This repository is a curated and continuously updated list of research papers and resources focusing on the application of Diffusion Models within Reinforcement Learning (RL). It serves as a valuable reference for researchers and practitioners exploring this emerging intersection, aiming to track the frontier of Diffusion RL advancements.

How It Works

Diffusion Models in RL are primarily applied in two ways: as a method for trajectory optimization and planning, and as an expressive policy class for offline RL. The former casts trajectory optimization as a diffusion probabilistic model that iteratively refines trajectories, bypassing bootstrapping and avoiding short-sighted behaviors. The latter frames policies as conditional diffusion models, leveraging the scalability and multi-modal adaptability of diffusion models for policy optimization.

Quick Start & Requirements

This repository is a collection of papers and does not have a direct installation or execution command. The primary requirement is an interest in Diffusion Models and Reinforcement Learning. Links to official codebases and experiment environments are provided for individual papers.

Highlighted Details

• Broad Coverage: Encompasses papers from major conferences (ICLR, NeurIPS, ICML, CVPR, ICRA) and arXiv, spanning various applications like robotics, autonomous driving, and general control tasks.
• Key Applications: Highlights diffusion models for trajectory optimization, offline RL, imitation learning, world modeling, and multi-agent RL.
• Methodological Variety: Features diverse approaches including classifier-guided diffusion, classifier-free diffusion, and diffusion for policy gradient methods.
• Code Availability: Many listed papers include links to official codebases, facilitating practical implementation and experimentation.

Maintenance & Community

The repository is marked as "continually updated" and welcomes contributions. Specific contributors or maintainers are not highlighted beyond the project's open nature.

Licensing & Compatibility

Awesome Diffusion Model in RL is released under the Apache 2.0 license. This license is permissive and generally compatible with commercial use and closed-source linking.

Limitations & Caveats

As a curated list, the repository itself does not provide implementations or benchmarks. The practical utility of the listed papers depends on the quality and availability of their associated codebases and experimental setups.