A modular, JAX-based framework and taxonomy for Reinforcement Learning with Diffusion and Flow policies.
March 31, 2026
Original Paper
FlowRL: A Taxonomy and Modular Framework for Reinforcement Learning with Diffusion Policies
arXiv · 2603.27450
The Takeaway
Diffusion policies are state-of-the-art for robotics but lack standardized, high-throughput implementations. This release democratizes research in generative RL by providing a unified codebase with JIT-compilation and benchmarks across multiple robotics suites.
From the abstract
Thanks to their remarkable flexibility, diffusion models and flow models have emerged as promising candidates for policy representation. However, efficient reinforcement learning (RL) upon these policies remains a challenge due to the lack of explicit log-probabilities for vanilla policy gradient estimators. While numerous attempts have been proposed to address this, the field lacks a unified perspective to reconcile these seemingly disparate methods, thus hampering ongoing development. In this