Assumptive Planning and Execution: a Simple, Working Robot Architecture

Abstract

In this paper, we present a simple approach to interleaving planning and execution based on the idea of making simplifying assumptions. Since assumptions can be tragically wrong, this assumptive approach must ensure both that the robot does not believe it has solved a problem when it has not and that it does not take steps that make a problem unsolvable. We present an assumptive algorithm that preserves goal-reachability and in addition we specify conditions under which the assumptive architecture is sound and complete.

We have successfully implemented the assumptive architecture on several real-world robots. Students in an introductory robot lab at Stanford University implement an assumptive system on robots that have incomplete information about their maze world. Dervish, our winning entry in the 1994 AAAI National Robot Competition, implements an assumptive architecture to cope with partially specified environments and unreliable effectors and sensors.