What happened
Ant Lingbo Technology open-sourced LingBot-Touch, a tactile foundation model pretrained on nearly 5,000 hours of real tactile data spanning magnetic, visuotactile, and piezoresistive sensors. In 26 real-robot tests across Sharpa Wave and XHand1 hands, four action models improved average task success rates by about 14 to 26 percentage points with LingBot-Touch. The technical report, model, and code are available through Hugging Face, GitHub, and ModelScope.
Why it matters
LingBot-Touch’s real-robot tests and cross-hardware transfer example show how a shared tactile representation can support different hands and action models.