The Code for the DIME paper submission at ICML2025.
This repository accompanies the paper "DIME: Diffusion-Based Maximum Entropy Reinforcement Learning" published at ICML 2025.
The file setup.sh provides a convenient way to set up the conda environment and install the required packages automatically via
chmod +x setup.sh
./setup.shAfter installation is finished, the conda environment can be activated, and the code can be run using
python run_dime.pySpecific parameters can be set in the terminal such as the learning environment using hydra's multirun function
python run_dime.py --multirun env_name=dm_control/humanoid-runDetailed hyperparameter specifications are available in the config directory.
Portions of the project are adapted from other repositories:
- https://github.com/DenisBless/UnderdampedDiffusionBridges is licensed under MIT,
- https://github.com/adityab/CrossQ is licensed under MIT and is built upon code from "Stable Baselines Jax"