Code to reproduce the experiments in Sample Efficient Reinforcement Learning via Model-Ensemble Exploration and Exploitation[abs].
It is noteworthy that our code is mainly based on MBPO, and we refer interested readers to the original code base MBPO for more details.
- Install
MuJoCo 2.0
at~/.mujoco/mujoco200
and copy your license key to~/.mujoco/mjkey.txt
, for example, you need to install the following dependencies first for Linux platform:
sudo yum install patchelf
sudo yum install mesa-libGL-devel mesa-libGLU-devel
sudo yum install mesa-libOSMesa-devel
sudo yum install mesa-libOSMesa
sudo yum install glfw
sudo yum install mesa-libGL
sudo yum install openmpi-devel
- Create a conda environment and install dependencies in
requirements.txt
cd code_meee
conda create -n "your_env_name" python=3.6
conda activate "your_env_name"
# install cuda to suport tf-gpu==1.13.1
conda install cudatoolkit==10.0.130
conda install cudnn==7.6.5
pip install -r requirements.txt
Configuration files can be found in examples/config
. Use the following command to conduct experiment on Humanoid-v2:
python main.py run_local examples.development --config=examples.config.humanoid.1 --trial-gpus=1
Currently only running locally is supported, so just keep the run_local
and examples.development
arguments. examples.config.humanoid.1
determines the configuration file you want to use, and --trial-gpus=1
indicate that you would like to experiment with one Nvidia GPU, you could change the experiment environment and GPU used by modifying relative arguments.
The results can be found in the default directory log_dir=~/ray_meee/
, you could also specify the directory in examples/config/configuration_files
.
If you use this code or results in your paper, please cite our work as:
@inproceedings{yao2021sample,
title={Sample efficient reinforcement learning via model-ensemble exploration and exploitation},
author={Yao, Yao and Xiao, Li and An, Zhicheng and Zhang, Wanpeng and Luo, Dijun},
booktitle={2021 IEEE International Conference on Robotics and Automation (ICRA)},
pages={4202--4208},
year={2021},
organization={IEEE}
}
The code in this repository is released under the MIT license as found in the LICENSE file.