ChinaTouch data lifted one AI model’s success on one robot hand from 45.6% to 72.5%.

Ant Group’s robotics unit Robbyant open-sourced the model, LingBot-Touch, trained on 4,983 hours of touch data.

It converts signals from different tactile sensor types into one shared encoding, so different robot hands can reuse the same learning.

Ant ran the tests itself: 26 tasks on 2 robot hands.

The model is on GitHub and Hugging Face for other labs to test.

How each outlet framed it
Zhidx leans favorable
reports Ant's LingBot-Touch as a cross-sensor fix, unifying magnetic, vision-tactile and piezoresistive signals into one token space, with 26-task real-robot results

Sources: Zhidx