vitro or in vivo, with very high spatial and temporal resolution. Tools such as optogenetics, for example, are making a significant impact in the neuroscience field by delivering optical firing control with the precision and resolution required for investigating information processing and plasticity in biological brains. This talk presents a spike-based training approach that is realizable in vitro or in vivo via neural stimulation and recording technologies, such as optogenetics and multielectrode arrays, and can be utilized to control synaptic plasticity in live neuronal circuits as well as neuromorphic circuits such as CMOS memristors. The approach is demonstrated by training a computational SNN to control the locomotion of a virtual cockroach in an unknown environment, and to stabilize the flight of the RoboBee in the presence of partially unmodeled dynamics.

Speaker’s Bio: Silvia Ferrari is a Professor of Mechanical and Aerospace Engineering at Cornell University. Prior to that, she was Professor of Engineering and Computer Science, and Founder and Director of the NSF Integrative Graduate Education and Research Traineeship (IGERT) and Fellowship program on Wireless Intelligent Sensor Networks (WISeNet) at Duke University. She is the Director of the Laboratory for Intelligent Systems and Controls (LISC), and her principal research interests include robust adaptive control of aircraft, learning and approximate dynamic programming, and optimal control of mobile sensor networks. She received the B.S. degree from Embry-Riddle Aeronautical University and the M.A. and Ph.D. degrees from Princeton University. She is a senior member of the IEEE, and a member of ASME, SPIE, and AIAA. She is the recipient of the ONR young investigator award (2004), the NSF CAREER award (2005), and the Presidential Early Career Award for Scientists and Engineers (PECASE) award (2006).
 
Winter 2017
01/13/2017
CSE 305
Matt Rueben
Oregon State University
Privacy-Sensitive Robotics
Abstract: Privacy is important for everybody, and the advent of robotics technology poses new and unique privacy concerns that need to be addressed. This talk introduces "privacy-sensitive robotics," the growing study of privacy issues and how to address them for useful and acceptable human-robot interactions. Two studies will be presented. The first looks at interfaces for specifying what's private to a robot. Physical and persistent interfaces are tested alongside a traditional point-and-click setup in a live user study with a real PR2 robot and realistic office environment. The second study is about the effect of context on people's privacy concerns. Subjects responded to a video of a telepresence robot in their home; we measured the impact of wrapping this scenario in different interpretive frames. It will become clear that privacy-sensitive robotics is a largely untouched research area despite deep privacy literature in neighboring fields. Much more research is needed soon to match the pace of technological development.

Speaker’s Bio: Matthew Rueben is a Robotics PhD candidate in the Personal Robotics group at Oregon State University. He earned the H.B.S. Degree in Mechanical Engineering, also from Oregon State University, in 2013. Matt's current research centers on privacy concerns about mobile robots. This includes the factors that determine people's privacy concerns about robots, what objects and places people consider private, and how they might communicate these preferences to robots.
02/17/2017
CSE 305
Sonia Chernova
Georgia Institute of Technology
Reliable Robot Autonomy through Learning and Interaction
Abstract: Robotics is undergoing an exciting transition from factory automation to the deployment of autonomous systems in less structured environments, such as warehouses, hospitals and homes. One of the critical barriers to the wider adoption of autonomous robotic systems in the wild is the challenge of achieving reliable autonomy in complex and changing human environments. In this talk, I will discuss ways in which innovations in learning from demonstration and remote access technologies can be used to develop and deploy autonomous robotic systems alongside and in collaboration with human partners. I will present applications of this research paradigm to robot learning, object manipulation and semantic reasoning, as well as explore some exciting avenues for future research in this area.

Speaker’s Bio: Sonia Chernova is the Catherine M. and James E. Allchin Early-Career Assistant Professor in the School of Interactive Computing at Georgia Tech, where she directs the Robot Autonomy and Interactive Learning research lab. Her research spans semantic reasoning, human-robot interaction, interactive machine learning and cloud robotics, with the focus on developing robots that are able to effectively operate in human environments. She is the recipient of the NSF CAREER, ONR Young Investigator and NASA Early Career Faculty awards.
 
Autumn 2016
10/07/2016
CSE 305
David Remy
University of Michigan
Gaits and Natural Dynamics in Robotic Legged Locomotion
Abstract: My research seeks to systematically exploit mechanical dynamics to make future robots faster, more efficient, and more agile then today’s kinematically controlled systems. Drawing inspiration from biology and biomechanics, I design and control robots whose motion emerges in great part passively from the interaction of inertia, gravity, and elastic oscillations. Energy is stored and returned periodically in springs and other dynamic elements, and continuous motion is merely initiated and shaped through the active actuator inputs. In this context, I am particularly interested in questions of gait selection. Should a legged robot use different gaits at different desired speeds? If so, what constitutes these gaits and what causes their existence? How do they relate to gaits observed in biology? We study these questions in conceptual models, in hardware implementations, and through biomechanical experiments. In the long term, this research will allow the development of systems that reach and even exceed the agility of humans and animals. It will enable us to build autonomous robots that can run as fast as a cheetah and as enduring as a husky, while mastering the same terrain as a mountain goat. And it will provide us with novel designs for prosthetics, orthotics, and active exoskeletons that help restoring the locomotion skills of the disabled and can be used as training and rehabilitation devices for the injured.

Speaker’s Bio: C. David Remy, is Assistant Professor of Mechanical Engineering at the University of Michigan, Ann Arbor. He received his Ph.D. from ETH Zurich (Prof. Roland Siegwart), and holds both a M.Sc. in Mechanical Engineering from the University of Wisconsin and a Diploma in Engineering Cybernetics from the University of Stuttgart. Dr. Remy is the head of the Robotics and Motion Laboratory. His research interests include the design, simulation, and control of legged robots, exoskeletons, and other nonlinear systems. Drawing inspiration from biology and biomechanics, he is particularly interested in the effects and exploitation of natural dynamic motions, the role of different gaits, and the possibility of force/torque controllable systems; both in conceptual models and in hardware realizations.
10/28/2016
CSE 305
Emo Todorov
University of Washington
Goal-directed Dynamics
Abstract: We develop a general control framework where a low-level optimizer is built into the robot dynamics. This optimizer together with the robot constitute a goal-directed dynamical system, which is controlled on a higher level. The high-level command is a cost function. It can encode desired accelerations, end-effector poses, center of pressure, and many other intuitive features that have been studied before. Unlike the currently popular quadratic programming framework, which comes with performance guarantees at the expense of modeling flexibility, the optimization problem we solve at each time step is non-convex and non-smooth. Nevertheless, by exploiting the unique properties of the soft-constraint physics model we have recently developed, we design an efficient optimizer for goal-directed dynamics. This new computational infrastructure can facilitate tele-operation, feature-based control, deep learning of control policies, and trajectory optimization. It will become a standard feature in future releases of the MuJoCo simulator.
11/04/2016
CSE 305
Sean Andrist
Microsoft Research
Gaze Mechanisms for Situated Interaction with Embodied Agents
Abstract: In this talk, I will discuss research I conducted for my PhD thesis on how embodied agents--both virtual agents and physical robots--can achieve positive social and communicative outcomes through the use of situated gaze mechanisms. I will discuss why social gaze is one of the most important nonverbal cues to consider for interactive embodied agents and present several projects I have carried out to design and test models of gaze behavior for agents in various contexts. These projects include (1) how agents can produce gaze shifts that target specific high-level interaction outcomes, (2) how agents can effectively utilize gaze aversions in conversation, (3) how agents can coordinate their gaze with the user’s gaze while collaborating on a physical task, and (4) how agents can adapt their gaze behaviors to the personality of their users for rehabilitation. I will also discuss current research directions into situated interaction with robots that I am now pursuing at Microsoft Research.

Speaker’s Bio: Sean Andrist is a researcher at Microsoft Research in the Adaptive Systems and Interaction group. His research interests involve designing, building, and evaluating socially interactive technologies that are physically situated in the open world. He recently received his PhD from the Department of Computer Sciences at the University of Wisconsin–Madison, where he conducted research on gaze mechanisms for the development of communicative characters, including both embodied virtual agents and social robots.
11/18/2016
CSE 305
Nicholas Roy
MIT
Planning to Fly (and Drive) Aggressively
Abstract: Getting a small unmanned aircraft to fly aggressively and autonomously through an unknown, cluttered environment creates substantial challenges for the vehicle's navigation and control. Without a prior map, the vehicle has to detect obstacles and avoid them, often on the basis of little sensor data, and make rapid decisions about how to move around given an uncertain and incomplete model of the world and the vehicle's position. I will discuss some recent results from my group in developing approximate inference and planning algorithms that have enabled fast and aggressive autonomous motion for unmanned vehicles in the air and on the ground.

Speaker’s Bio: Nicholas Roy is an Associate Professor in the Department of Aeronautics & Astronautics at the Massachusetts Institute of Technology and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT. He received his Ph. D. in Robotics from Carnegie Mellon University in 2003. His research interests include unmanned aerial vehicles, autonomous systems, human-computer interaction, decision-making under uncertainty and machine learning. He spent two years at Google [x] as the founder of Project Wing.
 
Spring 2016
4/1/16 Tomás Lozano-Pérez
MIT
Integrated task and motion planning in belief space

Abstract:

This talk describe an integrated strategy for planning, perception, state-estimation and action in complex mobile manipulation domains based on planning in the belief space of probability distributions over states, using hierarchical goal regression (pre-image back-chaining). We develop a vocabulary of logical expressions that describe sets of belief states, which are goals and subgoals in the planning process. We show that a relatively small set of symbolic operators can give rise to task-oriented perception in support of the manipulation goals. An implementation of this method is demonstrated in simulation and on a real PR2 robot, showing robust, flexible solution of mobile manipulation problems with multiple objects and substantial uncertainty.

This is joint work with Leslie Pack Kaelbling.

Bio:

Tomás Lozano-Pérez is currently the School of Engineering Professor in Teaching Excellence at MIT, where he is a member of the Computer Science and Artificial Intelligence Laboratory. His research has been in robot motion planning, computer vision, machine learning, medical imaging and computational chemistry. He received his degrees from MIT (SB 73, SM 77, PhD 80). He was a recipient of a 1985 Presidential Young Investigator Award and of the 2011 IEEE Robotics Pioneer Award. He is a Fellow of the AAAI and of the IEEE.

4/8/16 Henny Admoni
Yale / CMU
Recognizing Human Intent for Assistive Robotics

Abstract:

Assistive robots provide direct, personal help to people to address specific human needs. The mechanisms by which assistive robots provide help can vary widely. Socially assistive robots act as tutors, coaches, or therapy aides to shape human behavior through social interaction. In contrast, physically assistive robots help people through direct manipulation of their environment. While these different types of assistance involve different robot functions, there exist underlying principles that remain constant across all assistive human-robot interactions. For example, robots must be able to recognize people’s goals and intentions in order to assist them, whether that assistance is social or physical.

Identifying human intentions can be challenging, because the mapping from observed human behavior back to the underlying goals and beliefs which generated that behavior if often unclear. However, we can take advantage of findings from psychology, which show that people actually project their intentions in natural and often subconscious ways through their nonverbal behavior, such as eye gazes and gestures.

In this talk, I describe how we can extract human intent from behavior so that robots can assist people in accomplishing their goals. I discuss research across the socially and physically assistive domains, from autonomous robots designed to teach and collaborate with humans on a building task, to a robot arm operated through shared control that helps people with mobility impairments manipulate their environment. Throughout the talk, I show how nonverbal behavior can be incorporated into these systems to improve their understanding of human intentions, which leads to more effective assistance.

Bio:

Henny Admoni is a postdoctoral fellow at the Robotics Institute at Carnegie Mellon University, where she works on assistive robotics and human-robot interaction with Siddhartha Srinivasa in the Personal Robotics Lab. Henny develops and studies intelligent robots that improve people's lives by providing assistance through social and physical interactions. Henny completed her PhD in Computer Science at Yale University with Brian Scassellati. Her PhD dissertation was about modeling the complex dynamics of nonverbal behavior for socially assistive human-robot interaction. Henny also holds an MS in Computer Science from Yale University, and a BA/MA joint degree in Computer Science from Wesleyan University. Henny's scholarship has been recognized with awards such as the NSF Graduate Research Fellowship, the Google Anita Borg Memorial Scholarship, and the Palantir Women in Technology Scholarship.

4/11/16 Wolfram Burgard
University of Freiburg
Deep Learning for Robot Navigation and Perception

Abstract:

Autonomous robots are faced with a series of learning problems to optimize their behavior. In this presentation I will describe recent approaches developed in my group based on deep learning architectures for object recognition and body part segmentation from RGB(-D) images. In addition, I will present a terrain classification approach that utilizes sound. For all approaches I will describe expensive experiments quantifying in which way the corresponding algorithm extends the state of the art.

Bio:

Wolfram Burgard is a professor for computer science at the University of Freiburg and head of the research lab for Autonomous Intelligent Systems. His areas of interest lie in artificial intelligence and mobile robots. His research mainly focuses on the development of robust and adaptive techniques for state estimation and control. Over the past years Wolfram Burgard and his group have developed a series of innovative probabilistic techniques for robot navigation and control. They cover different aspects such as localization, map-building, SLAM, path-planning, exploration, and several other aspects. Wolfram Burgard coauthored two books and more than 300 scientific papers. In 2009, Wolfram Burgard received the Gottfried Wilhelm Leibniz Prize, the most prestigious German research award. In 2010, Wolfram Burgard received an Advanced Grant of the European Research Council. Since 2012, Wolfram Burgard is the coordinator of the Cluster of Excellence BrainLinks-BrainTools funded by the German Research Foundation. Wolfram Burgard is Fellow of the ECCAI, the AAAI, and the IEEE.

4/22/16 Travis Deyle
Cobalt Robotics
RFID-Enhanced Robots Enable New Applications in Healthcare, Asset Tracking, and Remote Sensing

Abstract:

Mounting long-range RFID readers to mobile robots or drones permits them to opportunistically relocate antennas to a virtually-infinite number of unique vantage points -- including hard-to-reach locations. Robotics researchers are employing mobile readers in unstructured environments (offices, homes, and outdoors) to make great strides in robotics, strides which would otherwise be extremely difficult or impossible without the use of long-range RFID tags. Examples include: Taking inventory and locating tagged assets in homes in lieu of perfect visual object recognition; fetching and retrieving tagged objects for older adults; and using drones to obtain remote sensor measurements from "sensorized" tags and performing tasks such as soil moisture sensing, remote crop monitoring, water quality monitoring, remote sensor deployment, and infrastructure monitoring of buildings, bridges, and dams.

Bio:

Travis Deyle is an expert in passive UHF RFID systems and their applications in healthcare and robotics. He earned a PhD from Professor Charlie Kemp's Healthcare Robotics Lab at Georgia Tech; he worked on "cyborg dragonflies" as a Postdoc at Duke University under Professor Matt Reynolds; and he worked on new, non-public projects within Google[x] Life Sciences (now Verily Life Sciences) alongside the team that developed the glucose-sensing "smart contact lens." He is now the co-founder and CEO of a stealthy robotics startup named Cobalt Robotics.

5/6/16 Brian Scassellati
Yale
Robots That Teach

Abstract:

Robots have long been used to provide assistance to individual users through physical interaction, typically by supporting direct physical rehabilitation or by providing a service such as retrieving items or cleaning floors. Socially assistive robotics (SAR) is a comparatively new field of robotics that focuses on developing robots capable of assisting users through social rather than physical interaction. Just as a good coach or teacher can provide motivation, guidance, and support without making physical contact with a student, socially assistive robots attempt to provide the appropriate emotional, cognitive, and social cues to encourage development, learning, or therapy for an individual.

In this talk, I will review some of the reasons why physical robots rather than virtual agents are essential to this effort, highlight some of the major research issues within this area, and describe some of our recent results building supportive robots for teaching 1st graders about nutrition, helping 2nd graders struggling to learn English as a second language, coaching 3rd graders on how to deal with bullies, and practicing social skills with children with autism spectrum disorder.

Bio:

Brian Scassellati is a Professor of Computer Science, Cognitive Science, and Mechanical Engineering at Yale University and Director of the NSF Expedition on Socially Assistive Robotics. His research focuses on building embodied computational models of human social behavior, especially the developmental progression of early social skills. Using computational modeling and socially interactive robots, his research evaluates models of how infants acquire social skills and assists in the diagnosis and quantification of disorders of social development (such as autism).

5/13/16 University of Washington ICRA 2016 Practice Talks

University of Washington papers at ICRA 2016: (In order of program)

Leah Perlmutter, Alex Fiannaca, Sahil Anand, Lindsey Arnold, Eric Kernfeld, Kimyen Truong, Akiva Notkin, and Maya Cakmak. Teaching English through Conversational Robotic Agents

Igor Mordatch, Nikhil Mishra, Clemens Eppner, and Pieter Abbeel. Combining Model-Based Policy Search with Online Model Learning for Control of Physical Humanoids

Vikash Kumar, Emanual Todorov, and Sergey Levine. Optimal Control with Learned Local Models: Application to Dexterous Manipulation

Luis Puig and Kostas Daniilidis. Monocular 3D Tracking of Deformable Surfaces

Yuyin Sun and Dieter Fox. NEOL: Toward Never-Ending Object Learning for Robots

Zhe Xu and Emanuel Todorov. Design of a Highly Biomimetic Anthropomorphic Robotic Hand towards Artificial Limb Regeneration

Muneaki Miyasaka, Mohammad Haghighipanah, Yangming Li, and Blake Hannaford. Hysteresis Model of Longitudinally Loaded Cable for Cable Driven Robots and Identification of the Parameters

Yangming Li, Muneaki Miyasaka, Mohammad Haghighipanah, and Blake Hannaford. Dynamic Modeling of Cable Driven Elongated Surgical Instruments for Sensorless Grip Force Estimation

Mohammad Haghighipanah, Muneaki Miyasaka, Yangming Li, and Blake Hannaford. Unscented Kalman Filter and 3D Vision to Improve Cable Driven Surgical Robot Joint Angle Estimation

Sarah Elliott, Michelle Valente, and Maya Cakmak. Making Objects Graspable in Confined Environments through Push and Pull Manipulation with a Tool

6/3/16 Siddhartha Srinivasa Physics-based Manipulation

Humans effortlessly push, pull, and slide objects, fearlessly reconfiguring clutter, and using physics and the world as a helping hand. But most robots treat the world like a game of pick-up-sticks: avoiding clutter and attempting to rigidly grasp anything they want to move. I'll talk about some of our ongoing efforts at harnessing physics for nonprehensile manipulation, and the challenges of deploying our algorithms on real physical systems. I'll specifically focus on whole-arm manipulation, state estimation for contact manipulation, and on closing the feedback loop on nonprehensile manipulation.

Bio:

Siddhartha Srinivasa is the Finmeccanica Associate Professor at The Robotics Institute at Carnegie Mellon University. He works on robotic manipulation, with the goal of enabling robots to perform complex manipulation tasks under uncertainty and clutter, with and around people. To this end, he founded and directs the Personal Robotics Lab, and co-directs the Manipulation Lab. He has been a PI on the Quality of Life Technologies NSF ERC, DARPA ARM-S and the CMU CHIMP team on the DARPA DRC.Sidd is also passionate about building end-to-end systems (HERB, ADA, HRP3, CHIMP, Andy, among others) that integrate perception, planning, and control in the real world. Understanding the interplay between system components has helped produce state of the art algorithms for object recognition and pose estimation (MOPED), and dense 3D modeling (CHISEL, now used by Google Project Tango). Sidd received a B.Tech in Mechanical Engineering from the Indian Institute of Technology Madras in 1999, an MS in 2001 and a PhD in 2005 from the Robotics Institute at Carnegie Mellon University. He played badminton and tennis for IIT Madras, captained the CMU squash team, and likes to run ultra marathons.

