Ant Group's Robbyant open-sources LingBot-Touch tactile AI model for robot hands

Ant Group's robotics unit Robbyant has released the model weights and inference code for LingBot-Touch. The model is designed to help robot hands use touch to handle objects.

Robbyant trained LingBot-Touch on nearly 5,000 hours of recordings. The data came from three types of tactile sensors, a setup intended to let the model work across different hardware configurations.

The company evaluated the model on 26 real-world tasks using two robotic hands. Those tests included handling fragile objects and coordinating both hands.

By open-sourcing both weights and inference code, Robbyant is making the tactile model available for outside use and testing. TechNode reported the release and the evaluation details.

The evaluation focused on physical manipulation tasks with two hands. Fragile-object handling and two-hand coordination were among the tested scenarios.