Option compatible reward inverse reinforcement learning

Reinforcement learning in complex environments is a challenging problem. In particular, the success of reinforcement algorithms depends on well-designed reward function. Inverse (IRL) solves problem recovering functions from expert demonstrations. this paper, we solve hierarchical inverse within options framework, which allows us to utilize intrinsic motivation A gradient method for parametrized used deduce defining equation Q-feature space, leads feature space. Using second-order optimality condition option parameters, an optimal function selected. Experimental results both discrete and continuous domains confirm that our recovered rewards provide solution IRL using temporal abstraction, turn are effective accelerating transfer tasks. We also show robust noises contained