6/10/16 Ashish Kapoor
Microsoft Research
Planetary Scale Swarm Sensing, Planning and Control for Weather Prediction

Abstract:

Weather forecasting is a canonical predictive challenge that relies on extensive data gathering operations. We explore new directions with forecasting weather as a data-intensive challenge that involves large-scale sensing of the required information via planning and control of a swarm of aerial vehicles. First, we will demonstrate how commercial aircraft can be used to sense the current weather conditions at a continental scale and help us create Bayesian deep-hybrid predictive model for weather forecasts. Beyond making predictions, these probabilistic models can provide the guidance of sensing with value-of-information analyses, where we consider uncertainties and needs of sets of routes and maximize information value in light of the costs of acquiring data from a swarm of sensors. The methods can be used to select ideal subsets of locations to sense and also to evaluate the value of trajectories of flights for sensing. Finally, we will discuss how to carry out such large sensing missions using novel algorithms for robot planning under uncertainty.

Bio:

Ashish Kapoor is a senior researcher at Microsoft Research, Redmond. His recent research focuses on machine learning with applications to controls and planning of aerial vehicles. In the past he has worked in many different areas that include quantum machine learning, computer vision, affective computing and human-computer-interaction. Ashish received a PhD at the MIT Media Lab in 2006 and prior to that graduated from Indian Institute of Technology, Delhi.

 
Winter 2016
03/11/16 Daniel Butler
UW CSE
Model-based Reinforcement Learning with Parametrized Physical Models and Optimism-Driven Exploration
Chris Xie, Sachin Patil, Teodor Moldovan Sergey, Levine Pieter Abbeel

Abstract: In this paper, we present a robotic model-based reinforcement learning method that combines ideas from model identification and model predictive control. We use a featurebased representation of the dynamics that allows the dynamics model to be fitted with a simple least squares procedure, and the features are identified from a high-level specification of the robot’s morphology, consisting of the number and connectivity structure of its links. Model predictive control is then used to choose the actions under an optimistic model of the dynamics, which produces an efficient and goal-directed exploration strategy. We present real time experimental results on standard benchmark problems involving the pendulum, cartpole, and double pendulum systems. Experiments indicate that our method is able to learn a range of benchmark tasks substantially faster than the previous best methods. To evaluate our approach on a realistic robotic control task, we also demonstrate real time control of a simulated 7 degree of freedom arm.
03/04/16 James Youngquist
UW CSE
DeepMPC: Learning Deep Latent Features for Model Predictive Control
Ian Lenz, Ross Knepper, and Ashutosh Saxena

Abstract: Designing controllers for tasks with complex nonlinear dynamics is extremely challenging, time-consuming, and in many cases, infeasible. This difficulty is exacerbated in tasks such as robotic food-cutting, in which dynamics might vary both with environmental properties, such as material and tool class, and with time while acting. In this work, we present DeepMPC, an online real-time model-predictive control approach designed to handle such difficult tasks. Rather than hand-design a dynamics model for the task, our approach uses a novel deep architecture and learning algorithm, learning controllers for complex tasks directly from data. We validate our method in experiments on a large-scale dataset of 1488 material cuts for 20 diverse classes, and in 450 real-world robotic experiments, demonstrating significant improvement over several other approaches.
02/26/16 Justin Huang
UW CSE
Place Recognition with ConvNet Landmarks: Viewpoint-Robust, Condition-Robust, Training-Free
Niko Sunderhauf, Sareh Shirazi, Adam Jacobson, Feras Dayoub, Edward Pepperell, Ben Upcroft, and Michael Milford

Abstract: Place recognition has long been an incompletely solved problem in that all approaches involve significant compromises. Current methods address many but never all of the critical challenges of place recognition – viewpoint-invariance, condition-invariance and minimizing training requirements. Here we present an approach that adapts state-of-the-art object proposal techniques to identify potential landmarks within an image for place recognition. We use the astonishing power of convolutional neural network features to identify matching landmark proposals between images to perform place recognition over extreme appearance and viewpoint variations. Our system does not require any form of training, all components are generic enough to be used off-the-shelf. We present a range of challenging experiments in varied viewpoint and environmental conditions. We demonstrate superior performance to current state-of-theart techniques. Furthermore, by building on existing and widely used recognition frameworks, this approach provides a highly compatible place recognition system with the potential for easy integration of other techniques such as object detection and semantic scene interpretation.
02/19/16 Daniel Gordon
UW CSE
Deep Neural Decision Forests
Peter Kontschieder, Madalina Fiterau, Antonio Criminisi, Samuel Rota Bulo

Abstract: We present Deep Neural Decision Forests – a novel approach that unifies classification trees with the representation learning functionality known from deep convolutional networks, by training them in an end-to-end manner. To combine these two worlds, we introduce a stochastic and differentiable decision tree model, which steers the representation learning usually conducted in the initial layers of a (deep) convolutional network. Our model differs from conventional deep networks because a decision forest provides the final predictions and it differs from conventional decision forests since we propose a principled, joint and global optimization of split and leaf node parameters. We show experimental results on benchmark machine learning datasets like MNIST and ImageNet and find onpar or superior results when compared to state-of-the-art deep models. Most remarkably, we obtain Top5-Errors of only 7.84%/6.38% on ImageNet validation data when integrating our forests in a single-crop, single/seven model GoogLeNet architecture, respectively. Thus, even without any form of training data set augmentation we are improving on the 6.67% error obtained by the best GoogLeNet architecture (7 models, 144 crops).
02/12/16 Harley Montgomery
UW CSE
End-to-End Training of Deep Visuomotor Policies
Sergey Levine, Chelsea Finn, Trevor Darrell, Pieter Abbeel

Abstract: Policy search methods can allow robots to learn control policies for a wide range of tasks, but practical applications of policy search often require hand-engineered components for perception, state estimation, and low-level control. In this paper, we aim to answer the following question: does training the perception and control systems jointly end-to-end provide better performance than training each component separately? To this end, we develop a method that can be used to learn policies that map raw image observations directly to torques at the robot's motors. The policies are represented by deep convolutional neural networks (CNNs) with 92,000 parameters, and are trained using a partially observed guided policy search method, which transforms policy search into supervised learning, with supervision provided by a simple trajectory-centric reinforcement learning method. We evaluate our method on a range of real-world manipulation tasks that require close coordination between vision and control, such as screwing a cap onto a bottle, and present simulated comparisons to a range of prior policy search methods.
02/05/16 Aaron Walsman
UW CSE
Mastering the game of Go with deep neural networks and tree search
David Silver, Aja Huang et al.

Abstract: The game of Go has long been viewed as the most challenging of classic games for artificial intelligence owing to its enormous search space and the difficulty of evaluating board positions and moves. Here we introduce a new approach to computer Go that uses ‘value networks’ to evaluate board positions and ‘policy networks’ to select moves. These deep neural networks are trained by a novel combination of supervised learning from human expert games, and reinforcement learning from games of self-play. Without any lookahead search, the neural networks play Go at the level of stateof-the-art Monte Carlo tree search programs that simulate thousands of random games of self-play. We also introduce a new search algorithm that combines Monte Carlo simulation with value and policy networks. Using this search algorithm, our program AlphaGo achieved a 99.8% winning rate against other Go programs, and defeated the human European Go champion by 5 games to 0. This is the first time that a computer program has defeated a human professional player in the full-sized game of Go, a feat previously thought to be at least a decade away.
01/29/16 Zachary Nehrenberg
UW CSE
Real-Time Trajectory Generation for Quadrocopters
Markus Hehn, and Raffaello D’Andrea

Abstract: This paper presents a trajectory generation algorithm that efficiently computes high-performance flight trajectories that are capable of moving a quadrocopter from a large class of initial states to a given target point that will be reached at rest. The approach consists of planning separate trajectories in each of the three translational degrees of freedom, and ensuring feasibility by deriving decoupled constraints for each degree of freedom through approximations that preserve feasibility. The presented algorithm can compute a feasible trajectory within tens of microseconds on a laptop computer; remaining computation time can be used to iteratively improve the trajectory. By replanning the trajectory at a high rate, the trajectory generator can be used as an implicit feedback law similar to model predictive control. The solutions generated by the algorithm are analyzed by comparing them with time-optimal motions, and experimental results validate the approach.
01/29/16 Patrick Lancaster
UW CSE
Towards Learning Hierarchical Skills for Multi-Phase Manipulation Tasks
Oliver Kroemer, Christian Daniel, Gerhard Neumann, Herke van Hoof, and Jan Peters

Abstract: Most manipulation tasks can be decomposed into a sequence of phases, where the robot’s actions have different effects in each phase. The robot can perform actions to transition between phases and, thus, alter the effects of its actions, e.g. grasp an object in order to then lift it. The robot can thus reach a phase that affords the desired manipulation. In this paper, we present an approach for exploiting the phase structure of tasks in order to learn manipulation skills more efficiently. Starting with human demonstrations, the robot learns a probabilistic model of the phases and the phase transitions. The robot then employs model-based reinforcement learning to create a library of motor primitives for transitioning between phases. The learned motor primitives generalize to new situations and tasks. Given this library, the robot uses a value function approach to learn a high-level policy for sequencing the motor primitives. The proposed method was successfully evaluated on a real robot performing a bimanual grasping task.
01/22/16 Tanner Schmidt
UW CSE
Pose Estimation of Kinematic Chain Instances via Object Coordinate Regression
Frank Michel, Alexander Krull, Eric Brachmann, Michael Ying Yang, Stefan Gumhold, Carsten Rother


Abstract: In this paper, we address the problem of one shot pose estimation of articulated objects from an RGB-D image. In particular, we consider object instances with the topology of a kinematic chain, i.e. assemblies of rigid parts connected by prismatic or revolute joints. This object type occurs often in daily live, for instance in the form of furniture or electronic devices. Instead of treating each object part separately we are using the relationship between parts of the kinematic chain and propose a new minimal pose sampling approach. This enables us to create a pose hypothesis for a kinematic chain consisting of K parts by sampling K 3D-3D point correspondences. To asses the quality of our method, we gathered a large dataset containing four objects and 7000+ annotated RGB-D frames1 . On this dataset we achieve considerably better results than a modified state-of-the-art pose estimation system for rigid objects

Related papers:
6-DOF Model Based Tracking via Object Coordinate Regression
Alexander Krull, Frank Michel, Eric Brachmann, Stefan Gumhold, Stephan Ihrke, Carsten Rother

Learning 6D Object Pose Estimation using 3D Object Coordinates
Eric Brachmann, Alexander Krull, Frank Michel, Stefan Gumhold, Jamie Shotton, and Carsten Rother

01/15/16 Kendall Lowrey
UW CSE
Combining the benefits of function approximation and trajectory optimization

Abstract: Neural networks have recently solved many hard problems in Machine Learning, but their impact in control remains limited. Trajectory optimization has recently solved many hard problems in robotic control, but using it online remains challenging. Here we leverage the high-fidelity solutions obtained by trajectory optimization to speed up the training of neural network controllers. The two learning problems are coupled using the Alternating Direction Method of Multipliers (ADMM). This coupling enables the trajectory optimizer to act as a teacher, gradually guiding the network towards better solutions. We develop a new trajectory optimizer based on inverse contact dynamics, and provide not only the trajectories but also the feedback gains as training data to the network. The method is illustrated on rolling, reaching, swimming and walking tasks.
01/15/16 Vladimir Korukov
UW CSE
Information-Theoretic Planning with Trajectory Optimization for Dense 3D Mapping

Abstract: We propose an information-theoretic planning approach that enables mobile robots to autonomously construct dense 3D maps in a computationally efficient manner. Inspired by prior work, we accomplish this task by formulating an information-theoretic objective function based on CauchySchwarz quadratic mutual information (CSQMI) that guides robots to obtain measurements in uncertain regions of the map. We then contribute a two stage approach for active mapping. First, we generate a candidate set of trajectories using a combination of global planning and generation of local motion primitives. From this set, we choose a trajectory that maximizes the information-theoretic objective. Second, we employ a gradientbased trajectory optimization technique to locally refine the chosen trajectory such that the CSQMI objective is maximized while satisfying the robot’s motion constraints. We evaluated our approach through a series of simulations and experiments on a ground robot and an aerial robot mapping unknown 3D environments. Real-world experiments suggest our approach reduces the time to explore an environment by 70% compared to a closest frontier exploration strategy and 57% compared to an information-based strategy that uses global planning, while simulations demonstrate the approach extends to aerial robots with higher-dimensional state.
FALL 2015
10/9/15 Dan Bohus
Microsoft Research
Physically Situated Dialog: Opportunities and Challenges

Abstract: Most research to date on spoken language interaction has focused on supporting dialog with single users in limited domains and contexts. Efforts in this space have led to significant progress, including wide-scale deployments of telephony-based systems and voice-enabled mobile assistants. At the same time, numerous and important challenges in the realm of physically situated, open-world interaction have remained largely unaddressed, e.g., language interaction with robots in a public space, in-car systems, ambient assistants, etc. In this talk, I will give an overview of the Situated Interaction project at Microsoft Research, which aims to address some of these challenges. Specifically, I will outline a core set of communicative competencies required for supporting dialog in physically situated settings – such as models of multiparty engagement, turn-taking and interaction planning, and I will present samples of our work as part of a broader research agenda in this space.

Speaker’s Bio: Dan Bohus is a Senior Researcher in the Adaptive Systems and Interaction Group at Microsoft Research. His research agenda is focused on physically situated, open-world spoken language interaction. Before joining Microsoft Research, Dan has received his Ph.D. degree (2007) in Computer Science from Carnegie Mellon University.
 
 
10/16/15 Sawyer Fuller
UW
Aerial autonomy at insect scale: What flying insects can tell us about robotics and vice versa

Abstract: Insect-sized aerial robots will be deployed where their small size, low cost, and maneuverability give them an advantage over larger robots. For example, they could deploy in swarms to follow airborne plumes to locate gas leaks in dense piping infrastructure. However, miniaturization poses challenges because the physics of scaling dictates that many technologies used in larger aircraft cannot operate effectively at the size of insects. These include propellers, the Global Positioning System, and general-purpose microprocessors. Insects have overcome these challenges by evolving a scale-appropriate flight apparatus whose robustness and agility surpasses anything man-made. For example, using only senses carried onboard, they can land on flowers buffeted by wind or deftly avoid a flyswatter. But how they do this is not fully understood. My research aims to better understand insect capabilities through experimental study and use this to create autonomous robot counterparts with competitive performance. I will describe experiments I performed on flies that revealed that they sense and compensate for wind to improve flight agility. And I will describe flight control demonstrations on a fly-sized flapping-wing robot stabilized by insect-inspired sensors. The results indicate that, under the severe power and weight constraints and unfamiliar physics at this scale, success will require designs with intimately coupled sensing and mechanics, new low-power control architectures, and careful observation of the techniques used by biology.
 
 
10/23/15 Russ Tedrake
MIT
From Polynomials to Humanoid Robots

Abstract: The last few years have seen absolutely incredible advances in the field of robotics, with massive new investments from major companies including Google, Apple, and Uber. At the heart of these advances are algorithms, often using mathematical optimization, which allow our machines to better interpret massive streams of incoming data, to decide how and where to move, and even to balance and not fall down while they are executing those plans. In this talk, I'll describe some of those advances in the context of a controlling a 400lb humanoid robot in a disaster response scenario and an airplane that can dart through a forest at 30 mph. And I'd like to send a clear message -- there is still a lot of work to be done! Even small improvements in our mathematical foundations, such as the algorithms which check if a polynomial equation is uniformly greater than zero, can make our robots more capable of moving through the world.

Speaker's Bio: Russ Tedrake is the X Consortium Associate Professor of Electrical Engineering and Computer Science, Aeronautics and Astronautics, and Mechanical Engineering at MIT, the Director of the Center for Robotics at the Computer Science and Artificial Intelligence Lab, and the leader of Team MIT's entry in the DARPA Robotics Challenge. He is a recipient of the NSF CAREER Award, the MIT Jerome Saltzer Award for undergraduate teaching, the DARPA Young Faculty Award in Mathematics, the 2012 Ruth and Joel Spira Teaching Award, and was named a Microsoft Research New Faculty Fellow. Russ received his B.S.E. in Computer Engineering from the University of Michigan, Ann Arbor, in 1999, and his Ph.D. in Electrical Engineering and Computer Science from MIT in 2004, working with Sebastian Seung. After graduation, he joined the MIT Brain and Cognitive Sciences Department as a Postdoctoral Associate. During his education, he has also spent time at Microsoft, Microsoft Research, and the Santa Fe Institute.
 
 
10/30/15 Frank Dellaert
Skydio
Factor Graphs for Flexible Inference in Robotics and Vision

Abstract: Simultaneous Localization and Mapping (SLAM) and Structure from Motion (SFM) are important and closely related problems in robotics and vision. I will show how both SLAM and SFM instances can be posed in terms of a graphical model, a factor graph, and that inference in these graphs can be understood as variable elimination. The overarching theme of the talk will be to emphasize the advantages and intuition that come with seeing these problems in terms of graphical models. For example, while the graphical model perspective is completely general, linearizing the non-linear factors and assuming Gaussian noise yields the familiar direct linear solvers such as Cholesky and QR factorization. Based on these insights, we have developed both batch and incremental algorithms defined on graphs in the SLAM/SFM domain. I will also discuss my recent work on using polynomial bases for trajectory optimization, inspired by pseudospectral optimal control, which is made easy by the new Expressions language in GTSAM 4, currently under development.

Speaker’s Bio: Frank Dellaert is currently on leave from the Georgia Institute of Technology for a stint as Chief Scientist of Skydio, a startup founded by MIT grads to create intuitive interfaces for micro-aerial vehicles. When not on leave, he is a Professor in the School of Interactive Computing and Director of the Robotics PhD program at Georgia Tech. His research interests lie in the overlap of Robotics and Computer vision, and he is particularly interested in graphical model techniques to solve large-scale problems in mapping and 3D reconstruction. You can find out about his group’s research and publications at https://borg.cc.gatech.edu and http://www.cc.gatech.edu/~dellaert. The GTSAM toolbox which embodies many of the ideas his group has worked on in the past few years is available for download at http://tinyurl.com/gtsam.
 
 
 
 
11/13/15 Louis Phillip Morency
CMU
Modeling Human Communication Dynamics

Abstract: Human face-to-face communication is a little like a dance, in that participants continuously adjust their behaviors based on verbal and nonverbal cues from the social context. Today's computers and interactive devices are still lacking many of these human-like abilities to hold fluid and natural interactions. Leveraging recent advances in machine learning, audio-visual signal processing and computational linguistic, my research focuses on creating computational technologies able to analyze, recognize and predict human subtle communicative behaviors in social context. I formalize this new research endeavor with a Human Communication Dynamics framework, addressing four key computational challenges: behavioral dynamic, multimodal dynamic, interpersonal dynamic and societal dynamic. Central to this research effort is the introduction of new probabilistic models able to learn the temporal and fine-grained latent dependencies across behaviors, modalities and interlocutors. In this talk, I will present some of our recent achievements modeling multiple aspects of human communication dynamics, motivated by applications in healthcare (depression, PTSD, suicide, autism), education (learning analytics), business (negotiation, interpersonal skills) and social multimedia (opinion mining, social influence).

