What Can RL Bring to VLA Generalization? An Empirical Study

TL;DR

This study systematically evaluates how reinforcement learning, especially PPO, improves the generalization of large vision-language action models across various tasks and dimensions, outperforming supervised fine-tuning.

Contribution

It introduces a comprehensive benchmark for VLA generalization and demonstrates that RL fine-tuning, particularly with PPO, enhances semantic understanding and robustness over supervised methods.

Findings

RL fine-tuning with PPO improves semantic understanding.

PPO outperforms DPO and GRPO for VLAs.

A simple PPO training recipe enhances VLA generalization.

Abstract

Large Vision-Language Action (VLA) models have shown significant potential for embodied AI. However, their predominant training via supervised fine-tuning (SFT) limits generalization due to susceptibility to compounding errors under distribution shifts. Reinforcement learning (RL) offers a path to overcome these limitations by optimizing for task objectives via trial-and-error, yet a systematic understanding of its specific generalization benefits for VLAs compared to SFT is lacking. To address this, our study introduces a comprehensive benchmark for evaluating VLA generalization and systematically investigates the impact of RL fine-tuning across diverse visual, semantic, and execution dimensions. Our extensive experiments reveal that RL fine-tuning, particularly with PPO, significantly enhances generalization in semantic understanding and execution robustness over SFT, while maintaining comparable visual robustness. We identify PPO as a more effective RL algorithm for VLAs than LLM-derived methods like DPO and GRPO. We also develop a simple recipe for efficient PPO training on VLAs, and demonstrate its practical utility for improving VLA generalization. The project page is at https://rlvla.github.io