Bingwen Wei

We conduct an empirical study to evaluate the generalization benefits of reinforcement learning fine-tuning versus supervised fine-tuning for vision-language-action models and provide some findings and analyses.