SOLE-R1: Video-Language Reasoning as the Sole Reward for On-Robot Reinforcement Learning Paper • 2603.28730 • Published Mar 30 • 1