tedmoskovitz / WNPG

implementation of Wasserstein Natural Policy Gradients and Wasserstein Natural Evolution Strategies

Wasserstein Natural Policy Gradients (WNPG) + Wasserstein Natural Evolution Strategies (WNES)

Implementations of WNPG and WNES from our paper Efficient Wasserstein Natural Gradients for Reinforcement Learning

To run WNPG:

python run_ppo_kwng.py

To run WNES, see WNES.ipynb.

Requirements:

• TensorFlow >= 1.14
• MuJoCo
• OpenAI Gym
• OpenAI Baselines
• PyBullet Gym

If you find this code useful, it would be great if you could cite us using:

@misc{moskovitz2020efficient,
      title={Efficient Wasserstein Natural Gradients for Reinforcement Learning}, 
      author={Ted Moskovitz and Michael Arbel and Ferenc Huszar and Arthur Gretton},
      year={2020},
      eprint={2010.05380},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}