Ant Group’s robotics unit Robbyant releases open-source assets for a tactile AI model designed to help robot hands manipulate objects using touch. The release includes model weights and inference code for LingBot-Touch, with materials intended to let researchers and developers reproduce or build on the system.

Robbyant says the model was trained using nearly 5,000 hours of recordings from three types of tactile sensors, aiming to support use across different robot hand hardware. Across the outlets, emphasis differs: TechNode highlights the multi-sensor training and the breadth of inference support for varied tactile setups, while Pandaily focuses on how the model unifies sensor modalities and reports performance improvements on real-robot tasks.

Both accounts describe LingBot-Touch as a tactile sensing-and-action approach for robot hands, and both reference Robbyant’s evaluation on real robotic tasks, including gains across multiple action models.