An Efficient Reactive Planner for Synthesizing Reactive Plans

Abstract

We present a nonlinear forward-search method suitable for planning the reactions of an agent operating in a highly unpredictable environment. We show that this method is more efficient than existing linear methods. We then introduce the notion of safety and liveness rules. This makes possible a sharper exploitation of the information retrieved when exploring the future of the agent. Introduction Classically, a plan is a set of actions to guide an agent from its current situation to another situation called the goal. If the result of these actions is not always the expected one, the agent is said to be operating in an unpredictable environment. Under this assumption, the agent may be deviated at any time from the intermediate situations expected in its plan. Whenever there is such a deviation, the agent has to replan from its new current situation. In real-time applications, the agent does not always have the time to replan. This prompted the development of new agent architectures whe...