Explaining Temporal Plans with Incomplete Knowledge and Sensing Information

The challenge of explaining AI solutions is driven by the need for trust, transparency in the decision process, and interaction between humans and machines, which allows the first to comprehend the reasoning behind an AI algorithm decision. In recent years, Explainable AI Planning (XAIP) has emerged to provide the grounds for querying AI planner behaviour in multiple settings, such as problems requiring temporal and numeric reasoning. This paper introduces an analysis of explainability for temporal planning problems that require reasoning about incomplete knowledge and sensing information. We present an approach called Explainable AI Planning for Temporally-Contingent Problems (XAIP-TCP) that defines a set of interesting questions from the temporal and contingent planning perspective, covering numeric, temporal, and contingent notions in the presence of incomplete knowledge and sensing information. We present an analysis of the main elements required to deliver compelling explanations for a new set of domains motivated by real-world problems.