arxivcs.ROcs.AIcs.CLcs.CV2026-06-30
RCT: A Robot-Collected Touch-Vision-Language Dataset for Tactile Generalization
Jingbo He, Michael Färber, Roberto Calandra
For robots manipulating open-world objects, tactile representations must generalize to unseen materials. We introduce RCT (Robotic Contact Tactile), a robot-collected touch-vision-language dataset with 29,279 tactile frames from full robot presses on 122 industrial reference mate…