Softmax Deep Double Deterministic Policy Gradients

This repository is the implementation of Softmax Deep Deterministic Policy Gradients in NeurIPS 2020, and is based on the open-source TD3 codebase.

Please cite our paper if you use this codebase:

@inproceedings{pan2020softmax,
  title={Softmax Deep Double Deterministic Policy Gradients},
  author={Pan, Ling and Cai, Qingpeng and Huang, Longbo},
  booktitle={Neural Information Processing System},
  year={2020}
}

Requirements

• python: 3.5.2
• mujoco_py: 1.50.1.68
• torch: 1.3.0
• gym: 0.15.3
• box2d-py

Usage

To replicate the results in the paper:

./run.sh

To run the SD3 algorithm in each single environment:

python main.py --env <environment_name>