Speaker's Bio: Louis-Philippe Morency is Assistant Professor in the Language Technology Institute at the Carnegie Mellon University where he leads the Multimodal Communication and Machine Learning Laboratory (MultiComp Lab). He received his Ph.D. and Master degrees from MIT Computer Science and Artificial Intelligence Laboratory. In 2008, Dr. Morency was selected as one of "AI's 10 to Watch" by IEEE Intelligent Systems. He has received 7 best paper awards in multiple ACM- and IEEE-sponsored conferences for his work on context-based gesture recognition, multimodal probabilistic fusion and computational models of human communication dynamics. Dr. Morency was lead Co-PI for the DARPA-funded multi-institution effort called SimSensei which was recently named one of the year’s top ten most promising digital initiatives by the NetExplo Forum, in partnership with UNESCO.
 
 
11/20/15 Tom Whelan
Oculus Research
Real-time dense methods for 3D perception

Abstract: In the past few years real-time dense methods have exploded onto the scene of robotics and general 3D perception. Key to the high level algorithms which can exploit this kind of data the most is the underlying method for generating and reconstructing a dense 3D representation. This talk will firstly contain an overview of the Kintinuous system for large scale real-time dense SLAM as well as a number of more recent results in use cases such as object detection, semantics and in-the-loop robotic control. Secondly, the recently published ElasticFusion system for real-time comprehensive dense 3D reconstruction will be presented, posing an alternative map-centric approach to the SLAM problem. Finally, some very recent results on extracting advanced surface information from the scene in real-time will be shown, paving the way for the future of real-time dense methods.

Speaker's Bio: Dr. Thomas Whelan is currently a Research Scientist at Oculus Research in Redmond working with the Surreal Vision team. Previous to this he spent one year as a post doctoral research fellow at the Dyson Robotics Laboratory at Imperial College London, lead by Prof. Andrew J. Davison. He was previously a Ph.D. student at the National University of Ireland Maynooth under a 3 year post-graduate scholarship from the Irish Research Council. In 2012 he spent 3 months as a visiting researcher at Prof. John Leonard’s group in CSAIL, MIT funded by a Science Foundation Ireland Short-Term Travel Fellowship. He received his B.Sc. (Hons) in Computer Science & Software Engineering from the National University of Ireland Maynooth in 2011. His research focuses on developing methods for dense real-time perception and its applications in SLAM and robotics.
 
 
12/4/15 Seth Hutchinson
UIUC
Robust Distributed Control Policies for Multi-Robot Systems

Abstract: In this talk, I will describe our recent progress in developing fault-tolerant distributed control policies for multi-robot systems. We consider two problems: rendezvous and coverage. For the former, the goal is to bring all robots to a common location, while for the latter the goal is to deploy robots to achieve optimal coverage of an environment. We consider the case in which each robot is an autonomous decision maker that is anonymous (i.e., robots are indistinguishable to one another), memoryless (i.e., each robot makes decisions based upon only its current information), and dimensionless (i.e., collision checking is not considered). Each robot has a limited sensing range, and is able to directly estimate the state of only those robots within that sensing range, which induces a network topology for the multi-robot system. We assume that it is not possible for the fault-free robots to identify the faulty robots (e.g., due to the anonymous property of the robots). For each problem, we provide an efficient computational framework and analysis of algorithms, all of which converge in the face of faulty robots under a few assumptions on the network topology and sensing abilities.

Speaker's Bio: Seth Hutchinson received his Ph.D. from Purdue University in 1988. In 1990 he joined the faculty at the University of Illinois in Urbana-Champaign, where he is currently a Professor in the Department of Electrical and Computer Engineering, the Coordinated Science Laboratory, and the Beckman Institute for Advanced Science and Technology. He served as Associate Department Head of ECE from 2001 to 2007. He currently serves on the editorial boards of the International Journal of Robotics Research and the Journal of Intelligent Service Robotics, and chairs the steering committee of the IEEE Robotics and Automation Letters. He was Founding Editor-in-Chief of the IEEE Robotics and Automation Society's Conference Editorial Board (2006-2008), and Editor-in-Chief of the IEEE Transaction on Robotics (2008-2013). He has published more than 200 papers on the topics of robotics and computer vision, and is coauthor of the books "Principles of Robot Motion: Theory, Algorithms, and Implementations," published by MIT Press, and "Robot Modeling and Control," published by Wiley. Hutchinson is a Fellow of the IEEE.
 
 
12/11/15 Dmitry Berenson
WPI
Toward General-Purpose Manipulation of Deformable Objects

Abstract: Imagine a robot that could perceive and manipulate rigid objects as skillfully as a human adult. Would a robot that had such amazing capabilities be able to perform the range of practical manipulation tasks we expect in settings such as the home? Consider that this robot would still be unable to prepare a meal, do laundry, or make a bed because these tasks involve deformable object manipulation. Unlike in rigid-body manipulation, where methods exist for general-purpose pick-and-place tasks regardless of the size and shape of the object, no such methods exist for a similarly broad and practical class of deformable object manipulation tasks. The problem is indeed challenging, as these objects are not straightforward to model and have infinite-dimensional configuration spaces, making it difficult to apply established motion planning approaches. Our approach seeks to bypass these difficulties by representing deformable objects using simplified geometric models at both the global and local planning levels. Though we cannot predict the state of the object precisely, we can nevertheless perform tasks such as cable-routing, cloth folding, and surgical probe insertion in geometrically-complex environments. Building on this work, our new projects in this area aim to blend exploration of the model space with goal-directed manipulation of deformable objects and to generalize the methods we have developed to motion planning for soft robot arms, where we can exploit contact to mitigate the actuation uncertainty inherent in these systems.

Speaker's Bio: Dmitry Berenson received a BS in Electrical Engineering from Cornell University in 2005 and received his Ph.D. degree from the Robotics Institute at Carnegie Mellon University in 2011, where he was supported by an Intel PhD Fellowship. He completed a post-doc at UC Berkeley and started as an Assistant Professor in Robotics Engineering and Computer Science at WPI in 2012. He founded and directs the Autonomous Robotic Collaboration (ARC) Lab at WPI, which focuses on motion planning, manipulation, and human-robot collaboration.
 
Spring 2015
04/17/15 Neil Lebeck and Natalie Brace
UW CSE
Multirotor Aerial Vehicles: Modeling, Estimation, and Control of Quadrotor

Abstract: This article provides a tutorial introduction to modeling, estimation, and control for multirotor aerial vehicles that includes the common four-rotor or quadrotor case. Aerial robotics is a fast-growing field of robotics and multirotor aircraft, such as the quadrotor, are rapidly growing in popularity. In fact, quadrotor aerial robotic vehicles have become a standard platform for robotics research worldwide. They already have sufficient payload and flight endurance to support a number of indoor and outdoor applications, and the improvements of battery and other technology is rapidly increasing the scope for commercial opportunities. They are highly maneuverable and enable safe and low-cost experimentation in mapping, navigation, and control strategies for robots that move in three-dimensional (3-D) space. This ability to move in 3-D space brings new research challenges compared with the wheeled mobile robots that have driven mobile robotics research over the last decade. Small quadrotors have been demonstrated for exploring and mapping 3-D environments; transporting, manipulating, and assembling objects; and acrobatic tricks such as juggling, balancing, and flips. Additional rotors can be added, leading to generalized N-rotor vehicles, to improve payload and reliability.
04/24/15 Peter Henry
UW CSE
LSD-SLAM: Large-Scale Direct Monocular SLAM

Abstract: We propose a direct (feature-less) monocular SLAM algorithm which, in contrast to current state-of-the-art regarding direct methods, allows to build large-scale, consistent maps of the environment. Along with highly accurate pose estimation based on direct image alignment, the 3D environment is reconstructed in real-time as pose-graph of keyframes with associated semi-dense depth maps. These are obtained by filtering over a large number of pixelwise small-baseline stereo comparisons. The explicitly scale-drift aware formulation allows the approach to operate on challenging sequences including large variations in scene scale. Major enablers are two key novelties: (1) a novel direct tracking method which operates on sim(3), thereby explicitly detecting scale-drift, and (2) an elegant probabilistic solution to include the effect of noisy depth values into tracking. The resulting direct monocular SLAM system runs in real-time on a CPU.
05/01/15 Dan Butler
UW CSE
Probabilistic Segmentation and Targeted Exploration of Objects in Cluttered Environments

Abstract: Creating robots that can act autonomously in dynamic, unstructured environments requires dealing with novel objects. Thus, an off-line learning phase is not sufficient for recognizing and manipulating such objects. Rather, an autonomous robot needs to acquire knowledge through its own interaction with its environment, without using heuristics encoding human insights about the domain. Interaction also allows information that is not present in static images of a scene to be elicited. Out of a potentially large set of possible interactions, a robot must select actions that are expected to have the most informative outcomes to learn efficiently. In the proposed bottom-up, probabilistic approach, the robot achieves this goal by quantifying the expected informativeness of its own actions in information-theoretic terms. We use this approach to segment a scene into its constituent objects. We retain a probability distribution over segmentations. We show that this approach is robust in the presence of noise and uncertainty in real-world experiments. Evaluations show that the proposed information-theoretic approach allows a robot to efficiently determine the composite structure of its environment. We also show that our probabilistic model allows straightforward integration of multiple modalities, such as movement data and static scene features. Learned static scene features allow for experience from similar environments to speed up learning for new scenes.
05/07/15 Marc Deisenroth
Imperial College, London
Statistical Machine Learning for Autonomous Systems and Robots

Abstract: Statistical machine learning has been a promising direction in control and robotics for more than a decade since learning models and controllers from data allows us to reduce the amount of engineering knowledge that is otherwise required. In real systems, such as robots, many experiments, which are often required for machine learning and reinforcement learning methods, can be impractical and time consuming. To address this problem, current learning approaches typically require task-specific knowledge in form of expert demonstrations, pre-shaped policies, or the underlying dynamics.

In the first part of the talk, I follow a different approach and speed up learning by efficiently extracting information from sparse data. In particular, I propose to learn a probabilistic, non-parametric Gaussian process dynamics model. By explicitly incorporating model uncertainty in long-term planning and controller learning my approach reduces the effects of model errors, a key problem in model-based learning. Compared to state-of-the art reinforcement learning my model-based policy search method achieves an unprecedented speed of learning. I demonstrate its applicability to autonomous learning from scratch in real robot and control tasks.

In the second part of my talk, I will discuss an alternative method for learning controllers for bipedal locomotion based on Bayesian Optimization, where it is hard to learn models of the underlying dynamics due to ground contacts. Using Bayesian optimization, we sidestep this modeling issue and directly optimize the controller parameters without the need of modeling the robot's dynamics.

In the third part of my talk, I will discuss state estimation in dynamical systems (filtering and smoothing) from a machine learning perspective. I will present a unifying view on Bayesian latent-state estimation, which allows both to re-derive common filters (e.g., the Kalman filter) and devise novel smoothing algorithms in dynamical systems. I will demonstrate the applicability of this approach to intention inference in robot table tennis.

Related papers:
Marc P. Deisenroth, Dieter Fox, Carl E. Rasmussen, Gaussian Processes for Data-Efficient Learning in Robotics and Control, IEEE Transactions on Pattern Analysis and Machine Intelligence, volume 37, pp.408–423, 2015.

Roberto Calandra, Jan Peters, Andre Seyfarth, Marc P. Deisenroth, An Experimental Evaluation of Bayesian Optimization on Bipedal Locomotion, Proceedings of the IEEE International Conference on Robotics and Automation (ICRA), 2014.

Marc P. Deisenroth, Ryan Turner, Marco Huber, Uwe D. Hanebeck, Carl E. Rasmussen, Robust Filtering and Smoothing with Gaussian Processes, IEEE Transactions on Automatic Control, volume 57, pp.1865–1871, 2012.

Bio: Dr Marc Deisenroth is an Imperial College Junior Research Fellow and head of the Statistical Machine Learning Group in the Department of Computing at Imperial College London (UK). From December 2011 to August 2013 he was a Senior Research Scientist at TU Darmstadt (Germany). From February 2010 to December 2011, he was a full-time Research Associate at the University of Washington (Seattle). He completed his PhD at the Karlsruhe Institute for Technology (Germany). Marc conducted his PhD research at the Max Planck Institute for Biological Cybernetics (2006-2007) and at the University of Cambridge (2007-2009). Marc was Program Chair of the "European Workshop on Reinforcement Learning" (EWRL) in 2012 and Workshops Chair of "Robotics: Science & Systems" (RSS) in 2013. His interdisciplinary research expertise centers around machine learning, control, robotics, and signal processing.
05/15/15 Arunkumar Byravan and Kendall Lowrey
UW CSE
Reinforcement Learning in Robotics: A Survey

Abstract: Reinforcement learning offers a framework and set of tools for the design of sophisticated and hard-to-engineer behaviours. In the general reinforcement learning setting, an agent tries to autonomously discover an optimal behaviour through trial-and-error interactions with its environment. Instead of explicitly detailing the solution to a problem, in reinforcement learning the designer of a control task provides feedback in terms of a scalar objective function that measures the one-step performance of the agent.

Robotics as a reinforcement learning domain differs considerably from most well-studied reinforcement learning benchmark problems. Problems in robotics are often high-dimensional, have continuous state and action spaces and partially-observable states. Additionally, experience on robots is tedious to obtain, expensive and often hard to reproduce. In spite of these difficulties, there have been many successful applications of reinforcement learning to robotics.

In this talk, we attempt to give a high-level overview of reinforcement learning for robotics. In the first part of the talk, we go over the formulation of the reinforcement learning problem and its inherent challenges as compared to other machine learning problems. We discuss two approaches to solving the reinforcement learning problem (value function methods and policy search) and their applicability to the robotics domain. In the second part of the talk, we discuss a few recent applications of reinforcement learning in learning robot tasks. We conclude by highlighting some open questions and practical difficulties in applying reinforcement learning to robotics.
05/22/15 Sonya Alexandrova
Zach Tatlock
Maya Cakmak
RoboFlow: A Flow-based Visual Programming Language for Mobile Manipulation Tasks

Abstract: General-purpose robots can perform a range of useful tasks in human environments; however, programming them to robustly function in all possible environments that they might encounter is unfeasible. Instead, our research aims to develop robots that can be programmed by its end-users in their context of use, so that the robot needs to robustly function in only one particular environment. This requires intuitive ways in which end-users can program their robot. To that end, this paper contributes a flow-based visual programming language, called RoboFlow, that allows programming of generalizable mobile manipulation tasks. RoboFlow is designed to (i) ensure a robust low-level implementation of program procedures on a mobile manipulator, and (ii) restrict the high-level programming as much as possible to avoid user errors while enabling expressive programs that involve branching, looping, and nesting. We present an implementation of RoboFlow on a PR2 mobile manipulator and demonstrate the generalizability and error handling properties of RoboFlow programs on everyday mobile manipulation tasks in human environments.
05/22/15 Maxwell Forbes
Rajesh Rao
Luke Zettlemoyer
Maya Cakmak
Robot Programming by Demonstration with Situated Spatial Language Understanding

Abstract: Robot Programming by Demonstration (PbD) allows users to program a robot by demonstrating the desired behavior. Providing these demonstrations typically involves moving the robot through a sequence of states, often by physically manipulating it. This requires users to be co-located with the robot and have the physical ability to manipulate it. In this paper, we present a natural language based interface for PbD that removes these requirements and enables hands-free programming. We focus on programming object manipulation actions—our key insight is that such actions can be decomposed into known types of manipulator movements that are naturally described using spatial language; e.g., object reference expressions and prepositions. Our method takes a natural language command and the current world state to infer the intended movement command and its parametrization. We implement this method on a two-armed mobile manipulator and demonstrate the different types of manipulation actions that can be programmed with it. We compare it to a kinesthetic PbD interface and we demonstrate our method’s ability to deal with incomplete language.
05/22/15 Danying Hu
Yuanzheng Gong
Blake Hannaford
Eric J. Seibel
Semi-autonomous Simulated Brain Tumor Ablation with RavenII Surgical Robot using Behavior Tree

Abstract: Medical robots have been widely used to assist surgeons to carry out dexterous surgical tasks via various ways. Most of the tasks require surgeon’s operation directly or indirectly. Certain level of autonomy in robotic surgery could not only free the surgeon from some tedious repetitive tasks, but also utilize the advantages of robot: high dexterity and accuracy. This paper presents a semi-autonomous neurosurgical procedure of brain tumor ablation using RAVEN Surgical Robot and stereo visual feedback. By integrating with the behavior tree framework, the whole surgical task is modeled flexibly and intelligently as nodes and leaves of a behavior tree. This paper provides three contributions mainly: (1) describing the brain tumor ablation as an ideal candidate for autonomous robotic surgery, (2) modeling and implementing the semi-autonomous surgical task using behavior tree framework, and (3) designing an experimental simulated ablation task for feasibility study and robot performance analysis.
05/22/15 Kevin Huang
Liang-Ting Jiang
Joshua R. Smith
Howard Jay Chizeck
Sensor-Aided Teleoperated Grasp of Transparent Objects

Abstract: This paper presents a method of augmenting streaming point cloud data with pretouch proximity sensor information for the purposes of teleoperated grasping of transparent targets. When using commercial RGB-Depth (RGB-D) cameras, material properties can significantly affect depth measurements. In particular, transparent objects are difficult to perceive with RGB images and commercially available depth sensors. Geometric information of such objects needs to be gathered with additional sensors, and in many scenarios, it is of interest to gather this information without physical contact. In this work, a non-contact pretouch sensor fixed to the robot end effector is used to sense and explore physical geometries previously unobserved. Thus, the point cloud representation of an unknown, transparent grasp target, can be enhanced through telerobotic exploration in real-time. Furthermore, real-time haptic rendering algorithms and haptic virtual fixtures used in combination with the augmented streaming point clouds assist the teleoperator in collision avoidance during exploration. Theoretical analyses are performed to design virtual fixtures suitable for pretouch sensing, and experiments show the effectiveness of this method to gather geometry data without collision and eventually to successfully grasp a transparent object.
05/22/15 Tanner Schmidt
Katharina Hertkorn
Richard Newcombe
Zoltan Marton
Michael Suppa
Dieter Fox
Depth-Based Tracking with Physical Constraints for Robot Manipulation

Abstract: This work integrates visual and physical constraints to perform real-time depth-only tracking of articulated objects, with a focus on tracking a robot’s manipulators and manipulation targets in realistic scenarios. As such, we extend DART, an existing visual articulated object tracker, to additionally avoid interpenetration of multiple interacting objects, and to make use of contact information collected via torque sensors or touch sensors. To achieve greater stability, the tracker uses a switching model to detect when an object is stationary relative to the table or relative to the palm and then uses information from multiple frames to converge to an accurate and stable estimate. Deviation from stable states is detected in order to remain robust to failed grasps and dropped objects. The tracker is integrated into a shared autonomy system in which it provides state estimates used by a grasp planner and the controller of two anthropomorphic hands. We demonstrate the advantages and performance of the tracking system in simulation and on a real robot. Qualitative results are also provided for a number of challenging manipulations that are made possible by the speed, accuracy, and stability of the tracking system.
05/22/15 Eric Schoof
Airlie Chapman
Mehran Mesbahi
Efficient Leader Selection for Translation and Scale of a Bearing-Compass Formation

Abstract: The paper considers the efficient selection of leader agents in a swarm running a distributed bearing-compass formation controller. The leaders apply external control which induces translation and scaling of the formation, providing manipulation methods useful to a human operator. The selection algorithm for maximizing translation and scale draws from modularity and submodularity theory. Consequently, the algorithms exhibit guaranteed optimal and suboptimal performance, respectively. For more restricted human-swarm interaction requiring pure translation and scale, a relaxed integer programming algorithm is described to reduce the combinatorial optimization problem to a computationally tractable semidefinite program. The leader selection strategies are supported through demonstration on a swarm testbed.
06/05/15 Jim Youngquist
UW CSE
A Strictly Convex Hull for Computing Proximity Distances With Continuous Gradients

Abstract: We propose a new bounding volume that achieves a tunable strict convexity of a given convex hull. This geometric operator is named sphere-tori-patches bounding volume (STP-BV), which is the acronym for the bounding volume made of patches of spheres and tori. The strict convexity of STP-BV guarantees a unique pair of witness points and at least C1 continuity of the distance function resulting from a proximity query with another convex shape. Subsequently, the gradient of the distance function is continuous. This is useful for integrating distance as a constraint in robotic motion planners or controllers using smooth optimization techniques. For the sake of completeness, we compare performance in smooth and nonsmooth optimization with examples of growing complexity when involving distance queries between pairs of convex shapes.
 
