Authors: Feihu Huang, Shangqian Gao, Pei Jian and Huang Heng

PyTorch Implementation of Momentum-Based Policy Gradient Methods (ICML 2020).

Code uploaded.

pytorch 1.1.0
garage 2019.10.1
mujuco
gym
If you do not install mujuco, then only CartPole environment is available.

To run IS-MBPG

python MBPG_test.py --env CartPole

To run IS-MBPG*

python MBPG_test.py --env CartPole --IS_MBPG_star True

To run HA-MBPG

python MBPG_HA_test.py --env CartPole

To run different environments change --env to one of the followings: "CartPole", "Walker", "Hopper" or "HalfCheetah". If you want to use our algorithms on different enviroment, you need to implement it by yourself, but it should be pretty straightforward.

@InProceedings{huang2020accelerated,
  author    = {Huang, Feihu and Gao, Shangqian and Pei, Jian and Huang, Heng},
  title     = {Momentum-Based Policy Gradient Methods},
  booktitle = {Proceedings of the 37th International Conference on Machine Learning},
  year      = {2020},}