MIT's blind grasp reflex: robots grasp diverse objects by feel alone, no vision needed
arankomatsuzaki · x · 2026-09-28
- MIT Biomimetic Robotics Lab, with Seoul National University and Yonsei, published "See to Reach, Feel to Grasp": an architecture where an anthropomorphic hand grasps diverse objects using proprioception only — no images, object poses, or geometry.
- The design is modular: a separately controlled arm guides the palm near the object, while an RL hand-local policy performs the grasp, and a learned stable-grasp score decides when the object is secure so the arm can continue post-grasp manipulation.
- Author Arankomatsuzaki's framing: blind people handle most tasks robots fail at, so the bottleneck may be proprioception and touch, not vision — "see to reach, feel to grasp."
- Experiments in simulation and on hardware show robust blind grasping, and the learned grasp score closely tracks an independent physics-based grasp-stability measure; the skill composes with arbitrary arm controllers.
Related event: Robots may need proprioception, not better vision, to grasp(2 posts)→
More from Embodied
- ETH Zurich trains robotic hand to walk on its five fingers across 14 surfaces — burny_tech · 2026-09-28
- Builder brings a physical Codex Pet to San Francisco for OpenAI Dev Day — paw_lean · 2026-09-28
- Researchers debate GPT-6 Astra: a generalist VLA that could beat custom VLAs — YouJiacheng · 2026-09-28
- Eric Jang: now is the window to test OSS models on robotics before the gap closes — ericjang11 · 2026-09-28
- 10 NeurIPS 2026 robotics papers to watch: VLA inference and long-horizon tasks — shaohua0116 · 2026-09-28
- Claude Opus 5 Shows Creative Tool-Regrasp Skills in Robot Manipulation Tasks — ericjang11 · 2026-09-28