Ant Group's Robbyant gets LingBot-Touch to read force it was never taught
heyshrutimishra · x · 2026-11-10
Robbyant, an embodied AI company under Ant Group, showed a striking emergent capability: its tactile model LingBot-Touch learned to read force it was never explicitly trained on.
- The model was pretrained on 4,983 hours of touch recordings
- Pretraining data included task-category info (opening, pouring, etc.) but no frame-level labels for force, contact location, or contact timing
- Training masked most of each touch clip and had the model predict the hidden portions, yielding emergent force perception
Paper link in the original post.
More from Embodied
- Ring enters robotics with Rover security robot and indoor Drone Cam — BenBajarin · 2026-11-11
- Stanford's BEHAVIOR Challenge: full-task success jumps 12% to 48% in one year, Huawei team wins — drfeifei · 2026-11-11
- CoRL 2026 slide trains an LSTM in 1 minute to predict robot joint torque — shaohua0116 · 2026-11-11
- Musk claims Cybercab will do majority of passenger miles; commenters cite unmet demand — RachelVT42 · 2026-11-11
- Newton: An open-source GPU-accelerated physics engine built on NVIDIA Warp — ChongZzZhang · 2026-11-11
- From quantum to AI: Ge Yang on CoRL's boom and where robot learning goes next — EpisodeYang · 2026-11-11