Winter 2015
01/15/15 Mike Chung
UW CSE
Accelerating Imitation Learning through Crowdsourcing

Abstract: Although imitation learning is a powerful technique for robot learning and knowledge acquisition from na ̈ıve human users, it often suffers from the need for expensive human demonstrations. In some cases the robot has an insufficient number of useful demonstrations, while in others its learning ability is limited by the number of users it directly interacts with. We propose an approach that overcomes these short- comings by using crowdsourcing to collect a wider variety of examples from a large pool of human demonstrators online. We present a new goal-based imitation learning framework which utilizes crowdsourcing as a major source of human demonstration data. We demonstrate the effectiveness of our approach experimentally on a scenario where the robot learns to build 2D object models on a table from basic building blocks using knowledge gained from locals and online crowd workers. In addition, we show how the robot can use this knowledge to support human-robot collaboration tasks such as goal inference through object-part classification and missing-part prediction. We report results from a user study involving fourteen local demonstrators and hundreds of crowd workers on 16 different model building tasks.

Bio:Mike Chung is a third year graduate student at UW in CSE. His research interests are human-robot interaction and machine learning. His advisors are Rajesh Rao and Maya Cakmak, and he has collaborated with Dieter Fox, Su-In Lee and Jeff Bilmes.
01/15/15 Tanner Schmidt
UW CSE
Dense Articulated Real-Time Tracking

Abstract: We have developed DART, a general framework for tracking articulated objects, such as human bodies, human hands, and robots, with RGB-D sensors. We took a generative model approach, where the model is an extension the recently-popular signed distance function representation to articulated objects. Articulated poses are estimated via gradient descent on an error function which combines a standard articulated ICP formulation with additional terms which penalize violation of apparent free space and model self-intersection. Importantly, all error terms are trivially parallelizable, and optimized on a GPU, allowing for real-time performance while tracking many degrees of freedom. The practical applicability of the fast and accurate tracking provided by DART has been demonstrated in a robotics application in which live estimates of robot hands and of a target object are used to plan and execute grasps.

Bio: Tanner Schmidt is a graduate student in Computer Science and Engineering at the University of Washington, working with Dieter Fox in the Robotics and State Estimation lab. His primary interests are robotics, computer vision, and artificial intelligence. He received his bachelor's degree in Electrical and Computer Engineering and Computer Science from Duke University in 2012, and began at UW in the fall of 2012.
02/06/15 Joseph Xu
UW CSE
Design and Control of an Anthropomorphic Robotic Hand: Learning Advantages From the Human Body & Brain

Abstract: According to the cortical homunculus, our hand function requires over one quarter of the brain power allocated for the whole body's motor/sensory activities. The evolutionary role of the human hand is more than just being the manipulation tool that allows us to physically interact with the world. Recent study shows that our hands can also affect the mirror neuron system that enables us to cognitively learn and imitate the actions of others. However the state-of-art technologies only allow us to make cosmetically true-to-life prosthetic hands with cadaver-like stiff joints made of mechanical substitutes. And very few research group know how to design robotic hands that can closely mimic the salient biological features of the human hand. The goal of our project is to reduce cognitive and physical discrepancy, in the cases where we need a pair of our hands interacting with a different environment remotely. Our project will try to answer the following questions: With the great advance of 3D-printing technologies, and promising new materials for artificial muscles and ligaments, can we design a personalized anthropomorphic robotic hand that possesses all the favorable functions of our very own hand? With such a robotic hand, can we reduce the control space, and establish a easy mapping for the human user to effectively control it? Is it possible to teleoperate the robotic hand to perform amazingly dexterous tasks without force feedback as those surgical robots demonstrated? To answer these questions, we are going to investigate the design and control of our proposed anthropomorphic robotic hand.

Bio:Zhe (Joseph) Xu is a graduate student at University of Washington's Movement Control Laboratory lab working under the supervision of Emanuel Todorvo and Joshua Smith. His research interest is in the area of biomimetics, soft robotics, rehabilitation robotics, control systems, and robotic surgery. He hold degrees in three different fields: mechanical engineering, bioengineering, and computer science & engineering. His current research focuses on designing and analysing highly biomimetic robots with biological “soft” artificial joints through rapid prototyping technologies like 3D scanning and printing.
02/06/15 Vikash Kumar
UW CSE
Dimensionality Augmentation: A tool towards synthesizing complex and expressive behaviors

Abstract: Dexterous hand manipulation is one of the most complex types of biological movement, and has proven very difficult to replicate in robots. The usual approaches to robotic control – following predefined trajectories or planning online with reduced models – are both not applicable. Dexterous manipulation is so sensitive to small variations in contact forces and object location that it seems to require online planning without any simplifications. This entail searching in high dimensional spaces full of discontinuities (due to contacts and constraints) and dynamic phenomena (such as rolling, sliding and deformation).

This talk will introduce ‘Dimensionality Augmentation’ as a primary tool towards synthesizing complex and expressive behaviors in high dimensional, non-smooth search spaces. Although somewhat counterintuitive, these methods involve smartly augmenting the dimensionality of an already high dimensional search space in order to make optimizers amenable to the curse of dimensionality. Optimizers make quick progress along these augmented dimensions first, and then search the neighborhood exhaustively. Unlike other methods, which hinders the dexterity by constraining search spaces, faster convergence and improved search capabilities on the full search space results in more expressive and dynamic behaviors. Dimensionality Augmentation, in association with other tools, enabled us to demonstrate for the first time online planning (i.e. model-predictive control) with a full physics model of a humanoid hand with 28 degrees of freedom and 48 pneumatic actuators. Results include full hand behaviors like prehensile and non-prehensile object manipulation and finger focused behaviors like typing. In both cases the input to the system is a high level task description, while all details of the hand movement emerge from fully automated online numerical optimization.

Bio:Vikash Kumar is a graduate student at the University of Washington's Movement Control Lab, working under the supervision of Prof. Emanuel Todorov. He previously completed a master's and undergraduate degree in Mathematics and Scientific Computing at Indian Institute of Technology, Kharagpur. His research interests lie in developing bio-mimetic systems and behavior synthesis for dexterous hand manipulation.
02/13/15 Sofia Alexandrova
UW CSE
RoboFlow: A Flow-based Visual Programming Language for Mobile Manipulation Tasks

Abstract: General-purpose robots can perform a range of useful tasks in human environments. However, programming them requires many hours of expert work, and programming a robot to perform robustly in any possible environment is infeasible. We describe a system that allows non-expert users to program the robot for their specific environment. We show that the system, implemented for the PR2 mobile manipulator, is intuitive and can be used by users unfamiliar with robotics. We further extend the system into a visual programming language - RoboFlow - that allows looping, branching and nesting of programs. We demonstrate the generalizability and error handling properties of RoboFlow programs on everyday mobile manipulation tasks.

Bio: Sofia Alexandrova is a third-year graduate student in Computer Science and Engineering at the University of Washington. She is part of the Human-Centered Robotics lab, working with Maya Cakmak. Her main interests in robotics are human-robot interaction and programming by demonstration. She received a Masters degree in Software Engineering from St. Petersburg Academic University in 2011, and her bachelor's degree in Physics from St. Petersburg Polytechnic University in 2009.
02/20/15 Igor Mordatch
UW CSE
Synthesis of Interactive Control for Diverse Complex Characters with Neural Networks

Abstract: We present a method for automatic synthesis of interactive real-time controllers, applicable to complex three-dimensional characters. The same method is able to generate stable and realistic behaviors in a range of diverse tasks -- swimming, flying, biped and quadruped walking. It does not require motion capture or task-specific features or state machines. Instead, our method creates controllers de novo just from the physical model of the character and the definition of the control objectives. The controller is a neural network, having a large number of feed-forward units that learn elaborate state-action mappings, and a small number of recurrent units that implement memory states beyond the physical state of the character. The action generated by the network is defined as velocity. Thus the network is not learning a control policy, but rather the physics of the character under an implicit policy. Learning relies on a combination of supervised neural network training and trajectory optimization. Essential features include noise injected during training, training for unexpected changes in the task specification, and using the trajectory optimizer to obtain optimal feedback gains in addition to optimal actions. Although training is computationally-expensive and relies on cloud and GPU computing, the interactive animation can run in real-time on any processor once the network parameters are learned.

This is joint work with Kendall Lowrey, Galen Andrew, Zoran Popovic, and Emanuel Todorov

Bio:Igor Mordatch is a graduate student at the University of Washington's Graphics and Imaging Laboratory lab, working under the supervision of Emanuel Todorov and Zoran Popovic. He previously completed a master's and undergraduate degree in Computer Science and Mathematics at University of Toronto. His research interests lie in the use of physics-based methods, optimization, and machine learning techniques for graphics content creation, robotics, and biomechanics.
02/27/15 Richard Newcombe
UW CSE
DynamicFusion: Reconstruction and Tracking of Non-Rigid Scenes in Real-Time

Abstract: We present the first dense SLAM system capable of reconstructing non-rigidly deforming scenes in real-time, by fusing together RGBD scans captured from commodity sensors. Our DynamicFusion approach reconstructs scene geometry whilst simultaneously estimating a dense volumetric 6D motion field that warps the estimated geometry into a live frame. Like KinectFusion, our system produces increasingly denoised, detailed, and complete reconstructions as more measurements are fused, and displays the updated model in real time. Because we do not require a template or other prior scene model, the approach is applicable to a wide range of moving objects and scenes. In this talk I will outline how DynamicFusion works and motivate some of things we hope to do with it soon.

Bio: Richard Newcombe is a Postdoctoral research associate at the University of Washington, working on computer vision with Steve Seitz and Dieter Fox. He researched Robot Vision for his Ph.D. with Andrew Davison and Murray Shanahan at Imperial College, London, and before that he studied with Owen Holland at the University of Essex where he received his BSc. and MSc. in robotics, machine learning and embedded systems.
03/06/15 Aaron Steinfeld
Carnegie Mellon University
Understanding and Creating Appropriate Robot Behavior

Abstract: End users expect appropriate robot actions, interventions, and requests for human assistance. As with most technologies, robots that behave in unexpected and inappropriate ways face misuse, abandonment, and sabotage. Complicating this challenge are human misperceptions of robot capability, intelligence, and performance. This talk will summarize work from several projects focused on this human-robot interaction challenge. Findings and examples will be shown from work on human trust in robots, deceptive robot behavior, robot motion, and robot characteristics. It is also important to examine the human-robot system, rather than just the robot. To this end, it is possible to draw lessons learned from related work in crowdsourcing (e.g., Tiramisu Transit) to help inform methods for enabling and supporting contributions by end users and local experts.

Bio:Aaron Steinfeld is an Associate Research Professor in the Robotics Institute (RI) at Carnegie Mellon University. He received his BSE, MSE, and Ph.D. degrees in Industrial and Operations Engineering from the University of Michigan and completed a Post Doc at U.C. Berkeley. He is the Co-Director of the Rehabilitation Engineering Research Center on Accessible Public Transportation (RERC-APT), Director of the DRRP on Inclusive Cloud and Web Computing, and the area lead for transportation related projects in the Quality of Life Technology Center (QoLT). His research focuses on operator assistance under constraints, i.e., how to enable timely and appropriate interaction when technology use is restricted through design, tasks, the environment, time pressures, and/or user abilities. His work includes intelligent transportation systems, crowdsourcing, human-robot interaction, rehabilitation, and universal design.
02/27/15 Luis Puig
UW CSE
Overview of Omnidirectional Vision

Abstract: We will present an overview of omnidirectional vision, whose major advantage over conventional systems is its wide field of view. In particular we will discuss the catadioptric systems, which are a combination of conic mirrors and conventional cameras. As any other vision system, its ultimate goal is to provide useful 3D information about the environment. In order to achieve this goal, several hierarchical steps are performed. In this talk we will cover several of this steps, from camera calibration to the two-view geometry of such systems and its combination with conventional cameras. Moreover, we will show higher level applications using this type of systems, such as robot localization, image stabilization and SLAM and their advantages over conventional systems.

Bio: Luis Puig is a post-doctoral researcher in the Department of Computer Science & Engineering at the University of Washington under the supervision of Prof. Dieter Fox. He obtained his PhD degree from the University of Zaragoza at the Robotics, Perception and Real Time group. He is interested in omnidirectional vision, visual odometry, SLAM, object recognition and Structure from Motion.
Autumn 2014
10/03/14 Danny Kaufman
Adobe Creative Technologies Lab, Seattle
Geometric Algorithms for Computing Frictionally Contacting Systems

Abstract: Algorithms to accurately capture the combined effects of dissipation and contact processes are essential for the physical modeling of many poorly understood phenomena. These extend from prosaic domestic phenomena such as the chattering of a chair dragged across the floor to emergent pattern-formation in driven granular assemblies.
Yet the fundamental features of contact mechanics expose significant challenges to computation including strong nonlinearity, nonsmoothness, nonconvexity, and nonuniqueness, compounded by the difficulties of scaling to the high-dimensional systems and interactive rates required by modern research, entertainment, and industrial applications. In this talk I will discuss how these fundamental challenges can be successfully addressed by geometric algorithms that respect core properties of modeled physical systems. I will explain how these critical geometric features are identified and incorporated as fundamental algorithmic building blocks so that predictive and convincing simulations follow by construction. I will present examples of how algorithms I have developed with such "baked-in" geometry have enabled the efficient and scalable computation of highly difficult and, in some cases, previously intractable simulation problems in contact modeling, animation, interactive design, and haptic rendering. Moving forward I will argue that building geometry into our computations is key for developing the next generation of physical simulation and design algorithms that are both simple, thus easing adoption and code maintenance, and yet efficiently predictive, thus producing reliable and visually compelling results.

Bio: Danny Kaufman is a research scientist at Adobe Creative Technologies Lab in Seattle. His research focuses on developing geometric algorithms and frameworks to obtain predictive, expressive, and efficient simulations of physical systems for applications in computer animation, interactive design, robotics, and computational physics. His work on physical simulation algorithms has led to ongoing collaborations with industrial partners including Weta Digital, Disney, and Thunderlily. He completed his PhD from Rutgers University in 2009, was a visiting scholar in the Imager Lab at The University of British Columbia from 2006 through 2010, and was a Postdoc in the Computer Science department at Columbia University from 2011 to 2013.
10/10/14 Dubi Katz & Michael Abrash
Oculus VR
VR, the future, and you

Abstract: In the surprisingly near future, VR is very likely to transform how we interact with information, computers, and each other. This talk will discuss why VR is likely to be a key part of our future, why it's different from anything that's come before, and what that implies for researchers and developers.

Bio: Over the last 30 years, Michael has worked at companies that made graphics hardware, computer-based instrumentation, and rendering software, been the GDI lead for the first couple of versions of Windows NT, worked with John Carmack on Quake, worked on Xbox and Xbox 360, written or co-written at least four software rasterizers (the last one of which, written at RAD Game Tools, turned into Intel’s late, lamented Larrabee project), and worked on VR at Valve. Along the way he wrote a bunch of magazine articles and columns for Dr. Dobb’s Journal, PC Techniques, PC Tech Journal, and Programmer’s Journal, as well as several books. He’s been lucky enough to have more opportunities to work on interesting stuff than he could ever have imagined when he almost failed sixth grade because he spent all his time reading science fiction. He thinks VR is going to be the most interesting project of all.
10/17/14
CSE 503
3:30pm
Kira Mourao
PostDoc, Institute for Language, Cognition and Computation, University of Edinburgh
What happens if I push this button? Learning planning operators from experience

Abstract: When a robot, dialog manager or other agent operates autonomously in a real-world domain, it uses a model of the dynamics of its domain to plan its actions. Typically, pre-specified domain models are used by AI planners to generate plans. However, creating these domain models is notoriously difficult. Furthermore, to be truly autonomous, agents must be able to learn their own models of world dynamics. An alternative therefore is to learn domain models from observations, either via known successful plans or through exploration of the world. This route is also challenging, as agents often do not operate in a perfect world: both actions and observations may be unreliable. In this talk I will present a method which, unlike other approaches, can learn from both observed successful plans and from action traces generated by exploration. Importantly, the method is robust in a variety of settings, able to learn useful domain models when observations are noisy and incomplete, or when action effects are noisy or non-deterministic. The approach first builds a classification model to predict the effects of actions, and then derives explicit planning operators from the classifiers. Through a range of experiments using International Planning Competition domains and a real robot domain, I will show that this approach learns accurate domain models suitable for use by standard planners. I also demonstrate that where settings are comparable, the results equal or surpass the performance of state-of-the-art methods.

Bio: Kira Mourao is a Postdoctoral Research Associate based in the Institute for Language, Cognition and Computation (ILCC) in the School of Informatics, at the University of Edinburgh. Currently she works on the EU project Xperience developing new methods for grounding action representations for robots. Her broad research interests are in both using cognitive robotics to inform theories of grounded cognition, and also in applying theories of grounded cognition to develop cognitive robots.
10/24/14 Sam Burden 
PostDoc, University of California, Berkeley
Hybrid Models for Dynamic and Dexterous Robots

Abstract: To move through and interact with the world, a robot must intermittently contact its environment. When contacts are established or broken, the equations of motion change abruptly. Models of these piecewise-defined (or "hybrid") dynamics exhibit discontinuities and inconsistencies that generally limit their utility. In this talk I present techniques that exploit intrinsic properties of the mechanics of locomotion and manipulation to circumvent these pathologies. By topologically quotienting and smoothing the hybrid state space, I remove discontinuities that arise when limbs impact terrain. By restricting the class of impact restitution laws, I resolve inconsistencies that emerge when several limbs touch down nearly simultaneously (as with a quadruped's trot or hexapod's alternating-tripod). In addition to broadening the applicability of hybrid models for the development of dynamic and dexterous robots, these results provide novel mechanisms for stabilization of rhythmic behaviors and aperiodic maneuvers.

