GTR: Guided Thought Reinforcement Prevents Thought Collapse in RL-based VLM Agent Training

Open in new window