Robotics / news brief
Robbyant opens LingBot-Touch for cross-sensor robot touch
- Event
- Published

Robbyant released Apache-2.0 code and Tiny and Small checkpoints for LingBot-Touch, a JEPA-based tactile representation model trained on 4,983 hours of piezoresistive, magnetic and visuotactile recordings. The team reports tests on 26 real-robot tasks with 20 trials each, plus 12 simulated tasks. Results are author-reported, the paper is bundled in the repository rather than peer reviewed, and checkpoint access requires platform permission.
Why it matters
A shared touch representation across sensor types could reduce per-hand retraining, but the transfer and manipulation gains still need independent replication.
Sources
- LingBot-Touch: A Foundation Model for Heterogeneous Tactile Representation Learning (external site)Robbyant · primary source ·
Single source: Major same-day open research release with public Apache-2.0 code, model checkpoints, a technical report and a dated repository; no independent English editorial coverage was found before this round closed.