Learning non-Markovian Decision-Making from State-only Sequences

Conventional imitation learning assumes access to the actions of demonstrators, but these motor signals are often non-observable in naturalistic settings.