Bio: Sam Burden earned his BS with Honors in Electrical Engineering from the University of Washington in Seattle. He earned his PhD in Electrical Engineering and Computer Sciences at the University of California in Berkeley where he is currently a post-doctoral researcher. In Fall 2015, Sam will return to UW EE as an Assistant Professor. He focuses on discovering and formalizing principles that enable dynamic locomotion and dexterous manipulation in robotics, biomechanics, and human motor control. Broadly, he is interested in developing a sensorimotor control theory for neuromechanical and cyberphysical systems. In his spare time, he enjoys teaching robotics to students of all ages in K-12 classrooms, Maker Fairs, and campus events.
10/31/14 Sergey Levine  
PostDoc, University of California, Berkeley
Learning to Move: Machine Learning for Robotics and Animation

Abstract: Being able to acquire new motion skills autonomously could help robots build rich motion repertoires suitable for tackling complex, varied environments. I will discuss my work on motion skill learning for robotics, including methods for learning from demonstration and reinforcement learning. In particular, I will describe a class of "guided" policy search algorithms, which combine reinforcement learning and learning from demonstration to acquire multiple simple, trajectory-centric policies, with a supervised learning phase to obtain a single complex, high-dimensional policy that can then generalize to new situations. I will show applications of this method to simulated bipedal locomotion, as well as a range of robotic manipulation tasks, including putting together two parts of a plastic toy and screwing bottle caps onto bottles. I will also discuss how such techniques can be applied to character animation in computer graphics, and how this field can inform research in robotics.

Bio: Sergey Levine is a postdoctoral researcher working with Professor Pieter Abbeel at the University of California at Berkeley. He previously completed his PhD with Professor Vladlen Koltun at Stanford University. His research areas include robotics, reinforcement learning and optimal control, machine learning, and computer graphics. His work includes the development of new algorithms for learning motor skills, methods for learning behaviors from human demonstration, and applications in robotics and computer graphics, ranging from robotic manipulation to animation of martial arts and conversational hand gestures.
11/14/14 Sachin Patil
PostDoc, University of California, Berkeley
Coping with Uncertainty in Robotic Navigation, Exploration, and Grasping

Abstract: A key challenge in robotics is to robustly complete navigation, exploration, and manipulation tasks when the state of the world is uncertain. This is a fundamental problem in several application areas such as logistics, personal robotics, and healthcare where robots with imprecise actuation and sensing are being deployed in unstructured environments. In such a setting, it is necessary to reason about the acquisition of perceptual knowledge and to perform information gathering actions as necessary. In this talk, I will present an approach to motion planning under motion and sensing uncertainty called "belief space" planning where the objective is to trade off exploration (gathering information) and exploitation (performing actions) in the context of performing a task. In particular, I will present how we can use trajectory optimization to compute locally-optimal solutions to a determinized version of this problem in Gaussian belief spaces. I will show that it is possible to obtain significant computational speedups without explicitly optimizing over the covariances by considering a partial collocation approach. I will also address the problem of computing such trajectories, given that measurements may not be obtained during execution due to factors such as limited field of view of sensors and occlusions. I will demonstrate this approach in the context of robotic grasping in unknown environments where the robot has to simultaneously explore the environment and grasp occluded objects whose geometry and positions are initially unknown.

Bio: Sachin Patil is a postdoctoral researcher working with Prof. Pieter Abbeel and Prof. Ken Goldberg at the University of California at Berkeley. He previously completed his PhD with Prof. Ron Alterovitz at University of North Carolina at Chapel Hill. His research focuses on developing rigorous motion planning algorithms to enable new, minimally invasive medical procedures and to facilitate reliable operation of robots in unstructured environments.
12/05/14 Marianna Madry
Ph.D. candidate at the Computer Vision and Active Perception Lab at the Royal Institute of Technology (KTH) in Stockholm, Sweden
Representing Objects in Robotics from Visual, Depth and Tactile Sensing

Abstract: Being able to localize, identify and manipulate objects is of key importance for a large range of tasks in robotics. The recent developments in depth cameras and haptic sensors have led to a wide availability and use of both visual, 3D and tactile data creating a need for defining suitable representations that enable detection, recognition and manipulation of objects. In this talk, I will discuss the desired characteristics of an object representation and the main challenges in real applications. I will start with demonstrating how simultaneous encoding of object appearance and affordance can enable transfer of grasp information to a novel object and facilitate object manipulation by a humanoid robot. Then, I will present our 3D data descriptor, the Global Structure Histogram (GSH), that by encoding global object structure of local surface properties can robustly generalize over different object poses, scales and data incompleteness outperforming the state-of-the-art global descriptors in real world conditions. Finally, I will introduce our new descriptor named Spatio-Temporal Hierarchical Matching Pursuit (ST-HMP) that captures properties of a time series of tactile sensor measurements. ST-HMP is based on the concept of unsupervised hierarchical feature learning realized using sparse coding.

Bio: Marianna is a Ph.D. candidate at the Computer Vision and Active Perception (CVAP) Lab at the Royal Institute of Technology (KTH) in Sweden advised by Professor Danica Kragic. Her research spans the areas of robotics and computer vision. She is interested in developing a representation of household objects that serves a wide range of robotics applications, such as object detection and classification, inferring object affordances, object grasping and manipulation. Recently, she has been visiting the RSE Lab at the University of Washington in USA working with Dieter Fox and Liefeng Bo. She was also involved in the EU GRASP project directed towards the development of a cognitive robots capable of performing grasping and manipulation tasks. The project involved collaboration with the High Performance Humanoid Technologies Lab at the Karlsruhe Institute of Technology (KIT), Germany and the Vision4Robotics Lab at the Vienna University of Technology (TUW), Austria.
12/18/14 Scott Niekum
Carnegie Mellon University
Structure Discovery in Robotics with Demonstrations and Active Learning

Abstract: Future co-robots in the home and workplace will require the ability to quickly characterize new tasks and environments without the intervention of expert engineers. Human demonstrations and active learning can play complementary roles when learning complex, multi-step tasks in novel environments—demonstrations are a fast, natural way to broadly provide human insight into task structure and environmental dynamics, while active learning can fine-tune models by exploiting the robot’s knowledge of its own internal representations and uncertainties.
Using these complementary data sources, I will focus on three types of structure discovery that can help robots quickly produce robust control strategies for novel tasks: 1) learning high-level task descriptions from unstructured demonstrations, 2) inferring physically-grounded models of task goals and environmental dynamics, and 3) interactive perception for refinement of physically-grounded models. These techniques draw from Bayesian nonparametrics, time series analysis, filtering, and control theory to characterize complex tasks like IKEA furniture assembly that challenge the state of the art in manipulation.

Bio: Scott Niekum is a postdoctoral fellow at the Carnegie Mellon Robotics Institute, working with Chris Atkeson. He received his Ph.D. in Computer Science from the University of Massachusetts Amherst in 2013 under the supervision of Andrew Barto, and his B.S from Carnegie Mellon University in 2005. His research interests include learning from demonstration, robotic manipulation, time-series analysis, and reinforcement learning.
Winter 2014
01/17/14 Byron Boots
University of Washington
Learning Better Models of Dynamical Systems

Abstract: The majority of sequential data in scientific and technological domains is high-dimensional, noisy, and collected in a raw and unstructured form. In order to interpret, track, predict, or control such data, we need to hypothesize a model. For this purpose, an appealing model representation is a dynamical system. Although we can sometimes use extensive domain knowledge to write down a dynamical system, specifying a model by hand can be a time consuming process. This motivates an alternative approach: learning dynamical systems directly from sensor data.

Unfortunately, this is hard. To discover the right state representation and model parameters, we must solve difficult temporal and structural credit assignment problems. In addition, popular maximum likelihood based approaches to learning dynamical systems often must contend with optimization environments that are plagued with bad local optima. In this talk, I will present a number of tools that we have used to learn better dynamical system models including predictive representations, moment-based learning algorithms, and kernel methods. These tools have allowed us to design a family of learning algorithms that are computationally efficient, statistically consistent, and have no local optima; in addition, they can be simple to implement, and have state-of-the-art practical performance for some interesting learning problems.

Bio: Byron Boots is a postdoctoral research associate in the Robotics and State Estimation Lab at the University of Washington working with Dieter Fox. He received his Ph.D. in Machine Learning from Carnegie Mellon University in 2012 under Geoffrey Gordon. His research focuses on on statistical machine learning, artificial intelligence, and robotics. Byron’s work won the 2010 Best Paper award at the International Conference on Machine Learning (ICML-2010).

01/24/14 Julie Shah
MIT
Integrating Robots into Team-Oriented Environments

Abstract: Recent advances in computation, sensing, and hardware enable robotics to perform an increasing percentage of traditionally manual tasks in manufacturing. Yet, often the assembly mechanic cannot be removed entirely from the process. This provides new economic motivation to explore opportunities where assembly mechanics and industrial robots may work in close physical collaboration. In this talk, I present adaptive work-sharing and scheduling algorithms to collaborate with industrial robots on two levels: one-to-one human robot teamwork, and factory-level sequencing and scheduling of human and robotic tasks. I discuss our recent work developing adaptive control methods that incorporate high-level, person-specific planning and execution mechanisms to promote predictable, convergent team behavior. We apply human factors modeling coupled with statistical methods for planning and control to derive quantitative methods for assessing the quality and convergence of learnt teaming models, and to perform risk-sensitive robot control on the production line. I also discuss computationally efficient methods for coordinating human and robotic sequencing and scheduling at the factory-level. Tight integration of human workers and robotic resources involves complex dependencies. Even relatively small increases in process time variability lead to schedule inefficiencies and performance degradation. Our methods allow fast, dynamic computation of robot tasking and scheduling to respond to people working and coordinating in shared physical space, and provide real-time guarantees that schedule deadlines and other operational constraints will be met.

Bio: Julie Shah is an Assistant Professor in the Department of Aeronautics and Astronautics and leads the Interactive Robotics Group of the Computer Science and Artificial Intelligence Laboratory. Shah received her SB (2004) and SM (2006) from the Department of Aeronautics and Astronautics at MIT, and her PhD (2010) in Autonomous Systems from MIT. Before joining the faculty, she worked at Boeing Research and Technology on robotics applications for aerospace manufacturing. She has developed innovative methods for enabling fluid human-robot teamwork in time-critical, safety-critical domains, ranging from manufacturing to surgery to space exploration. Her group draws on expertise in artificial intelligence, human factors, and systems engineering to develop interactive robots that emulate the qualities of effective human team members to improve the efficiency of human-robot teamwork. This work was recognized by the Technology Review as one of the 10 Breakthrough Technologies of 2013, and has received international recognition in the form of best paper awards and nominations from the International Conference on Automated Planning and Scheduling, the American Institute of Aeronautics and Astronautics, the IEEE/ACM International Conference on Human-Robot Interaction, and the International Symposium on Robotics.

01/31/14 Ryan Calo
UW Law
Robotics & The New Cyberlaw

Abstract: The ascendance of the Internet wrought great social, cultural, and economic changes. It also launched the academic movement known as “cyberlaw.” The themes of this movement reflect the essential qualities of the Internet, i.e., the set of characteristics that distinguish the Internet from predecessor and constituent technologies. Now a new set of technologies is ascending, one with arguably different essential qualities. This project examines how the mainstreaming of robotics—for instance, drones and driverless cars—will affect legal and policy discourse, and explores whether cyberlaw is still the right home for the resulting doctrinal and academic conversation.

Bio: Professor Calo researches the intersection of law and emerging technology, with an emphasis on robotics and the Internet. His work on drones, driverless cars, privacy, and other topics has appeared in law reviews and major news outlets, including the New York Times, the Wall Street Journal, and NPR. Professor Calo has also testified before the full Judiciary Committee of the United States Senate.
Professor Calo serves on numerous advisory boards, including the Electronic Privacy Information Center (EPIC), the Electronic Frontier Foundation (EFF), the Future of Privacy Forum, and National Robotics Week. Professor Calo co-chairs the Robotics and Artificial Intelligence committee of the American Bar Association and is a member of the Executive Committee of the American Association of Law Schools (AALS) Section on Internet and Computer Law.
Professor Calo previously served as a director at the Stanford Law School Center for Internet and Society (CIS) where he remains an Affiliate Scholar. He also worked as an associate in the Washington, D.C. office of Covington & Burling LLP and clerked for the Honorable R. Guy Cole on the U.S. Court of Appeals for the Sixth Circuit. Prior to law school at the University of Michigan, Professor Calo investigated allegations of police misconduct in New York City.

02/07/14 James McLurkin
Rice University
Distributed Algorithms for Robot Recovery, Multi-Robot Triangulation, and Advanced Low-Cost Robots: An Overview of the Rice Multi-Robot Systems Lab

Abstract: In this talk we present results from three different projects: 1. A distributed recovery algorithm to extract a multi-robot system from complex environments. The goal is to maintain network connectivity while allowing efficient recovery. Our approach uses a maximal-leaf spanning tree as a communication and navigation backbone, and routes robots along this tree to the goal. Simulation and experimental results demonstrate the efficacy of this approach. 2. Triangulations of regions is a staple technique in almost every geometric computation. When robots triangulate their workspace, they build a "physical data structure" that supports geometric and computational reasoning about the environment using the topology of the triangulated graph. We demonstrate multi-robot triangulation construction, dual-graph navigation, and patrolling using a distributed data structure. 3. We introduce the "r-one" robot, a low-cost design suitable for research, education, and outreach. We provide tales of joy and disaster from using 90 of these platforms for our research, college courses, and museum outreach exhibits.

Bio: James McLurkin is an Assistant Professor at Rice University in the Department of Computer Science. His research focuses on developing distributed algorithms for multi-robot systems, which is software that produces complex group behaviors from the interactions of many simple individuals. These ideas are not new: ants, bees, wasps, and termites have be running this type of software for 120 million years. His research group has one of the largest collections of robots in the world, with over 200 different robots at last count. The new r-one robots are an advanced open-source platform to support the "Robots for Everyone" movement. McLurkin was a Lead Research Scientist at iRobot, and was the 2003 recipient of the Lemelson-MIT student prize for invention. He holds a S.B. in Electrical Engineering with a Minor in Mechanical Engineering from M.I.T., a M.S. in Electrical Engineering from University of California, Berkeley, and a S.M. and Ph.D. in Computer Science from M.I.T.

02/21/14 Mihai Jalobeanu
Microsoft Research
Towards ubiquitous robots

Abstract: Despite significant advances in robotics research, commercial robots are nowhere near as pervasive as we hoped and imagined. This talk explores some of the reasons behind this apparent discrepancy and discusses the approach taken by Microsoft Robotics to bridge the gap between research and commercialization.

Bio: Mihai Jalobeanu leads the Microsoft Robotics development team. He joined Microsoft in 1998 as a software developer and worked on email servers and cloud services before joining the Robotics group in 2012. His interests include software reliability, large scale systems and machine learning, particularly as applied to autonomous navigation and manipulation. Mihai received a B.S. degree in computer engineering in 1996 and a M.S. degree in computer science in 1997, both from Technical University of Cluj-Napoca, Romania.

02/28/14 Cynthia Matuszek
UW, CSE
Talking to Robots: Learning to Ground Human Language in Perception and Execution

Abstract: Advances in computation, sensing, and hardware are enabling robots to perform an increasing variety of tasks in ever less constrained settings. It is now possible to consider near-term robots that will operate in traditionally human-centric spaces. If these robots understand language, they can take instructions and learn about tasks from nonspecialists; at the same time, a robot's real-world interactions can help it learn to better understand physically grounded language. Combining these areas is a fundamentally multidisciplinary problem, involving natural language processing, machine learning, robotics, and human-robot interaction. In this talk, I describe my work on learning natural language in a physical context; such language, learned from end users, allows a person to communicate their needs in a natural, unscripted fashion. I demonstrate that this approach can enable a robot to follow directions, learn about novel objects in the world, and perform simple tasks such as navigating an unfamiliar map or putting away objects, entirely from examples provided by users.

Bio: Cynthia Matuszek is a Ph.D. candidate in the University of Washington Computer Science and Engineering department, where she is a member of both the Robotics and State Estimation lab and the Language, Interaction, and Learning group. She earned a B.S. in Computer Science from the University of Texas at Austin, and M.Sc. from the University of Washington in 2009. She is published in the areas of artificial intelligence, robotics, and human-robot interaction.

03/14/14 Peter H. Kahn, Jr.
UW, Psychology
Social and Moral Relationships with Robots

Abstract: As social robots – and more broadly personified computational environments, as in the smart car and smart home of the future – become more prevalent, they will pose us with significant challenges, socially and morally. In this presentation, I’ll discuss some of my lab’s psychological research on this topic. My lab’s studies are in collaboration with Hiroshi Ishiguro and Takayuki Kanda, using ATR’s humanoid robot, Robovie. I’ll present and show video clips from three empirical studies, where we investigated 3 questions, respectively: (a) Do children and adults believe that humanoid robots can have moral standing? (b) Do adults hold humanoid robots morally accountable for causing harm to humans? and (c) Can adults form psychologically intimate relationships with humanoid robots such that they will keep a robot’s secret from a human experimenter? Then I’ll speak about a current project wherein we seek to provide a new vision for HRI: of how interacting with networked social robots can enhance human creativity. I’ll suggest that in HRI and more broadly HCI we need to hold out a vision of creating technology so that people can flourish. In my view, that involves integrating exponential technological growth with deep authentic connection with other humans, and with a natural world that we're destroying too quickly, and at our peril.

Bio: Peter H. Kahn, Jr. is Professor in the Department of Psychology and Director of the Human Interaction with Nature and Technological Systems (HINTS) Laboratory at the University of Washington. He is also Editor-in-Chief of the academic journal Ecopsychology. His research seeks to address two world trends that are powerfully reshaping human existence: (1) The degradation if not destruction of large parts of the natural world, and (2) unprecedented technological development, both in terms of its computational sophistication and pervasiveness. He received his Ph.D. from the University of California, Berkeley in 1988. His publications have appeared in such journals as Child Development, Developmental Psychology, Human-Computer Interaction, and Journal of Systems Software, as well as in such proceedings as CHI, HRI, and Ubicomp. His 5 books (all with MIT Press) include Technological Nature: Adaptation and the Future of Human Life (2011). His research projects are currently funded by The National Science Foundation (http://faculty.washington.edu/pkahn/).

03/21/14 Gur Kimchi
Amazon
Amazon Prime Air

Abstract: We're excited to share Prime Air — something our team has been working on in our next generation R&D lab right here in Seattle. The goal of this new delivery system is to get packages into customers' hands in 30 minutes or less using unmanned aerial vehicles. Putting Prime Air into commercial use will take some number of years as we advance the technology and work with the Federal Aviation Administration (FAA) on necessary rules and regulations. From a technology point of view, we'll be ready to enter commercial operations as soon as the necessary regulations are in place. One day, Prime Air vehicles will be as normal as seeing mail trucks on the road today.

Bio: Gur Kimchi is the VP of Profit Systems and Prime Air at Amazon.com. Gur joined Amazon’s Worldwide Retail Systems organization in 2012, building key platforms to manage Amazon’s “back office” and automating various retail processes. Gur leads the Prime Air team, a project garnering enormous attention since going public in November of 2013. The goal of Prime Air is to get packages into customers’ hands in 30 minutes or less using unmanned aerial vehicles. Prior to Amazon, Gur spent 10 years at Microsoft on the Contextual Mobile Search team, the MSN/Virtual Earth Core Platform team, and the Unified Communications team. He is a voracious reader, an avid skier, and enjoys spending time with his family.

