LingBot-Touch unifies three tactile sensor types using 4,983 hours of real recordings
CyberRobooo · x · 2026-11-10
LingBot-Touch is pretrained on 4,983 hours of real tactile recordings — 464K episodes and 1.33B frames total.
The key challenge: touch sensors don't speak one language. Camera-based fingertips, magnetic sensors, and pressure-sensitive ones each measure different physical quantities. The project maps all three onto a shared fingertip space and reads them as one set of tactile tokens via a shared Haptic Transformer.
It was tested on 26 real-world tasks with two dexterous hands (Sharpa Wave and XHand1) on dual-arm setups: twisting caps, inserting straws, closing zippers, and moving simulated eggs.
Related event: Ant Group's Robbyant Open-Sources Tactile Foundation Model LingBot-Touch(10 posts)→
More from Embodied
- CoRL 2026 to host twin workshops on adapting pretrained robot policies — huihan_liu · 2026-11-10
- Foldable dual-screen Apple Watch concept render goes viral on X — TechPreacher · 2026-11-10
- Unverified: Claude designed a working voice assistant from one prompt — FinanceYF5 · 2026-11-10
- Robbyant releases LingBot-Touch, a tactile model giving robot hands human-like touch sensing — CyberRobooo · 2026-11-10
- Dyna Robotics shows RL vs IK head-to-head on wheeled robot, unveils workflow agent Dyna-2.1 — chris_j_paxton · 2026-11-10
- AI researcher Chris Paxton comes away positive from a Tesla Cybercab ride — chris_j_paxton · 2026-11-10