A novel optimization approach is proposed for application to policy gradient methods and evolution strategies for reinforcement learning (RL). The procedure uses a computationally efficient Wasserstein natural gradient (WNG) descent that takes advantage of the geometry induced by a Wasserstein penalty to speed optimization. This method follows the recent theme in RL of including a divergence penalty in the objective to establish a trust region. Experiments on challenging tasks demonstrate improvements in both computational cost and performance over advanced baselines.
@article{arxiv.2010.05380,
title = {Efficient Wasserstein Natural Gradients for Reinforcement Learning},
author = {Ted Moskovitz and Michael Arbel and Ferenc Huszar and Arthur Gretton},
journal= {arXiv preprint arXiv:2010.05380},
year = {2021}
}