Flow Policy Gradients for Robot Control

TL;DR

This paper introduces flow matching policy gradients, a novel approach that enables training more expressive robot control policies beyond simple distributions, improving success and robustness in complex tasks and sim-to-real transfer.

Contribution

It presents an improved objective for flow matching policy gradients, demonstrating effectiveness in challenging robot control tasks and robust transfer from simulation to real robots.

Findings

Successful application to legged locomotion and manipulation tasks

Enhanced exploration capabilities during training from scratch

Improved fine-tuning robustness over baseline methods

Abstract

Likelihood-based policy gradient methods are the dominant approach for training robot control policies from rewards. These methods rely on differentiable action likelihoods, which constrain policy outputs to simple distributions like Gaussians. In this work, we show how flow matching policy gradients -- a recent framework that bypasses likelihood computation -- can be made effective for training and fine-tuning more expressive policies in challenging robot control settings. We introduce an improved objective that enables success in legged locomotion, humanoid motion tracking, and manipulation tasks, as well as robust sim-to-real transfer on two humanoid robots. We then present ablations and analysis on training dynamics. Results show how policies can exploit the flow representation for exploration when training from scratch, as well as improved fine-tuning robustness over baselines.