On Exploiting Hitting Sets for Model Reconciliation

In human-aware planning, a planning agent may need to provide an explanation human user on why its plan is optimal. A popular approach do this called model reconciliation, where the tries reconcile differences in and human's such that also optimal model. paper, we present logic-based framework for reconciliation extends beyond realm of planning. More specifically, given knowledge base KB1 entailing formula phi second KB2 not it, seeks explanation, form cardinality-minimal subset KB1, whose integration into makes entailment possible. Our approach, based ideas originating context analysis inconsistencies, exploits existing hitting set duality between minimal correction sets (MCSes) unsatisfiable (MUSes) order identify appropriate explanation. However, differently from those works targeting inconsistent formulas, which assume single base, MCSes MUSes are computed over two distinct bases. We conclude our paper with empirical evaluation newly introduced instances, show how it outperforms state-of-the-art solver, generic non-planning instances recent SAT competitions, no other solver exists.