Centibots: The 100 Robots Project
RSE-lab

The Centibots system is a framework for very large teams of robots that are able to perceive, explore, plan and collaborate in unknown environments. The Centibots were developed in collaboration with SRI International, funded under DARPA's SDR program. The Centibots team currently consist of approximately 100 robots. These robots can be deployed in unexplored areas, and can efficiently distribute tasks among themselves; the system also makes use of a mixed initiative mode of interaction in which a user can influence missions as necessary.

Our research group focused on the problem of distributed mapping and exploration. In the mapping phase, multiple robots explore the environment in order to build a map that can be used in the subsequent search and surveillance phases. We developed a decentralized system that goes beyond the state of the art in multi-robot mapping in that it does not depend on reliable communication between robots and makes no assumptions about the robots' relative start locations. Below are some results obtained during the final demo (see also the SRI web site).

Project Contributors

Dieter Fox, Jonathan Ko, Benson Limketkai, Benjamin Stewart, Kurt Konolige (SRI International)

Mapping Example

Click on the image to see three of our robots as they build a map of the Allen Center (shown is the perspective of the blue robot).
The animation might not look correct for about 7 seconds.

Results from the Final Project Evaluation in Fort APHill, 01/2004

• More information on our research on mapping and exploration can be found on the RSE-lab projects web site.
• The evaluation of the Centibots system is described in the Proc. of the IEEE 2006 paper Distributed Multi-Robot Exploration and Mapping.
  
• Our coordination strategy is discussed in an IROS-03 paper. We also developed a hierarchical Bayesian technique for the map merging problem, see our ISRR-03 and a UAI-03 papers.
  
  

One of the maps built during the evaluation overlayed with manually generated ground truth map (click on image for large version).


Three maps built during different runs, under different circumstances. The fact that the maps are virtually identical confirms our belief that these maps are more accurate than the manually generated ground truth map.

Some Pictures

Amigobots on the loose	Command center during final demo
UW development team	Developers united
Robots taking a break	Our fearless leader