Data-Efficient Robot Reinforcement Learning at the RSE-lab

How long does it take for a robot to learn a task from scratch if no informative prior knowledge is given? Typically, very long. This project aims at developing and applying novel reinforcement learning methods to low-cost off-the-shelf robots to make them learn tasks in a few trials only. We use a standard robot arm by Lynxmotion and a Kinect-depth camera (total cost is 500 USD) and demonstrate that fully autonomous learning (with random intializations) requires only a few trials.

Keywords

• robot learning
• reinforcement learning
• Gaussian processes
• approximate inference
• planning with collision avoidance

Videos

Project Contributors

Marc Peter Deisenroth, Carl Edward Rasmussen, Dieter Fox

Project Publications

• Marc Peter Deisenroth, Carl Edward Rasmussen, Dieter Fox
  Learning to Control a Low-Cost Manipulator using Data-Efficient Reinforcement Learning
  Proceedings of the International Conference on Robotics: Science & Systems (R:SS), 2011


• Marc Peter Deisenroth and Dieter Fox
  Multiple-Target Reinforcement Learning with a Single Policy
  ICML-2011 Workshop on Planning and Acting with Uncertain Models