MetaFine proposes a diagnostic meta-evaluation framework for fine-grained robot manipulation

刺破成功率幻象，直面具身智能的「真灵巧」，机器人精细操作评测新范式来了！

Southeast University and Peking University researchers introduced MetaFine, a diagnostic meta-evaluation framework that tests fine-grained robot manipulation across understanding, perception, and behavior, and the article says traditional binary success metrics can overestimate fine-manipulation capability by up to 70%.

Why it matters: HKR-H comes from the success-rate illusion hook; HKR-K adds MetaFine’s three-axis diagnostic and a 70% overestimation claim; HKR-R fits robotics eval trust. Research scope keeps it at the low end of 78-84.