Handling Model Uncertainty and Multiplicity in Explanations via Model Reconciliation

• DOI: 10.1609/icaps.v28i1.13930
• Publication Date: 2022-09-27T20:01:12Z
• Authors: Sarath Sreedharan; Tathagata Chakra...; Subbarao Kambhampati
• Keywords: Supervisor; Mental model

Model reconciliation has been proposed as a way for an agent to explain its decisions human who may have different understanding of the same planning problem by explaining in terms these model differences.However, often human's mental (and hence difference) is not known precisely and such explanations cannot be readily computed.In this paper, we show how explanation generation process evolves presence uncertainty or incompleteness generating {\em conformant explanations} that are applicable set possible models.We also can contain superfluous informationand redundancies reduced using conditional iterate with attain common ground. Finally, will introduce anytime version approach empirically demonstrate trade-offs involved forms computational overhead communication human.We illustrate concepts three well-known domains well demonstration on robot typical search reconnaissance scenario external supervisor.