A General Framework for Synthesizing and Executing Self-Explaining Plans for Human-AI Interaction

Anonymous

Abstract: In this work, we present a general formulation for decision making in human-in-the-loop planning problems where the human's expectations about an autonomous agent may differ from the agent's own model. We show how our formulation for such multi-model planning problems allows us to capture existing approaches to this problem and also be used to generate novel explanatory behaviors. Our formulation also reveals a deep connection between multi-model planning and epistemic planning and we show how we can leverage classical planning compilations designed for epistemic planning for solving multi-model planning problems. We empirically show how this new compilation provides a computational advantage over previous approaches that separate reasoning about model reconciliation and identifying the agent's plan.