https://github.com/cardwing/codes-for-rl-per

A novel DDPG method with prioritized experience replay (IEEE SMC 2017)
https://github.com/cardwing/codes-for-rl-per

openai-gym prioritized-experience-replay reinforcement-learning

Last synced: 7 months ago
JSON representation

A novel DDPG method with prioritized experience replay (IEEE SMC 2017)

• Host: GitHub
• URL: https://github.com/cardwing/codes-for-rl-per
• Owner: cardwing
• License: mit
• Created: 2018-10-02T14:28:09.000Z (about 7 years ago)
• Default Branch: master
• Last Pushed: 2018-11-13T08:57:23.000Z (almost 7 years ago)
• Last Synced: 2025-03-21T15:41:54.748Z (7 months ago)
• Topics: openai-gym, prioritized-experience-replay, reinforcement-learning
• Language: Python
• Homepage:
• Size: 17 MB
• Stars: 50
• Watchers: 0
• Forks: 16
• Open Issues: 3
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
Codes for conference paper ["A novel DDPG method with prioritized experience replay"](https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8122622)

### Demo videos

The following videos record the performance of our trained model running on five tasks in the OpenAI gym:

![demo1](https://github.com/cardwing/Codes-for-RL-PER/blob/master/demo_gif/inverted_pendulum.gif) 

![demo1](https://github.com/cardwing/Codes-for-RL-PER/blob/master/demo_gif/inverted_double_pendulum.gif)

![demo1](https://github.com/cardwing/Codes-for-RL-PER/blob/master/demo_gif/hopper.gif)

![demo1](https://github.com/cardwing/Codes-for-RL-PER/blob/master/demo_gif/halfcheetah.gif)

![demo1](https://github.com/cardwing/Codes-for-RL-PER/blob/master/demo_gif/walker.gif)

### Requirements

- Tensorflow 1.4.0

- MuJoCo 

- Gym 0.7.4

### Install necessary components

    conda create -n tensorflow_gpu pip python=2.7

    source activate tensorflow_gpu

    pip install --upgrade tensorflow-gpu==1.4

    pip install gym==0.7.4

    pip install mujoco-py==0.5.5

    

    

### Run the code

    source activate tensorflow_gpu

    cd PER-in-RL

    CUDA_VISIBLE_DEVICES=0 python run_ddpg_mujoco.py

### Notes

    export MUJOCO_PY_MJKEY_PATH=/path/to/mjpro131/bin/mjkey.txt

    export MUJOCO_PY_MJPRO_PATH=/path/to/mjpro131

You need to have the above mujoco key file in your path. Now, you can reproduce the results in our paper.

### Cite RL-PER

```

@inproceedings{hou2017novel,

  title={A novel DDPG method with prioritized experience replay},

  author={Hou, Yuenan and Liu, Lifeng and Wei, Qing and Xu, Xudong and Chen, Chunlin},

  booktitle={Systems, Man, and Cybernetics (SMC), 2017 IEEE International Conference on},

  pages={316--321},

  year={2017},

  organization={IEEE}

}

```

### Acknowledgement

This repo is built upon [Tensorflow-Reinforce](https://github.com/yukezhu/tensorflow-reinforce) and [prioritized-experience-replay](https://github.com/Damcy/prioritized-experience-replay)

### Contact

If you have any problems in reproducing the results, just raise an issue in this repo.