Information-theoretic policy learning from partial observations with fully informed decision makers

In this work we formulate and treat an extension of the Imitation from Observations problem. is a generalisation well-known Learning problem where state-only demonstrations are considered. our treatment extend scope to feature-only which could arguably be described as partial observations. Therewith mean that full state decision makers unknown imitation must take place on basis limited set features. We out for methods extract executable policy directly those features which, in literature, would referred Behavioural Cloning methods. Our combines elements probability information theory draws connections with entropy regularized Markov Decision Processes.