On Model Reconciliation: How to Reconcile When Robot Does not Know Human's Model?

The Model Reconciliation Problem (MRP) was introduced to address issues in explainable AI planning. A solution a MRP is an explanation for the differences between models of human and planning agent (robot). Most approaches solving MRPs assume that robot, who needs provide explanations, knows model. This assumption not always realistic several situations (e.g., might decide update her model robot unaware updates). In this paper, we propose dialog-based approach computing explanations under assumptions (i) does know model; (ii) share set predicates domain their exchanges are about action descriptions fluents' values; (iii) communication parties perfect; (iv) truthful. computed through dialog, defined as sequence rounds exchanges, human. each round, sends potential explanation, called proposal, replies with evaluation response. We develop algorithms proposals by responses implement these system combines imperative means answer programming using multi-shot feature clingo.