Tag
UniCAR-RL introduces a reinforcement learning framework that decouples perception and reasoning to improve multimodal large language models' visual mathematical reasoning without annotation.