RSS 2026 时间检验奖:形态即计算,具身智能的“逻辑刹车”已至
TL;DR - RSS 2026 honored the low-cost RBO pneumatic hand with its Test of Time Award, highlighting how compliant hardware can perform robust grasping without complex sensing or control. The work challenges embodied-AI approaches that rely primarily on larger models and more training data.
- The underactuated silicone hand adapts its grasp through physical compliance, effectively using morphology as computation.
- It deliberately exploits contact with tables and other environmental constraints rather than treating all collisions as errors.
- Identical open-loop commands can grasp objects of different shapes and fragility, reducing sensing and control requirements.
- Newer prototypes add inexpensive acoustic sensing only when passive compliance alone is insufficient.