Policy Agnostic RL: Offline RL and Online RL Fine-Tuning of Any Class and Backbone

TL;DR

This paper introduces policy-agnostic RL (PA-RL), a versatile offline and online RL method that effectively trains and fine-tunes diverse policy architectures, including diffusion and transformer models, with improved performance and efficiency.

Contribution

PA-RL replaces traditional policy improvement with a universal supervised learning loss, enabling training of various policy classes via action optimization, and demonstrates significant performance gains.

Findings

PA-RL doubles sample efficiency compared to existing methods.

Successfully fine-tuned a 7B generalist robot policy in real-world in 40 minutes.

Enables training and fine-tuning of diverse policy architectures with a unified approach.

Abstract

Recent advances in learning decision-making policies can largely be attributed to training expressive policy models, largely via imitation learning. While imitation learning discards non-expert data, reinforcement learning (RL) can still learn from suboptimal data. However, instantiating RL training of a new policy class often presents a different challenge: most deep RL machinery is co-developed with assumptions on the policy class and backbone, resulting in poor performance when the policy class changes. For instance, SAC utilizes a low-variance reparameterization policy gradient for Gaussian policies, but this is unstable for diffusion policies and intractable for autoregressive categorical policies. To address this issue, we develop an offline RL and online fine-tuning approach called policy-agnostic RL (PA-RL) that can effectively train multiple policy classes, with varying architectures and sizes. We build off the basic idea that a universal supervised learning loss can replace the policy improvement step in RL, as long as it is applied on "optimized" actions. To obtain these optimized actions, we first sample multiple actions from a base policy, and run global optimization (i.e., re-ranking multiple action samples using the Q-function) and local optimization (i.e., running gradient steps on an action sample) to maximize the critic on these candidates. PA-RL enables fine-tuning diffusion and transformer policies with either autoregressive tokens or continuous action outputs, at different sizes, entirely via actor-critic RL. Moreover, PA-RL improves the performance and sample-efficiency by up to 2 times compared to existing offline RL and online fine-tuning methods. We show the first result that successfully fine-tunes OpenVLA, a 7B generalist robot policy, autonomously with Cal-QL, an online RL fine-tuning algorithm, improving from 40% to 70% in the real world in 40 minutes.