Robbyant open-sources LingBot-Touch: a tactile foundation model trained on 4,983 hours
kimmonismus · x · 2026-11-11
Robbyant released LingBot-Touch, an open (Apache 2.0) JEPA-based tactile foundation model positioned as a 'robot tactile cortex'. Key facts: pretrained on 4,983 hours / 1.33B frames of piezoresistive, magnetic and visuotactile recordings; a shared Haptic Transformer with modality-specific tokenizers unifies heterogeneous sensors; the tiny 5.46M-param encoder supports high-frequency inference. On the unseen XHand1 sensor, the frozen encoder beat sensor-specific adapters on all five probes, cutting contact-center error from 5.89mm to 0.16mm. Validated on 26 real + 12 sim tasks, with scaling gains as tactile data grows. kimmonismus notes touch starts near zero data unlike vision, and an open cross-sensor encoder gives every lab a shared starting point.
Related event: Ant's Robbyant Open-Sources LingBot-Touch Tactile Foundation Model(30 posts)→
More from Embodied
- Long-horizon mobile manipulation called the standout demo at CoRL — JasonMa2020 · 2026-11-11
- UCLA's TRIMI turns the data glove itself into the robot hand — DennisHongRobot · 2026-11-11
- Supersplat XR update: Quest 3 streamed scenes jump from 9.7 to 30+ fps, Vision Pro gets WebGPU — willeastcott · 2026-11-11
- Cartwheel releases three open motion-tracking policies for humanoid robots on Hugging Face — andrew_n_carr · 2026-11-11
- Claude Fable 5.2 designs a humanoid robot overnight with CAD, sim and reel — davidpattersonx · 2026-11-11
- Russ Tedrake's CoRL talk: deploy robots early, train on intervention data to scale — DominiqueCAPaul · 2026-11-11