OPIRL: Sample Efficient Off-Policy Inverse Reinforcement Learning via Distribution Matching

Official implementation for OPIRL: Sample Efficient Off-Policy Inverse Reinforcement Learning via Distribution Matching.
Accepted to IEEE International Conference on Robotics and Automation (ICRA) 2022.

Installation

Run the following command to install all Python dependencies:

$ pip install -e .
$ pip install -r requirements.txt

Other dependencies:

• Python 3.8+

• TensorFlow 2.4+

• CUDA=11.0

• cuDNN=8.0

• Experts/reward functions are provided on Google Drive

Run Experiments

First, unzip the expert/reward files from Google Drive.
Then, to simply run experiments on MuJoCo tasks, run the bash scripts in /scripts directory.
E.g.

$ sh ./scripts/run_halfcheetah.sh

Others

• OPOLO: code
• f-IRL: code