Ant Group's Robbyant open-sources LingBot-Touch, a touch model that lifts four robot policies by 14 to 26 points
Robbyant's tactile encoder, trained on nearly 5,000 hours from three kinds of fingertip sensors, raised average success on 26 real two-arm tasks for π0.5, LingBot-VLA and two other policies. Feeding the same policies raw touch data helped half as much, and sometimes hurt.
2 min readSources verified