LingBot-Touch combines data from three tactile sensor types for robot manipulation

Researchers report that LingBot-Touch, a tactile foundation model, gives robot control models a shared representation of signals from piezoresistive, magnetic and visuotactile sensors. Their preprint describes experiments in simulation and on real robots across multiple hardware platforms, reporting improved performance on complex manipulation tasks. The model converts tactile readings into tokens that can feed diffusion policies and vision-language-action models, including π₀.₅ and LingBot-VLA. This lets those control systems use touch data across sensor types with different physical operating principles. The demonstrated scope is experimental manipulation across the tested hardware platforms.