Autumn 2013
10/11/13 Ashutosh Saxena
Cornell University
How should a robot perceive the world?

In order to perform assistive tasks, a robot should perceive a functional understanding of the environment. This comprises learning how the objects in the environment could be used (i.e., their affordances). In this talk, I will discuss what types of object representations could be useful. One challenge is to model the object's context with each other and with the (hidden) humans. In order to model such data, I will present Infinite Latent CRFs (ILCRFs) that allow modeling the data with different plausible graph structures. Unlike CRFs, where the graph structure is fixed, ILCRFs learn distributions over possible graph structures in an unsupervised manner.
We then show that our idea of modeling environments using object affordances and hidden humans is not only useful for robot manipulation tasks such as arranging a disorganized house, haptic manipulation, and unloading items from a dishwasher, but also in significantly improving standard robotic tasks such as scene segmentation, 3D object detection, human activity detection and anticipation, and task and path planning.

Bio: Ashutosh Saxena is an assistant professor in computer science department at Cornell University. His research interests include machine learning and robotics perception, especially in the domain of personal robotics. He received his MS in 2006 and Ph.D. in 2009 from Stanford University, and his B.Tech. in 2004 from Indian Institute of Technology (IIT) Kanpur. He is a recipient of National Talent Scholar award in India, Google Faculty award, Alfred P. Sloan Fellowship, Microsoft Faculty Fellowship, and NSF Career award.
In the past, Ashutosh developed Make3D (http://make3d.cs.cornell.edu), an algorithm that converts a single photograph into a 3D model. Tens of thousands of users used this technology to convert their pictures to 3D. He has also developed algorithms that enable robots (such as STAIR, POLAR, see http://pr.cs.cornell.edu) to perform household chores such as unload items from a dishwasher, place items in a fridge, etc. His work has received substantial amount of attention in popular press, including the front-page of New York Times, BBC, ABC, New Scientist, Discovery Science, and Wired Magazine. He has won best paper awards in 3DRR, IEEE ACE and RSS, and was named a co-chair of the IEEE technical committee on robot learning.

10/25/13 Kat Steele
UW, Mechanical Engineering
The Ultimate Machine: Strategies for understanding and improving movement disorders

Abstract: The human body is the ultimate machine. With billions of connections, hundreds of actuators, and adaptive learning, the human body provides a unique and versatile platform for us to explore with the world. However, the same complexity that empowers the human body also makes it extremely difficult to treat when things go awry. For individuals with movement disorders, such as cerebral palsy and stroke, the ability to move, manipulate, and interact with the world is impaired and negatively impacts quality of life. In this talk, I will discuss how we have been using a combination of musculoskeletal simulation, medical imaging, and device design to understand how movement is altered after brain injury, evaluate the impacts of current treatments, and design new treatment strategies.

Bio: Kat Steele is an assistant professor in mechanical engineering at the University of Washington. Her research focuses on integrating dynamic simulation, motion analysis, medical imaging, and device designto improve mobility for individuals with movement disorders. She earned her BS in Engineering from the Colorado School of Mines and MS and PhD in Mechanical Engineering from Stanford University. To integrate engineering and medicine, she has worked extensively in hospitals including the Cleveland Clinic, Denver Children’s Hospital, Lucile Packard Children’s Hospital, and, for the past year, the Rehabilitation Institute of Chicago. She has also helped to develop a free, open-source software platform for dynamic simulation of movement (http://opensim.stanford.edu).

*11/07/13
Thursday
Ross A. Knepper
MIT
Autonomous Assembly In a Human World

Abstract: The IkeaBot system autonomously plans and executes furniture assembly by incorporating capabilities for geometric reasoning, symbolic planning, multi-robot coordination, manipulation, and custom, modular tooling. After giving an overview of the basic system, I highlight two recent developments in IkeaBot designed to meet the challenge of operating in complex human environments. The first development, RF-Compass, is a centimeter-scale localization system that is based on inexpensive, off-the-shelf RFID technology. I explain how we overcame several of the challenges of RF-based localization to achieve this unprecedented accuracy. The second development is a system for handling the failures that inevitably occur during autonomous planning and execution. Failures take many forms, including errors in perception, reasoning, and action, as well as fundamental limitations of the robot hardware. Autonomous systems must detect, diagnose, and remedy failures when they occur. In cases where the robot cannot remedy the problem autonomously, it may ask a human to assist. It generates natural language help requests targeted at humans who may lack situational awareness of the failure or even the task. By grounding candidate requests in salient features of the environment and modeling how they would be understood by a human, we select the request that minimizes ambiguity.

Bio: Ross A. Knepper is a Research Scientist in the Distributed Robotics Laboratory at the Massachusetts Institute of Technology. His research focuses on the theory and algorithms of automated assembly. Taking IKEA furniture assembly as a challenge problem, he is exploring motion and task planning, manipulation, custom tooling, coordination, localization, failure handling, and human-robot interaction. Ross received his M.S and Ph.D. degrees in Robotics from Carnegie Mellon University in 2007 and 2011. Before his graduate education, Ross worked in industry at Compaq, where he designed high-performance algorithms for scalable multiprocessor systems; and also in commercialization at the National Robotics Engineering Center, where he adapted robotics technologies for customers in government and industry. Ross has served as a volunteer for Interpretation at Death Valley National Park, California.

*11/08/13
12-1pm
Brian Ziebart
University of Illinois, Chicago
Beyond Conditionals: Structured Prediction for Interacting Processes

Abstract: The principle of maximum entropy provides a powerful framework for estimating joint, conditional, and marginal probability distributions. Markov random fields and conditional random fields can be viewed as the maximum entropy approach in action. However, beyond joint and conditional distributions, there are many other important distributions with elements of interaction and feedback where its applicability has not been established. In this talk, I will present the principle of maximum causal entropy—an approach based on directed information theory for estimating an unknown process based on its interactions with a known process. I will discuss applications of this approach to assistive technologies and human-robot interaction.

Bio: Brian Ziebart is an Assistant Professor in the Department of Computer Science at the University of Illinois at Chicago. He received his PhD in Machine Learning from Carnegie Mellon University in 2010, where he was also a postdoctoral fellow. His research has been recognized with best paper awards, runner-ups, and nominations at ICML (2010, 2011), ECCV (2012), and IUI (2012).

11/01/13 Maya Cakmak
University of Washington
Towards Seamless Human-Robot Hand-overs

Abstract: Handing over different objects to humans is a key functionality for robots that will assist or cooperate with humans. A robot could fetch objects for elderly living in their homes or hand tools to a worker in a factory. While there are infinite ways that a robot can transfer an object to a human, including very simple ones, achieving this seamlessly, like humans and objects to humans, is a challenge. This talk overviews two research projects that aim at characterizing robot hand-over actions that result in seamless object transfer. The first focuses on the efficiency and fluency of the hand-over and explores the notion of contrast in the hand-over action. The second focuses on the ease with which the object can be taken when presented by the robot and explores learning from demonstration to learn appropriate hand-over configurations. I present empirical results from human-robot interaction studies in both projects and conclude with recommendations for designing robot hand-over behaviors.

Bio: Maya Cakmak is an Assistant Professor in Computer Science and Engineering at the University of Washington. She received her Ph.D. in Robotics from the Georgia Institute of Technology in 2012 and she was a post-doctoral research fellow at Willow Garage afterwards. Maya's research aims to develop functionalities and interfaces for personal robots that can be programmed by their end-users to assist everyday tasks. Her work has been published at major Robotics and AI conferences and journals and has been featured in numerous media outlets.

11/08/13 Jenay Beer
University of South Carolina
Considerations for Designing Assistive Robotics to Promote Aging-in-Place

Abstract: Many older adults wish to age-in-place, that is, to remain in their own homes as they age. However, challenges threaten an older adult’s ability to age-in-place. In fact, even healthy independently living older adults experience challenges in maintaining their home. Challenges with aging in place can be compensated through technology, such as home assistive robots. However, for home robots to be adopted by older adult users they must be designed to meet older adults’ needs for assistance and the older users must be amenable to robot assistance for those needs. I will discuss a range of projects (both quantitative and qualitative in nature) assessing older adults’ social interpretation, attitudes, and acceptance of assistive robotics. Study findings suggest that older adults’ assistance preferences discriminated between tasks, and the data suggest insights as the why older adults hold such preferences. The talk will detail multidisciplinary approaches to studying human-robot interaction (HRI) and how findings from user studies can apply to preliminary design recommendations for future assistive robots to support aging-in-place.

Bio: Jenay Beer is an Engineering Psychologist and an Assistant Professor with a joint appointment in the Department of Computer Science and Engineering and the College of Social Work at the University of South Carolina. She is the director of the Assistive Robotics and Technology Lab, and a member of the USC SeniorSmart initiative. Her research intersects the fields of Human Robot Interaction (HRI) and Psychology. Specifically, she studies home-based robots designed to assist older adults to maintain their independence and age-in-place. She has studied a variety of robotic systems and topics such as emotion expression of agents, user acceptance of robots, healthcare robotics, and the role of robot autonomy in HRI.
Jenay has published and presented at a number of major human factors- and HRI-related conferences, including Human-Robot Interaction and Human Factors and Ergonomics Society. Her work has been featured with TEDxGeorgiaTech 2012, CNET, and WABE NPR News. She has been awarded the American Psychological Association (APA) Early Graduate Student Researcher Award in 2010, and has been selected as a Georgia Tech Foley Scholar Finalist two years in a row, 2011 and 2012. Jenay received a B.A. degree in Psychology from the University of Dayton, Ohio, in 2006. She also earned an M.S. and Ph.D. in Engineering Psychology from the Georgia Institute of Technology in 2010 and 2013 respectively.

11/15/13 Dinei Florencio
Microsoft Research
Navigation for telepresence robots and some thoughts on robot learning

Abstract: This informal talk will be divided in three parts, corresponding to three papers (IROS’12, ICRA’13 and IROS’13) that cover work on telepresence robots, and learning aspects of HRI. We first present a method for a mobile robot to follow a person autonomously where there is an interaction between the robot and human during following. Contrary to traditional motion planning, instead of determining goal points close to the person, we introduce a task dependent goal function which provides a map of desirable areas for the robot to be at, with respect to the person. We implemented our approach on a telepresence robot and conducted a controlled user study to evaluate the experiences of the users on the remote end of the telepresence robot.

On the second part, we investigate “semi-autonomous driving” for a telepresence robot. Traditional aided driving is mostly based on “collision avoidance”, i.e., it limits or avoids movements that would lead to a collision. Instead, we borrow concepts from collaborative driving, and use the input from the operator as a general guidance to the target direction, then couple that with a variable degree of autonomy to the robot, depending on the task and the environment.

Finally, in the third part possible we investigate the problem of making a robot learn how to approach a person in order to increase the chance of a successful engagement. We propose the use of Gaussian Process Regression (GPR), combined with ideas from reinforcement learning to make sure the space is properly and continuously explored. In the proposed example scenario, this is used by the robot to predict the best decisions in relation to its position in the environment and approach distance, each one accordingly to a certain time of the day.

Bio: Dinei Florêncio received the B.S. and M.S. from University of Brasília (Brazil), and the Ph.D. from Georgia Tech, all in Electrical Engineering. He is a researcher with Microsoft Research since 1999, currently with the Multimedia, Interaction, and Communication group. From 1996 to 1999, he was a member of the research staff at the David Sarnoff Research Center. He was also a co-op student with AT&T Human Interface Lab (now part of NCR) from 1994 to 1996, and a summer intern at the (now defunct) Interval Research in 1994.
Dr. Florencio’s current research focus includes signal processing and computer security. In the area of signal processing, he works in audio and video processing, with particular focus to real time communication. He has numerous contributions in Speech Enhancement, Microphone arrays, Image and video coding, Spectral Analysis, and non-linear algorithms. In the area of computer security, his interest focuses in cybercrime and problems that can be assisted by algorithmic research. Topics include phishing prevention, user authentication, sender authentication, human interactive proofs, and economics of cybercrime.
Dr. Florencio has published over 100 referred papers, and 50 granted US patents (with another 13 currently pending). His papers received awards at MMSP’09, ICME’2010, SOUPS’2010, MMSP’12. He also received the 1998 Sarnoff Achievement Award, the 1996 SAIC best paper award, and an NCR inventor award. His research has enhanced the lives of millions of people, through high impact technology transfers to many Microsoft products, including Live Messenger, Exchange Server, RoundTable, and the MSN toolbar. Dr. Florencio was general co-chair of CBSP’08, MMSP'09, Hot3D’10, and WIFS’11, and technical chair of WIFS’10, ICME’11, and MMSP’13.
Dr. Florencio is a senior member of the IEEE, an elected member of the IEEE Information Forensics and Security Technical Committee, and of the IEEE Multimedia and Signal Processing Technical Committee (for which he will serve as chair for 2014-15). He is also a member of the IEEE ICME steering committee, and an associate editor of the IEEE Transactions on Information Forensics and Security.

11/22/13 Andrzej Pronobis
University of Washington, CSE
Semantic Knowledge in Mobile Robotics: Perception, Reasoning, Communication and Actions

Abstract: As robotic technologies mature, we can imagine an increasing number of applications in which robots would be useful in human environments and in human-robot collaborative scenarios. In fact, many believe that it is in the area of service and domestic robotics that we will see the largest growth within the next few years. A fundamental capability for such systems is to understand the dynamic, complex and unstructured human environments in which they are to operate. Such understanding is not only crucial for solving typical human tasks efficiently. More importantly, it can support communication and cooperation with untrained human users. In this talk, I will discuss how such understanding can be achieved by combining uncertain multi-modal robotic perception with a probabilistic relational representation of human semantic concepts transferred from Internet databases. I will present a semantic mapping algorithm combining information such as object observations, shape, appearance of rooms and human input with conceptual common-sense knowledge and show its ability to infer semantic room categories, predict existence of objects as well as reason about unexplored space. Furthermore, I will show that exploiting uncertain semantics can lead to more efficient strategies for solving real-world problems in large-scale realistic environments previously unknown to the robot. Finally, I will highlight our current work on integration of semantic spatial understanding and reasoning with language- and gesture-based human-robot interaction.

Bio: Andrzej Pronobis is a research associate in the Robotics and State Estimation Lab at the University of Washington working with Dieter Fox. His research is focused on the development of perception and spatial understanding mechanisms for mobile robots and their interplay with components responsible for interaction with the world and human users. Before joining UW, he was the Head of Research at OculusAI Technologies AB, a Swedish company developing mobile, cloud-based computer vision solutions. Between 2006 and 2012, he was involved in three large EU robotics research initiatives CogX, CoSy, and DIRAC. He obtained his PhD in Computer Vision and Robotics from KTH Royal Institute of Technology, Stockholm, Sweden in 2011 and his M.Sc. in Computer Science from the Silesian University of Technology, Gliwice, Poland in 2005. He is an author of 40 publications in the areas of robotics, computer vision and machine learning and an organizer or several international events related to robotics and computer vision research, including workshops and competitions.

12/06/13 Steve Cousins
CEO of Savioke and Willow Garage
It's Time for Service Robots

Abstract: Willow Garage laid the foundation for a new industry in personal robotics by creating the Robot Operating System and spinning off a number of companies to seed the new industry. This new industry has the potential to change the world the way the PC industry did 30 years ago. We have very capable personal robots becoming available at ever-decreasing price points, and new low-cost sensors and actuators all the time. What has changed in the 30 years since the IBM PC was introduced is our understanding of the value of open source software, the power of crowd-sourcing, and our ability to personalize to satisfy the "long tail" of demand. Like the PC, personal robots will probably make their debut in businesses and then later find their way into homes. The first ones will likely show up in the service industry.

Bio: Steve Cousins, CEO, was formerly the President and CEO of Willow Garage. During his tenure, Willow Garage created the PR2 robot, the open source TurtleBot, and the robot operating system (ROS), and spun off 8 companies:
* Suitable Technologies (maker of the Beam remote presence system)
* Industrial Perception, Inc.
* Redwood Robotics
* HiDOF (ROS and robotics consulting)
* Unbounded Robotics
* The Open Source Robotics Foundation
* The OpenCV Foundation
* The Open Perception Foundation
Before joining Willow Garage, Steve was a senior manager at IBM's Almaden Research Center, and a member of the senior staff at Xerox PARC. Steve holds a Ph.D. from Stanford University, and BS and MS degrees in computer science from Washington University.

Spring 2013
04/05/13 Dieter Fox
Cynthia Matuszek
PechaKucha 20x20 for Robotics

PechaKucha 20x20 is a new approach to giving presentations. From the FAQ: "PechaKucha 20x20 is a simple presentation format where you show 20 images, each for 20 seconds. The images advance automatically and you talk along to the images." While the presentation format was originally developed for architecture presentations, it has been successfully applied to fields as diverse as art, cooking, design, and journalism. This talk will give an overview of the format and some examples, in the interest of stimulating discussion about the role of such a format in technology.

04/19/13 Robotics Students & Staff PechaKucha-style Robotics Research Overviews

In this session, eight interested (and interesting!) robotics researchers will use one of the popular "Flash presentation" styles to spend a few minutes covering information about their own work, related work that they think is worth knowing about, or any other robotics-related topic they wish.

04/24/13 Pete Wurman Coordinating Hundreds of Autonomous Vehicles in Warehouses
Special Wednesday Colloquium, CSE 203

Kiva's mobile fulfillment system blends techniques from AI, Controls Systems, Machine Learning, Operations Research and other engineering disciplines into the world's largest mobile robotic platform. Kiva uses hundreds of mobile robots to carry inventory shelves around distribution centers for customers like Staples, Walgreens, and The Gap. Kiva currently has equipment in over 30 warehouses in three countries. This talk will describe the application domain and the business solution, and some of the practical engineering problems that Kiva has solved along the way.

04/26/13 Matt Mason Learning to Use Simple Hands

We often assume that general-purpose robot hands should be complex, perhaps even as complex as human hands. Yet humans can do a lot even when using tongs. This talk describes ongoing work with simple hands - hands inspired by very simple tools like tongs. We explore a robot's ability to grasp, recognize, localize, place and even manipulate objects in the hand, with a very simple hand. The perception and planning algorithms are based on learned models, which are in turn based on thousands of experiments with the objects in question.

Dr. Matthew T. Mason earned the BS, MS, and PhD degrees in Computer Science and Artificial Intelligence at MIT, finishing his PhD in 1982. Since that time he has been on the faculty at Carnegie Mellon University, where he is presently Professor of Robotics and Computer Science, and Director of the Robotics Institute. His prior work includes force control, automated assembly planning, mechanics of pushing and grasping, automated parts orienting and feeding, and mobile robotics. He is co-author of "Robot Hands and the Mechanics of Manipulation" (MIT Press 1985), co-editor of "Robot Motion: Planning and Control" (MIT Press 1982), and author of "Mechanics of Robotic Manipulation" (MIT Press 2001). He is a Fellow of the AAAI, and a Fellow of the IEEE. He is a winner of the System Development Foundation Prize and the IEEE Robotics and Automation Society's Pioneer Award.

05/17/13 Tom Daniel Control and Dynamics of Animal Flight: Reverse Engineering Nature's Robots

All living creatures process information from multiple sensory modalities and, in turn, control movement through multiple actuators. They do so to navigate through spatially and temporally complex environments with amazing agility. Among the most successful of nature's robots are insects, occupying every major habitat. This talk will review sensorimotor control of movement in flying insects, with a focus on where the functional role of sensing and actuation become blurred.

