Bibtek download is not available in the pre-proceeding
Tim G. J. Rudner, Cong Lu, Michael Osborne, Yarin Gal, Yee Teh
KL-regularized reinforcement learning from expert demonstrations has proved highly successful in improving the sample efficiency of deep reinforcement learning algorithms, allowing them to be applied to challenging physical real-world tasks. However, we show that KL-regularized reinforcement learning with behavioral policies derived from expert demonstrations suffers from previously unrecognized pathological behavior that can lead to slow, unstable, and suboptimal online training. We show empirically that the pathology occurs for commonly chosen behavioral policy classes and demonstrate its impact on sample efficiency and online policy performance. Finally, we show that we can resolve this pathology by specifying a non-parametric behavioral policy and that doing so allows KL-regularized RL to significantly outperform state-of-the-art approaches on a variety of challenging locomotion and dexterous hand manipulation tasks - without ad-hoc algorithmic design choices.