Skip to content

Code for paper "Model-based Adversarial Meta-Reinforcement Learning" (https://arxiv.org/abs/2006.08875)

Notifications You must be signed in to change notification settings

LinZichuan/AdMRL

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AdMRL

This is the implementation for the paper Model-based Adversarial Meta-Reinforcement Learning.

If you use this code in your research, please cite the following paper:

@article{lin2020model, 
    title={Model-based Adversarial Meta-Reinforcement Learning}, 
    author={Lin, Zichuan and Thomas, Garrett and Yang, Guangwen and Ma, Tengyu},
    journal={arXiv preprint arXiv:2006.08875}, 
    year={2020} 
}

Requirements

  1. OpenAI Baselines (0.1.6)
  2. MuJoCo (>= 1.5)
  3. TensorFlow (>= 1.9)
  4. NumPy (>= 1.14.5)
  5. Python 3.6

🔧 Installation

To install, you need to first install MuJoCo. Set LD_LIBRARY_PATH to point to the MuJoCo binaries (/$HOME/.mujoco/mujoco200/bin) and MUJOCO_LICENSE_PATH to point to the MuJoCo license (/$HOME/.mujoco/mjkey.txt). You can then setup mujoco by running rllab/scripts/setup_mujoco.sh. To install the remaining dependencies, you can create our environment with conda env create -f environment.yml. To use rllab, you also need to run cd rllab; pip install -e ..

🚀 Run

You can run experiments:

python main.py --taskname=Ant2D

You can also specify the hyper-parameters in launch.py and run many experiments:

python launch.py

License

MIT License.

Releases

No releases published

Packages

No packages published