H-GRPO: Permutation-Invariant Reinforcement Learning for Grounded Visual Reasoning figure
AlphaXiv 中文概览(可滚动查看)