Dr. Tom Daniel holds the Joan and Richard Komen Endowed Chair and has appointments in the Department of Biology, Computer Science & Engineering, the Program on Neurobiology and Behavior Faculty at the University of Washington. He is currently the Interim Director of the National Science Foundation Center for Sensorimotor Neural Engineering (CSNE). He has served as a UW faculty member since his initial appointment in 1984. He was the founding chair of the Department of Biology at the University of Washington (2000-2008). Prior to the UW, he was the Myron A. Bantrell Postdoctoral Fellow in Engineering Sciences at the California Institute of Technology. He received his PhD degree from Duke University. He was awarded MacArthur Fellow in 1996, the University of Washington Distinguished Teaching Award, and the University of Washington Distinguished Graduate Mentor Award. He is on the editorial boards of the Science Magazine, Proceedings of the Royal Society (Biology Letters). He is also on the Board of Directors and the Scientific Advisory Board of the Allen Institute of Brain Science, and the Scientific Advisory Board for the NSF Mathematical Biosciences Institutes. His research programs focus on biomechanics, neurobiology, and sensory systems, addressing questions about the physics, engineering and neural control of movement in biological systems.

05/24/13 Katherine Kuchenbecker The Value of Tactile Sensations in Haptics and Robotics

Although physical interaction with the world is at the core of human experience, few computer and machine interfaces provide the operator with high-fidelity touch feedback, limiting their usability. Similarly, autonomous robots rarely take advantage of touch perception and thus struggle to match the manipulation capabilities of humans. My long-term research goal is to leverage scientific knowledge about the sense of touch to engineer haptic interfaces and robotic systems that increase the range and quality of tasks humans can accomplish. This talk will describe my group's three main research thrusts: haptic texture rendering, touch feedback for robotic surgery, and touch perception for autonomous robots. First, most haptic interfaces struggle to mimic the feel of a tool dragging along a surface due to both software and hardware limitations. We pioneered a data-driven method of capturing and recreating the high-bandwidth vibrations that characterize tool-mediated interactions with real textured surfaces. Second, although commercial robotic surgery systems are approved for use on human patients, they provide the surgeon with little to no haptic feedback. We have invented, refined, and studied a practical method for giving the surgeon realistic tactile feedback of instrument vibrations during robotic surgery. Third, household robots will need to know how to grasp and manipulate a wide variety of objects. We have invented a set of methods that enable a robot equipped with commercial tactile sensors to delicately and firmly grasp real-world objects and perceive their haptic properties. Our work in all three of these areas has been principally enabled by a single insight: although less studied than kinesthetic cues, tactile sensations convey much of the richness of physical interactions.

Dr. Katherine J. Kuchenbecker is the Skirkanich Assistant Professor of Innovation in Mechanical Engineering and Applied Mechanics at the University of Pennsylvania. Her research centers on the design and control of haptic interfaces for applications such as robot-assisted surgery, medical simulation, stroke rehabilitation, and personal computing. She directs the Penn Haptics Group, which is part of the General Robotics, Automation, Sensing, and Perception (GRASP) Laboratory. She has won several awards for her research, including an NSF CAREER Award in 2009, Popular Science Brilliant 10 in 2010, and the IEEE Robotics and Automation Society Academic Early Career Award in 2012. Prior to becoming a professor, she completed a postdoctoral fellowship at the Johns Hopkins University, and she earned her Ph.D. in Mechanical Engineering at Stanford University in 2006.

05/31/13 Pieter Abbeel Machine Learning and Optimization for Robotics

Robots are typically far less capable in autonomous mode than in teleoperated mode. The few exceptions tend to stem from long days (and more often weeks, or even years) of expert engineering for a specific robot and its operating environment. Current control methodology is quite slow and labor intensive. I believe advances in machine learning and optimization have the potential to revolutionize robotics. First I will present new machine learning techniques we have developed that are tailored to robotics. I will describe in depth "Apprenticeship learning," a new approach to high-performance robot control based on learning for control from ensembles of expert human demonstrations. Our initial work in apprenticeship learning has enabled the most advanced helicopter aerobatics to-date, including maneuvers such as chaos, tic-tocs, and auto-rotation landings which only exceptional expert human pilots can fly. Our most recent work in apprenticeship learning is inspired by challenges in surgical robotics. We are studying how a robot could learn to perform challenging robotic manipulation tasks, such as knot-tying. Then I will describe our recent advances in optimization based planning — both in state space and in belief space. Finally, I will briefly highlight our recent work on enabling robots to learn on their own through non-parametric model-based reinforcement learning.

Dr. Pieter Abbeel received a BS/MS in Electrical Engineering from KU Leuven (Belgium) and received his Ph.D. degree in Computer Science from Stanford University in 2008. He joined the faculty at UC Berkeley in Fall 2008, with an appointment in the Department of Electrical Engineering and Computer Sciences. He has won various awards, including best paper awards at ICML and ICRA, the Sloan Fellowship, the Air Force Office of Scientific Research Young Investigator Program (AFOSR-YIP) award, the Office of Naval Research Young Investigator Program (ONR-YIP) award, the Okawa Foundation award, the TR35, the IEEE Robotics and Automation Society (RAS) Early Career Award, and the Dick Volz Best U.S. Ph.D. Thesis in Robotics and Automation Award. He has developed apprenticeship learning algorithms which have enabled advanced helicopter aerobatics, including maneuvers such as tic-tocs, chaos and auto-rotation, which only exceptional human pilots can perform. His group has also enabled the first end-to-end completion of reliably picking up a crumpled laundry article and folding it. His work has been featured in many popular press outlets, including BBC, New York Times, MIT Technology Review, Discovery Channel, SmartPlanet and Wired. His current research focuses on robotics and machine learning with a particular emphasis on challenges in personal robotics, surgical robotics and connectomics.

Winter 2013
01/25/13 Joshua Smith Robotics Research in the Sensor Systems Group

After providing a brief overview of the Sensor Systems group, I will present our recent work in robotics. I will introduce pretouch sensing, our term for in-hand sensing that is shorter range than vision but longer range than tactile sensing. I will review Electric Field Pretouch sensing, introduce Seashell Effect Pretouch, and discuss strategies for using pretouch sensing in the context of robotic manipulation. As an active sensing modality, pretouch requires a choice of "next view." Since the robot hand is used for both sensing and actuation, pretouch-enabled grasping also requires us to consider an exploration/execution tradeoff. Finally, I will outline several new robotics projects that are underway.

02/08/13 Gaurav Sukhatme Persistent Autonomy at Sea

Underwater robotics is undergoing a transformation. Recent advances in AI and machine learning are enabling a new generation of underwater robots to make intelligent decisions (where to sample ? how to navigate ?) by reasoning about their environment (what is the shipping and water forecast ?). At USC, we are engaged in a long-term effort to develop persistent, autonomous underwater explorer robots. In this talk, I will give an overview of some of our recent results focusing on two problems in adaptive sampling: underwater change detection and biological sampling. I will also present our recent work on hazard avoidance, allowing robots to operate in regions where there is ship traffic. Bio: Gaurav S. Sukhatme is a Professor of Computer Science (joint appointment in Electrical Engineering) at the University of Southern California (USC). He is currently serving as the Chairman of the Computer Science department. His recent research is in networked robots.

Dr. Sukhatme has served as PI on numerous federal grants. He is Fellow of the IEEE and a recipient of the NSF CAREER award and the Okawa foundation research award. He is one of the founders of the RSS conference and has served as program chair of all three leading robotics conferences (ICRA, IROS and RSS). He is the Editor-in-Chief of the Springer journal Autonomous Robots.

02/15/13 Jiri Najemnik Sequence Optimization in Engineering, Artificial Intelligence and Biology

Part 1: Linear equivalent of dynamic programming. We show that Bellman equation for dynamic programming can be replaced by just as simple linear equation for the so-called optimal ranking function, which encodes the optimal sequence via its greedy maximization. This optimal ranking function represents Gibbs distribution which minimizes the expected sequence cost given the entropy level (set by a temperature parameter). Each temperature level gives rise to a linearly computable optimal ranking function.

Part 2: Predictive state representation with entropy level constraint. Building on part 1, we show that if one specifies the entropy level of the input's stochastic process, then its Bayesian inference for the purposes of optimal learning can be simplified greatly. We conceptualize an idealized nervous system that is an online input-output transformer of binary vectors representing the neurons' firing states, and we ask how one would adjust the input-output mapping optimally to minimize the expected cost. We will argue that predictive state representations could be employed by a nervous system.

Part 3: Evidence of optimal predictive control of human eyes. We present evidence of optimal-like predictive control of human eyes in visual search for a small camouflaged target. To a striking degree, human searchers behave as if maintaining a map of beliefs (represented as probabilities) about the target location, updating their beliefs with visual data obtained on each fixation using the Bayes Rule, and moving eyes online in order to maximize the expected information gain. Some of these results were published in Nature.

02/15/13 Richard Newcombe Beyond Point Clouds: Adventures in Real-time Dense SLAM

One clear direction for the near future of robotics makes use of the ability to build and keep up to date geometric models of the environment. In this talk I will present an overview of my work in monocular real-time dense surface SLAM (simultaneous localisation and mapping) which aims to provide such geometric models using only a single passive colour or depth camera and without further specific hardware or infrastructure requirements. In contrast to previous SLAM systems which utilised sparser point cloud scene representations, the systems I will present, which include KinectFusion and DTAM, simultaneously estimate a camera pose together with a full dense surface estimate of the scene. Such dense surface mapping results in physically predictive models that are more useful for geometry aware augmented reality and robotics applications. Crucially, representing the scene using surfaces enables elegant dense image tracking techniques to be used in estimating the camera pose, resulting in robustness to high speed agile camera motion. I'll provide a real-time demonstration of these techniques which are useful not only in robust camera tracking, but also in object tracking in general. Finally, I'll outline our latest work in moving beyond surface estimation to incorporating objects into the dense SLAM pipeline.

02/15/13 Tom Erez Model-Based Optimization for Intelligent Robot Control

Science-fiction robots can perform any task humans do and more. In reality, however, today's articulated robots are disappointingly limited in their motor skills. Current planning and control algorithms cannot provide the robot with the capacity for intelligent motor behavior - instead, control engineers must manually specify the motions of every task. This approach results in jerky motions (popularly stereotyped as “moving like a robot”) that cannot cope with unexpected changes. I study control methods that automate the job of the controls engineer. I give the robot only a cost function that encodes the task in high-level terms: move forward, remain upright, bring an object, etc. The robot uses a model of itself and its surroundings to optimize its behavior, finding a solution that minimizes the future cost. This optimization-based approach can be applied to different problems, and in every case the robot alone decides how to solve the task. Re-optimizing in real time allows the robot to deal with unexpected deviations from the plan, generating robust and creative behavior that adapts to modeling errors and dynamic environments. In this talk, I will present the theoretic and algorithmic aspects needed to control articulated robots using model-based optimization. I will discuss how machine learning can be used to create better controllers, and share some of my work on trajectory optimization.

A preview of some of the work discussed in this talk can be seen here: https://dl.dropbox.com/u/57029/MedleyJan13.mp4 [a lower-quality version is also available on youtube: http://www.youtube.com/watch?v=t4JdSklL8w0 ]

02/15/13 Byron Boots Spectral Approaches to Learning Dynamical Systems

If we hope to build an intelligent agent, we have to solve (at least!) the following problem: by watching an incoming stream of sensor data, hypothesize an external world model which explains that data. For this purpose, an appealing model representation is a dynamical system. Sometimes we can use extensive domain knowledge to write down a dynamical system, however, for many domains, specifying a model by hand can be a time consuming process. This motivates an alternative approach: *learning* a dynamical system directly from sensor data. A popular assumption is that observations are generated from a hidden sequence of latent variables, but learning such a model directly from sensor data can be tricky. To discover the right latent state representation and model parameters, we must solve difficult temporal and structural credit assignment problems, often leading to a search space with a host of (bad) local optima. In this talk, I will present a very different approach. I will discuss how to model a dynamical system's belief space as a set of *predictions* of observable quantities. These so-called Predictive State Representations (PSRs) are very expressive and subsume popular latent variable models including Kalman filters and input-output hidden Markov models. One of the primary advantages of PSRs over latent variable formulations of dynamical systems is that model parameters can be estimated directly from moments of observed data using a recently discovered class of spectral learning algorithms. Unlike the popular EM algorithm, spectral learning algorithms are statistically consistent, computationally efficient, and easy to implement using established matrix-algebra techniques. The result is a powerful framework for learning dynamical system models directly from data.

Spring 2012
3/08/12 Andrea Thomaz Designing Learning Interactions for Robots

In this talk I present recent work from the Socially Intelligent Machines Lab at Georgia Tech. One of the focuses of our lab is on Socially Guided Machine Learning, building robot systems that can learn from everyday human teachers. We look at standard Machine Learning interactions and redesign interfaces and algorithms to support the collection of learning input from naive humans. This talk starts with an initial investigation comparing self and social learning which motivates our recent work on Active Learning for robots. Then, I will present results from a study of robot active learning, which motivates two challenges: getting interaction timing right, and asking good questions. To address the first challenge we are building computational models of reciprocal social interactions. And to address the second challenge we are developing algorithms for generating Active Learning queries in embodied learning tasks.

Dr. Andrea L. Thomaz is an Assistant Professor of Interactive Computing at the Georgia Institute of Technology. She directs the Socially Intelligent Machines lab, which is affiliated with the Robotics and Intelligent Machines (RIM) Center and with the Graphics Visualization and Usability (GVU) Center. She earned a B.S. in Electrical and Computer Engineering from the University of Texas at Austin in 1999, and Sc.M. and Ph.D. degrees from MIT in 2002 and 2006. Dr. Thomaz is published in the areas of Artificial Intelligence, Robotics, Human-Robot Interaction, and Human-Computer Interaction. She received an ONR Young Investigator Award in 2008, and an NSF CAREER award in 2010. Her work has been featured on the front page of the New York Times, and in 2009 she was named one of MIT Technology Review’s TR 35.

4/6/12 Javier Movellan Towards a New Science of Learning

Advances in machine learning, machine perception, neuroscience, and control theory are making possible the emergence of a new science of learning. This discipline could help us understand the role of learning in the development of human intelligence, and to create machines that can learn from experience and that can accelerate human learning and education. I will propose that key to this emerging science is the commitment to computational analysis, for which the framework of probability theory and stochastic optimal control is particularly well suited, and to the testing of theories using physical real time robotic implementations. I will describe our efforts to help understand learning and development from a computational point of view. This includes development of machine perception primitives for social interaction, development of social robots to enrich early childhood education, computational analysis of rich databases of early social behavior, and development of sophisticated humanoid robots to understand the emergence of sensory-motor intelligence in infants.

4/13/12 Emanuel Todorov Automatic Synthesis of Complex Behaviors with Optimal Control

In this talk I will show videos of complex motor behaviors synthesized automatically using new optimal control methods, and explain how these methods work. The behaviors include getting up from an arbitrary pose on the ground, walking, hopping, swimming, kicking, climbing, hand-stands, and cooperative actions. The synthesis methods fall in two categories. The first is online trajectory optimization or model-predictive control (MPC). The idea is to optimize the movement trajectory at every step of the estimation-control loop up to some time horizon (in our case about half a second), execute only the beginning portion of the trajectory, and repeat the optimization at the next time step (say 10 msec later). This approach has been used extensively in domains such as chemical process control where the dynamics are sufficiently slow and smooth to make online optimization possible. We have now developed a number of algorithmic improvements, allowing us to apply MPC to robotic systems. This requires a fast physics engine (for computing derivatives via finite differencing) which we have also developed. The second method is based on the realization that most movements performed on land are made for the purpose of establishing contact with the environment, and exerting contact forces. This suggests that contact events should not be treated as side-effects of multi-joint kinematics and dynamics, but rather as explicit decision variables. We have developed a method where the optimizer directly specifies the desired contact events, using continuous decision variables, and at the same time optimizes the movement trajectory in a way consistent with the specified contact events. This makes it possible to optimize movement trajectories with many contact events, without need for manual scripting, motion capture or fortuitous choice of "features".

4/20/12 Andrew Barto Autonomous Robot Acquisition of Transferable Skills

A central goal of artificial intelligence is the design of agents that can learn to achieve increasingly complex behavior over time. An important type of cumulative learning is the acquisition of procedural knowledge in the form of skills, allowing an agent to abstract away from low-level motor control and plan and learn at a higher level, and thus progressively improving its problem solving abilities and creating further opportunities for learning. I describe a robot system that learns to sequence innate controllers to solve a task, and then extracts components of that solution as transferable skills. The resulting skills improve the robot’s ability to learn to solve a second task. This system was developed by Dr. George Konidaris, who received the Ph.D. from the University of Massachusetts Amherst in 2010 and is currently a Postdoctoral Associate in the Learning and Intelligent Systems Group in the MIT Computer Science and Artificial Intelligence Laboratory.

4/27/12 Dieter Fox Grounding Natural Language in Robot Control Systems

Robots are becoming more and more capable at reasoning about people, objects, and activities in their environments. The ability to extract high-level semantic information from sensor data provides new opportunities for human-robot interaction. One such opportunity is to explore interacting with robots via natural language. In this talk I will present our preliminary work toward enabling robots to interpret, or ground, natural language commands in robot control systems. We build on techniques developed by the semantic natural language processing community on learning grammars that parse natural language input to logic-based semantic meaning. I will demonstrate early results in two application domains: First, learning to follow natural language directions through indoor environments; and, second, learning to ground (simple) object attributes via weakly supervised training. Joint work with Luke Zettlemoyer, Cynthia Matuszek, Nicholas Fitzgerald, and Liefeng Bo. Support provided by Intel ISTC-PC, NSF, ARL, and ONR.

5/4/12 Allison Okamura Robot-Assisted Needle Steering

Robot-assisted needle steering is a promising technique to improve the effectiveness of needle-based medical procedures by allowing redirection of a needle's path within tissue. Our robot employs a tip-based steering technique, in which the asymmetric tips of long, thin, flexible needles develop tip forces orthogonal to the needle shaft due to interaction with surrounding tissue. The robot steers a needle though two input degrees of freedom, insertion along and rotation about the needle shaft, in order to achieve six-degree-of-freedom positioning of the needle tip. A closed-loop system for asymmetric-tip needle steering was developed, including devices, models and simulations, path planners, controllers, and integration with medical imaging. I will present results from testing needle steering in artificial and biological tissues, and discuss ongoing work toward clinical applications. This project is a collaboration between researchers at Johns Hopkins University, UC Berkeley, and Stanford University.

Dr. Allison M. Okamura received the BS degree from the University of California at Berkeley in 1994, and the MS and PhD degrees from Stanford University in 1996 and 2000, respectively, all in mechanical engineering. She is currently Associate Professor in the mechanical engineering department at Stanford University. She was previously Professor and Vice Chair of mechanical engineering at Johns Hopkins University. She has been an associate editor of the IEEE Transactions on Haptics, an editor of the IEEE International Conference on Robotics and Automation Conference Editorial Board, and co-chair of the IEEE Haptics Symposium. Her awards include the 2009 IEEE Technical Committee on Haptics Early Career Award, the 2005 IEEE Robotics and Automation Society Early Academic Career Award, and the 2004 NSF CAREER Award. She is an IEEE Fellow. Her interests include haptics, teleoperation, virtual environments and simulators, medical robotics, neuromechanics and rehabilitation, prosthetics, and engineering education. For more information about our work, please see the Collaborative Haptics and Robotics in Medicine (CHARM) Laboratory website: http://charm.stanford.edu.

