SOLE-R1: Video-Language Reasoning as the Sole Reward for On-Robot Reinforcement Learning | ScienceToStartup | ScienceToStartup