GTR: Guided Thought Reinforcement Prevents Thought Collapse in RL-based VLM Agent Training Paper • 2503.08525 • Published Mar 11 • 16 • 2