Policy Optimization as Wasserstein Gradient Flows

All videos are downloadable at this Google Drive folder: https://drive.google.com/drive/folders/0B_KFuCNKS7ZVRlFUOFBUSVZLOUE?usp=sharing

For more details, please refer to http://proceedings.mlr.press/v80/zhang18a.html

Our implementations are heavily based on the Soft-Q and Soft actor-critic (SAC). We thank authors of Soft-Q/SAC to make the code public. If you have questions about reproducing our results, feel free to contact me: ryzhang@cs.duke.edu