Online inverse reinforcement learning for systems with disturbances

This paper addresses the problem of online inverse reinforcement learning for nonlinear systems with modeling uncertainties and additive disturbances. In the developed approach, the learner measures state and input trajectories of the demonstrator and identifies its unknown reward function online. Sub-optimality introduced in the measured trajectories by the unknown external disturbance is compensated for using a novel model-based inverse reinforcement learning approach. The learner estimates the external disturbances and uses them to identify the dynamic model of the demonstrator. The learned model along with the observed sub-optimal trajectories are used for reward function estimation.