Model Recognition as Planning

Given a partially observed plan execution, and a set of pos-sible planning models (models that share the same state vari-ables but different action schemata),model recognitionis thetask of identifying the model that explains the observation.The paper formalizes this task and introduces a novel methodthat estimates the probability of a STRIPSmodel to producean observation of a plan execution. This method builds on topof off-the-shelf classical planning algorithms and it is robustto missing actions and intermediate states in the observation.The effectiveness of the method is tested in three experiments,each encoding a set of different STRIPSmodels and all us-ing empty-action observations: (1) a classical string classifi-cation task; (2) identification of the model that encodes a fail-ure present in an observation; and (3) recognition of a robotnavigation policy