DexTaG uses human tactile demos as RL reward guidance to teach robots human-like tool use

gan_chuang · x · 2026-09-30

• DexTaG (UMass Amherst + Genesis AI) is an RL pipeline that learns dexterous tool-use policies from mocap-glove demonstrations.
• Key idea: the glove's tactile readouts serve as reward guidance during RL training, steering contact acquisition and human-like grasp formation without relying on precise reference geometry.
• A single retargeting policy per task (marker pen, hammer — both requiring in-hand reorientation) generalizes to held-out references without per-trajectory retraining.
• The trained retargeter is distilled into a tactile-free student controller that deploys zero-shot in the real world, positioning mocap gloves as a scalable path for dexterous manipulation data.

Original post →

More from Embodied

• Simple-WAM: One forward pass replaces video denoising, keeping world action models' generalization — Tsinghua-LeapLab · 2026-09-30
• Open-source humanoid Roboparty shows up at IROS, alongside a Gundam you can touch — 4310sy · 2026-09-30
• Scoble says Google Glass is coming back: 'We were early, not wrong' — Scobleizer · 2026-09-30
• DexAgent learns bimanual dexterous manipulation from one human video, hitting 63.6% success — YunzhuLiYZ · 2026-09-30
• Wayve's supervised robotaxi ride in London via Uber impresses, closing in on Waymo — Sethwinterroth · 2026-09-30
• DynaRobotics trains robust whole-body controller for wheeled robot via Isaac Sim RL — zhengyiluo · 2026-09-30

Embodied channel →