Map Learning for Mobile Robotics

This project emphasizes the use of evidence grids - probabilistic representations of occupancy - in robotic localization and navigation. Our approach constructs a separate evidence grid for each distinct place using sonar and/or laser sensors, then retrieves and matches these grids later during place recognition. Experiments with both a Nomad 200 and simulated robots suggest this approach is robust with respect to short-term and long-term changes in the environment.

In more recent work, we have developed Magellan, an architecture for mobile robotics that embedded these place descriptions in a learned topological network, which it uses for path planning and execution monitoring. Magellan also incorporates a module for continuous localization, which it uses to correct position estimates as it moves through the environment. Experimental studies with Magellan in two robotics laboratories - at the Navy AI Center and at Stanford University - suggested that the same software works well in quite different settings.

This work was funded by Grant N00014-94-1-0505 from the Intelligent Systems Program, Office of Naval Research. The project involved a collaboration with researchers at the Navy AI Center and, though the ISLE portion has wound down, they continue to build on the ideas we developed jointly. Researchers who have contributed to this project include Pat Langley, Brian Yamauchi, Karl Pfleger, and Mehran Sahami at ISLE and Stanford, as well as Alan Schultz, Bill Adams, and John Grefenstette at the Navy AI Center.

The Nomad 200 robot used in our experiments on place learning and recognition.

Related Publications

Yamauchi, B., Langley, P., Schultz, A. C., Grefenstette, J., & Adams, W. (1998). Magellan: An integrated adaptive architecture for mobile robotics (Technical Report 98-2). Institute for the Study of Learning and Expertise, Palo Alto, CA.
Yamauchi, B., & Langley, P. (1997). Place recognition in dynamic environments. Journal of Robotic Systems, 14, 107-120.
Langley, P., Pfleger, K., & Sahami, M. (1997). Lazy acquisition of place knowledge. Artificial Intelligence Review, 11, 315-342.
Yamauchi, B., & Langley, P. (1996). Place learning in dynamic real-world environments. Proceedings of RoboLearn-96: International Workshop for Learning in Autonomous Robots (pp. 123-129). Key West, FL.
Langley, P., & Pfleger, K. (1995). Case-based acquisition of place knowledge. Proceedings of the Twelfth International Conference on Machine Learning (pp. 244-352). Lake Tahoe, CA: Morgan Kaufmann.

For more information, send electronic mail to langley@isle.org