‘Tactile Curiosity Drives Robot Interaction’ accepted to CoRL 2026

The paper’s author announced its acceptance to CoRL 2026 and said it studies how tactile feedback can help robots explore more efficiently. Another post describing the findings says only touch-curious agents learned to grasp without task rewards, although all variants observed touch. Their data could also be used for fully offline pick-and-place learning. When steering a frozen vision-language-action model that had never seen touch, adding touch helped, but touch curiosity produced the best results, especially on harder, contact-rich tasks.