Avoiding Failure via Pre-Planned Responses and Time-Bounded Planning

Abstract

Complex plans are required to safely operate an autonomous system in a dynamic environment. An ideal planner would build all reactions offline so it could deliberate as long as necessary to develop its response set. However, it may be impossible to build and store all plans offline, so online planning is required whenever previously unplanned-for situations arise. In a dynamic system, online planning must be terminated within time bounds imposed by the environment, with potentially adverse affects on plan quality. We propose that the primary goal of any system is to avoid failure (e.g., system destruction), and a secondary consideration is to achieve other goals. In this abstract, we propose a method to balance offline vs. online planning in terms of failure avoidance. We combine pre-planned responses with time-bounded online