by , , , , ,
We introduce a sampling perspective to tackle the challenging task of training robust Reinforcement Learning (RL) agents. Leveraging the powerful Stochastic Gradient Langevin Dynamics, we present a novel, scalable two-player RL algorithm, which is a sampling variant of the two-player policy gradient method. Our algorithm consistently outperforms existing baselines, in terms of generalization across different training and testing conditions, on several MuJoCo environments. Our experiments also show that, even for objective functions that entirely ignore potential environmental shifts, our sampling approach remains highly robust in comparison to standard RL algorithms.
Robust Reinforcement Learning via Adversarial training with Langevin Dynamics K. Parameswaran, Y. T. Huang, Y. P. Hsieh, P. T. Y. Rolland, C. Shi, V. CevherIn Neural Information Processing Systems (NeurIPS), 2020
Bibtex Entry:
      title = {Robust Reinforcement Learning via Adversarial training  with Langevin Dynamics},
      author = {Parameswaran, Kamalaruban and Huang, Yu-Ting and Hsieh,  Ya-Ping and Rolland, Paul Thierry Yves and Shi, Cheng and  Cevher, Volkan},
      booktitle = {Neural Information Processing Systems (NeurIPS)},
      year = {2020},
      month = {December},