Robot Planning, Execution, and Monitoring in an Uncertain Environment

Abstract

An intelligent robot, operating in an external environment that cannot be fully modeled in the robot's software, must be able to monitor the success of its execution of a previously generated plan. This paper outlines a unified formalism for describing and relating the various functions of a robot operating in such an environment. After exploring the distinction between the external world and the robot's internal model of it, and the distinction between actions that interact with the world and the robot's descriptions of those actions, we formalize the concepts of a plan and of its execution. Current developments at Stanford Research Institute, and the benchmark idea of an "ultimate" rational robot, are both analyzed in this framework.