Self Reliant Robots ------------------- Reid Simmons Weh7220 9thNov94 Autonomous mobile robots have been demonstrated that successfully navigate, manipulate, and amble in unstructured environments -- as long as the demonstration doesn't last too long. To operate continually over long periods of time, robots must do more than just repeating the same tasks over and over: In long-term operation, the main problem is that infrequently occurring events do occur. To be successful, robot systems must be able to detect such events and react to them in a timely and intelligent manner. Currently, we have enough experience with Xavier to indicate where reliabilty gaps remain. I will talk about some of the problems, and some of the ways we are starting to address those problems. In particular, I am looking at methods for incrementally adding monitors and exception handlers (via TCA) to increase system reliability. This involves identifying when the robot is not making sufficient progress towards its goals, reasoning about why that is occurring, and developing methods for detecting and reacting to similar situations in the future. While this is currently done by hand, I am investigating ways of automating it. To facilitate this, we are moving away from discrete representations of the robot's position and actions to probabilistic models. I'll discuss some of the work in progress in probabilistic navigation, but that is a topic for a future meeting...