5/11/12 Blake Hannaford Click the Scalpel -- Better Patient Outcomes by Advancing Robotics in Surgery Surgery is a demanding unstructured physical manipulation task involving highly trained humans, advanced tools, networked information systems, and uncertainty. This talk will review engineering and scientific research at the University of Washington Biorobotics Lab, aimed at better care of patients, including remote patients in extreme environments. The Raven interoperable robot surgery research system is a telemanipulation system for exploration and training in surgical robotics. We are currently near completion of seven "Raven-II" systems which will be deployed at leading surgical robotics research centers to create an interoperable network of testbeds. Highly effective and safe surgical teleoperation systems of the future will provide high quality haptic feedback. Research in systems theory and human perception addressing that goal will also be introduced.

Dr. Blake Hannaford, Ph.D., is Professor of Electrical Engineering, Adjunct Professor of Bioengineering, Mechanical Engineering, and Surgery at the University of Washington. He received the B.S. degree in Engineering and Applied Science from Yale University in 1977, and the M.S. and Ph.D. degrees in Electrical Engineering from the University of California, Berkeley, in 1982 and 1985 respectively. Before graduate study, he held engineering positions in digital hardware and software design, office automation, and medical image processing. At Berkeley he pursued thesis research in multiple target tracking in medical images and the control of time-optimal voluntary human movement. From 1986 to 1989 he worked on the remote control of robot manipulators in the Man-Machine Systems Group in the Automated Systems Section of the NASA Jet Propulsion Laboratory, Caltech. He supervised that group from 1988 to 1989. Since September 1989, he has been at the University of Washington in Seattle, where he has been Professor of Electrical Engineering since 1997, and served as Associate Chair for Education from 1999 to 2001. He was awarded the National Science Foundation's Presidential Young Investigator Award and the Early Career Achievement Award from the IEEE Engineering in Medicine and Biology Society and is an IEEE Fellow. His currently active interests include haptic displays on the Internet, and surgical robotics. He has consulted on robotic surgical devices with the Food and Drug Administration Panel on surgical devices.

5/25/12 Malcolm MacIver Robotic Electrolocation Electrolocation is used by the weakly electric fish of South America and Africa to navigate and hunt in murky water where vision is ineffective. These fish generate an AC electric field that is perturbed by objects nearby that differ in impedance from the water. Electroreceptors covering the body of the fish report the amplitude and phase of the local field. The animal decodes electric field perturbations into information about its surroundings. Electrolocation is fundamentally divergent from optical vision (and other imaging methods) that create projective images of 3D space. Current electrolocation methods are also quite different from electrical impedance tomography. We will describe current electrolocation technology, and progress on development of a propulsion system inspired by electric fish to provide the precise movement capabilities that this short-range sensing approach requires.

Dr. Malcolm MacIver is Associate Professor at Northwestern University with joint appointments in the Mechanical Engineering and Biomedical Engineering departments. He is interested in the neural and mechanical basis of animal behavior, evolution, and the implications of the close coupling of movement with gathering information for our understanding of intelligence and consciousness. He also develops immersive art installations that have been exhibited internationally.

6/1/12 Drew Bagnell Imitation Learning, Inverse Optimal Control and Purposeful Prediction

Programming robots is hard. While demonstrating a desired behavior may be easy, designing a system that behaves this way is often difficult, time consuming, and ultimately expensive. Machine learning promises to enable "programming by demonstration" for developing high-performance robotic systems. Unfortunately, many approaches that utilize the classical tools of supervised learning fail to meet the needs of imitation learning. I'll discuss the problems that result from ignoring the effect of actions influencing the world, and I'll highlight simple "reduction- based" approaches that, both in theory and in practice, mitigate these problems. I'll demonstrate the resulting approach on the development of reactive controllers for cluttered UAV flight and for video game systems. Additionally, robotic systems are often built atop sophisticated planning algorithms that efficiently reason far into the future; consequently, ignoring these planning algorithms in lieu of a supervised learning approach often leads to poor and myopic performance. While planners have demonstrated dramatic success in applications ranging from legged locomotion to outdoor unstructured navigation, such algorithms rely on fully specified cost functions that map sensor readings and environment models to a scalar cost. Such cost functions are usually manually designed and programmed. Recently, our group has developed a set of techniques that learn these functions from human demonstration by applying an Inverse Optimal Control (IOC) approach to find a cost function for which planned behavior mimics an expert's demonstration. These approaches shed new light on the intimate connections between probabilistic inference and optimal control. I'll consider case studies in activity forecasting of drivers and pedestrians as well as the imitation learning of robotic locomotion and rough-terrain navigation. These case-studies highlight key challenges in applying the algorithms in practical settings. J. Andrew Bagnell is an Associate Professor with the Robotics Institute, the National Robotics Engineering Center and the Machine Learning Department at Carnegie Mellon University. His research centers on the theory and practice of machine learning for decision making and robotics.

Dr. Bagnell directs the Learning, AI, and Robotics Laboratory (LAIRLab) within the Robotics Institute. Dr. Bagnell serves as the director of the Robotics Institute Summer Scholars program, a summer research experience in robotics for undergraduates throughout the world. Dr. Bagnell and his group's research has won awards in both the robotics and machine learning communities including at the International Conference on Machine Learning, Robotics Science and Systems, and the International Conference on Robotics and Automation. Dr. Bagnell's current projects focus on machine learning for dexterous manipulation, decision making under uncertainty, ground and aerial vehicle control, and robot perception. Prior to joining the faculty, Prof. Bagnell received his doctorate at Carnegie Mellon in 2004 with a National Science Foundation Graduate Fellowship and completed undergraduate studies with highest honors in electrical engineering at the University of Florida.

ing a model by hand can be a time consuming process. This motivates an alternative approach: *learning* a dynamical system directly from sensor data. A popular assumption is that observations are generated from a hidden sequence of latent variables, but learning such a model directly from sensor data can be tricky. To discover the right latent state representation and model parameters, we must solve difficult temporal and structural credit assignment problems, often leading to a search space with a host of (bad) local optima. In this talk, I will present a very different approach. I will discuss how to model a dynamical system's belief space as a set of *predictions* of observable quantities. These so-called Predictive State Representations (PSRs) are very expressive and subsume popular latent variable models including Kalman filters and input-output hidden Markov models. One of the primary advantages of PSRs over latent variable formulations of dynamical systems is that model parameters can be estimated directly from moments of observed data using a recently discovered class of spectral learning algorithms. Unlike the popular EM algorithm, spectral learning algorithms are statistically consistent, computationally efficient, and easy to implement using established matrix-algebra techniques. The result is a powerful framework for learning dynamical system models directly from data. 3/08/12 Andrea Thomaz Designing Learning Interactions for Robots

In this talk I present recent work from the Socially Intelligent Machines Lab at Georgia Tech. One of the focuses of our lab is on Socially Guided Machine Learning, building robot systems that can learn from everyday human teachers. We look at standard Machine Learning interactions and redesign interfaces and algorithms to support the collection of learning input from naive humans. This talk starts with an initial investigation comparing self and social learning which motivates our recent work on Active Learning for robots. Then, I will present results from a study of robot active learning, which motivates two challenges: getting interaction timing right, and asking good questions. To address the first challenge we are building computational models of reciprocal social interactions. And to address the second challenge we are developing algorithms for generating Active Learning queries in embodied learning tasks.

Dr. Andrea L. Thomaz is an Assistant Professor of Interactive Computing at the Georgia Institute of Technology. She directs the Socially Intelligent Machines lab, which is affiliated with the Robotics and Intelligent Machines (RIM) Center and with the Graphics Visualization and Usability (GVU) Center. She earned a B.S. in Electrical and Computer Engineering from the University of Texas at Austin in 1999, and Sc.M. and Ph.D. degrees from MIT in 2002 and 2006. Dr. Thomaz is published in the areas of Artificial Intelligence, Robotics, Human-Robot Interaction, and Human-Computer Interaction. She received an ONR Young Investigator Award in 2008, and an NSF CAREER award in 2010. Her work has been featured on the front page of the New York Times, and in 2009 she was named one of MIT Technology Review’s TR 35. 4/6/12 Javier Movellan Towards a New Science of Learning

Advances in machine learning, machine perception, neuroscience, and control theory are making possible the emergence of a new science of learning. This discipline could help us understand the role of learning in the development of human intelligence, and to create machines that can learn from experience and that can accelerate human learning and education. I will propose that key to this emerging science is the commitment to computational analysis, for which the framework of probability theory and stochastic optimal control is particularly well suited, and to the testing of theories using physical real time robotic implementations. I will describe our efforts to help understand learning and development from a computational point of view. This includes development of machine perception primitives for social interaction, development of social robots to enrich early childhood education, computational analysis of rich databases of early social behavior, and development of sophisticated humanoid robots to understand the emergence of sensory-motor intelligence in infants. 4/13/12 Emanuel Todorov Automatic Synthesis of Complex Behaviors with Optimal Control

In this talk I will show videos of complex motor behaviors synthesized automatically using new optimal control methods, and explain how these methods work. The behaviors include getting up from an arbitrary pose on the ground, walking, hopping, swimming, kicking, climbing, hand-stands, and cooperative actions. The synthesis methods fall in two categories. The first is online trajectory optimization or model-predictive control (MPC). The idea is to optimize the movement trajectory at every step of the estimation-control loop up to some time horizon (in our case about half a second), execute only the beginning portion of the trajectory, and repeat the optimization at the next time step (say 10 msec later). This approach has been used extensively in domains such as chemical process control where the dynamics are sufficiently slow and smooth to make online optimization possible. We have now developed a number of algorithmic improvements, allowing us to apply MPC to robotic systems. This requires a fast physics engine (for computing derivatives via finite differencing) which we have also developed. The second method is based on the realization that most movements performed on land are made for the purpose of establishing contact with the environment, and exerting contact forces. This suggests that contact events should not be treated as side-effects of multi-joint kinematics and dynamics, but rather as explicit decision variables. We have developed a method where the optimizer directly specifies the desired contact events, using continuous decision variables, and at the same time optimizes the movement trajectory in a way consistent with the specified contact events. This makes it possible to optimize movement trajectories with many contact events, without need for manual scripting, motion capture or fortuitous choice of "features". 4/20/12 Andrew Barto Autonomous Robot Acquisition of Transferable Skills

A central goal of artificial intelligence is the design of agents that can learn to achieve increasingly complex behavior over time. An important type of cumulative learning is the acquisition of procedural knowledge in the form of skills, allowing an agent to abstract away from low-level motor control and plan and learn at a higher level, and thus progressively improving its problem solving abilities and creating further opportunities for learning. I describe a robot system that learns to sequence innate controllers to solve a task, and then extracts components of that solution as transferable skills. The resulting skills improve the robot’s ability to learn to solve a second task. This system was developed by Dr. George Konidaris, who received the Ph.D. from the University of Massachusetts Amherst in 2010 and is currently a Postdoctoral Associate in the Learning and Intelligent Systems Group in the MIT Computer Science and Artificial Intelligence Laboratory. 4/27/12 Dieter Fox Grounding Natural Language in Robot Control Systems

Robots are becoming more and more capable at reasoning about people, objects, and activities in their environments. The ability to extract high-level semantic information from sensor data provides new opportunities for human-robot interaction. One such opportunity is to explore interacting with robots via natural language. In this talk I will present our preliminary work toward enabling robots to interpret, or ground, natural language commands in robot control systems. We build on techniques developed by the semantic natural language processing community on learning grammars that parse natural language input to logic-based semantic meaning. I will demonstrate early results in two application domains: First, learning to follow natural language directions through indoor environments; and, second, learning to ground (simple) object attributes via weakly supervised training. Joint work with Luke Zettlemoyer, Cynthia Matuszek, Nicholas Fitzgerald, and Liefeng Bo. Support provided by Intel ISTC-PC, NSF, ARL, and ONR. 5/4/12 Allison Okamura Robot-Assisted Needle Steering

Robot-assisted needle steering is a promising technique to improve the effectiveness of needle-based medical procedures by allowing redirection of a needle's path within tissue. Our robot employs a tip-based steering technique, in which the asymmetric tips of long, thin, flexible needles develop tip forces orthogonal to the needle shaft due to interaction with surrounding tissue. The robot steers a needle though two input degrees of freedom, insertion along and rotation about the needle shaft, in order to achieve six-degree-of-freedom positioning of the needle tip. A closed-loop system for asymmetric-tip needle steering was developed, including devices, models and simulations, path planners, controllers, and integration with medical imaging. I will present results from testing needle steering in artificial and biological tissues, and discuss ongoing work toward clinical applications. This project is a collaboration between researchers at Johns Hopkins University, UC Berkeley, and Stanford University.

Dr. Allison M. Okamura received the BS degree from the University of California at Berkeley in 1994, and the MS and PhD degrees from Stanford University in 1996 and 2000, respectively, all in mechanical engineering. She is currently Associate Professor in the mechanical engineering department at Stanford University. She was previously Professor and Vice Chair of mechanical engineering at Johns Hopkins University. She has been an associate editor of the IEEE Transactions on Haptics, an editor of the IEEE International Conference on Robotics and Automation Conference Editorial Board, and co-chair of the IEEE Haptics Symposium. Her awards include the 2009 IEEE Technical Committee on Haptics Early Career Award, the 2005 IEEE Robotics and Automation Society Early Academic Career Award, and the 2004 NSF CAREER Award. She is an IEEE Fellow. Her interests include haptics, teleoperation, virtual environments and simulators, medical robotics, neuromechanics and rehabilitation, prosthetics, and engineering education. For more information about our work, please see the Collaborative Haptics and Robotics in Medicine (CHARM) Laboratory website: http://charm.stanford.edu. 5/11/12 Blake Hannaford Click the Scalpel -- Better Patient Outcomes by Advancing Robotics in Surgery Surgery is a demanding unstructured physical manipulation task involving highly trained humans, advanced tools, networked information systems, and uncertainty. This talk will review engineering and scientific research at the University of Washington Biorobotics Lab, aimed at better care of patients, including remote patients in extreme environments. The Raven interoperable robot surgery research system is a telemanipulation system for exploration and training in surgical robotics. We are currently near completion of seven "Raven-II" systems which will be deployed at leading surgical robotics research centers to create an interoperable network of testbeds. Highly effective and safe surgical teleoperation systems of the future will provide high quality haptic feedback. Research in systems theory and human perception addressing that goal will also be introduced.

Dr. Blake Hannaford, Ph.D., is Professor of Electrical Engineering, Adjunct Professor of Bioengineering, Mechanical Engineering, and Surgery at the University of Washington. He received the B.S. degree in Engineering and Applied Science from Yale University in 1977, and the M.S. and Ph.D. degrees in Electrical Engineering from the University of California, Berkeley, in 1982 and 1985 respectively. Before graduate study, he held engineering positions in digital hardware and software design, office automation, and medical image processing. At Berkeley he pursued thesis research in multiple target tracking in medical images and the control of time-optimal voluntary human movement. From 1986 to 1989 he worked on the remote control of robot manipulators in the Man-Machine Systems Group in the Automated Systems Section of the NASA Jet Propulsion Laboratory, Caltech. He supervised that group from 1988 to 1989. Since September 1989, he has been at the University of Washington in Seattle, where he has been Professor of Electrical Engineering since 1997, and served as Associate Chair for Education from 1999 to 2001. He was awarded the National Science Foundation's Presidential Young Investigator Award and the Early Career Achievement Award from the IEEE Engineering in Medicine and Biology Society and is an IEEE Fellow. His currently active interests include haptic displays on the Internet, and surgical robotics. He has consulted on robotic surgical devices with the Food and Drug Administration Panel on surgical devices. 5/25/12 Malcolm MacIver Robotic Electrolocation Electrolocation is used by the weakly electric fish of South America and Africa to navigate and hunt in murky water where vision is ineffective. These fish generate an AC electric field that is perturbed by objects nearby that differ in impedance from the water. Electroreceptors covering the body of the fish report the amplitude and phase of the local field. The animal decodes electric field perturbations into information about its surroundings. Electrolocation is fundamentally divergent from optical vision (and other imaging methods) that create projective images of 3D space. Current electrolocation methods are also quite different from electrical impedance tomography. We will describe current electrolocation technology, and progress on development of a propulsion system inspired by electric fish to provide the precise movement capabilities that this short-range sensing approach requires.

Dr. Malcolm MacIver is Associate Professor at Northwestern University with joint appointments in the Mechanical Engineering and Biomedical Engineering departments. He is interested in the neural and mechanical basis of animal behavior, evolution, and the implications of the close coupling of movement with gathering information for our understanding of intelligence and consciousness. He also develops immersive art installations that have been exhibited internationally. 6/1/12 Drew Bagnell Imitation Learning, Inverse Optimal Control and Purposeful Prediction

Programming robots is hard. While demonstrating a desired behavior may be easy, designing a system that behaves this way is often difficult, time consuming, and ultimately expensive. Machine learning promises to enable "programming by demonstration" for developing high-performance robotic systems. Unfortunately, many approaches that utilize the classical tools of supervised learning fail to meet the needs of imitation learning. I'll discuss the problems that result from ignoring the effect of actions influencing the world, and I'll highlight simple "reduction- based" approaches that, both in theory and in practice, mitigate these problems. I'll demonstrate the resulting approach on the development of reactive controllers for cluttered UAV flight and for video game systems. Additionally, robotic systems are often built atop sophisticated planning algorithms that efficiently reason far into the future; consequently, ignoring these planning algorithms in lieu of a supervised learning approach often leads to poor and myopic performance. While planners have demonstrated dramatic success in applications ranging from legged locomotion to outdoor unstructured navigation, such algorithms rely on fully specified cost functions that map sensor readings and environment models to a scalar cost. Such cost functions are usually manually designed and programmed. Recently, our group has developed a set of techniques that learn these functions from human demonstration by applying an Inverse Optimal Control (IOC) approach to find a cost function for which planned behavior mimics an expert's demonstration. These approaches shed new light on the intimate connections between probabilistic inference and optimal control. I'll consider case studies in activity forecasting of drivers and pedestrians as well as the imitation learning of robotic locomotion and rough-terrain navigation. These case-studies highlight key challenges in applying the algorithms in practical settings. J. Andrew Bagnell is an Associate Professor with the Robotics Institute, the National Robotics Engineering Center and the Machine Learning Department at Carnegie Mellon University. His research centers on the theory and practice of machine learning for decision making and robotics.

Dr. Bagnell directs the Learning, AI, and Robotics Laboratory (LAIRLab) within the Robotics Institute. Dr. Bagnell serves as the director of the Robotics Institute Summer Scholars program, a summer research experience in robotics for undergraduates throughout the world. Dr. Bagnell and his group's research has won awards in both the robotics and machine learning communities including at the International Conference on Machine Learning, Robotics Science and Systems, and the International Conference on Robotics and Automation. Dr. Bagnell's current projects focus on machine learning for dexterous manipulation, decision making under uncertainty, ground and aerial vehicle control, and robot perception. Prior to joining the faculty, Prof. Bagnell received his doctorate at Carnegie Mellon in 2004 with a National Science Foundation Graduate Fellowship and completed undergraduate studies with highest honors in electrical engineering at the University of Florida.