Generating Perception Requests and Expectations to Verify the Execution of Plans

This paper addresses the problem of verifying plan execution. An implemented computer program which is part of the execution monitoring process for an experimental robot system is described. The program analyzes a plan and automatically inserts appropriate perception requests into the plan and generates anticipated sensor values. Real-time confirmation of these expectations implies successful plan execution. The implemented plan verification strategy and knowledge representation are described. Several issues and extensions of the method are discussed, including a language for plan verification, heuristics for constraining plan verification, and methods for analyzing plans at multiple levels of abstraction to determine context-dependent verification strategies.