4,983 hours of real touch data lift robot policy success rates across 26 tasks
dr_cintas · x · 2026-11-14
A team recorded 4,983 hours of real touch data from tactile gloves on human hands and robot hands, then fed the model as an extra input to 4 existing robot policies across 26 real-robot tasks on two dexterous hands. Every policy improved:
- T-Rex: 45.42% → 68.61% average success
- LingBot-VLA: 48.28% → 73.09%
- π₀.₅ and Diffusion Policy also improved markedly
The tactile encoder stays frozen and is only added as extra input during post-training. Model weights and inference code are open-sourced in Tiny and Small sizes optimized for efficient tactile inference, with project page, GitHub, Hugging Face, and paper all public.
More from Embodied
- UC Davis Neuroprosthetics Lab Brings 8 Posters on Speech BCIs to SfN 2026 — neuro_engineer · 2026-11-14
- Dev data: Quest peak CCU ~10x the top 2026 SteamVR release, backing 90% market share claim — ID_R_McGregor · 2026-11-14
- Tactile model weights open-sourced in Tiny and Small sizes, links included — dr_cintas · 2026-11-14
- Robbyant Open-Sources LingBot-Touch, Giving Robots a Sense of Touch — dr_cintas · 2026-11-14
- Runway open-sources Praxis-1: 27B robot model trained on web video, tops RoboLab-120 at 46.2% — iamfakhrealam · 2026-11-14
- Neural Implant User Says AI Turns Silently Mouthed Songs Into Real Creations — Kyrannio · 2026-11-14