Robot hands are getting better at seeing objects, but knowing how firmly to grip them remains a separate problem. Robbyant, the embodied AI company within China’s Ant Group, has released LingBot-Touch, a tactile foundation model intended to help different robotic hands interpret contact and pressure through a shared representation.
Announced on November 9, 2026, the model was pretrained on 4,983 hours of tactile recordings collected from three sensor families: magnetic, visuotactile and piezoresistive. Instead of requiring a robot control system to interpret each sensor’s raw output independently, LingBot-Touch maps the readings into a common fingertip coordinate system and produces features that can be used alongside vision and robot-state inputs.
Robbyant evaluated the system on 26 real-world manipulation tasks using Sharpa Wave and XHand robotic hands. Across four action-model configurations, the company reports average task-success improvements of approximately 14 to 26 percentage points over policies without tactile input. In one bottle-cap unscrewing test, success rose from 56.25% without touch feedback to 90% when LingBot-Touch was added. Feeding the same policy raw tactile signals achieved 62.5%.
The evaluations included the company’s LingBot-VLA as well as π0.5, Diffusion Policy and T-Rex. The results are company-reported and have not been independently reproduced.
Robbyant has made pretrained model weights and inference code available, including Tiny and Small variants. The smaller encoder is designed to add relatively little computational overhead when paired with an existing robot action model. The public release does not include the complete training dataset or training pipeline.
The release follows growing work on tactile feedback for dexterous manipulation, where visual information alone may not reveal whether a robot is slipping, pressing too hard or making the intended contact. LingBot-Touch’s technical report and inference implementation are available for researchers to examine and test on additional hardware.
