Learning to Feel the Future: DreamTacVLA for Contact-Rich Manipulation
arXiv:2512.23864v3 Announce Type: replace-cross
Abstract: Vision-Language-Action (VLA) models have shown remarkable generalization by mapping web-scale knowledge to robotic control, yet they remain blind to physical contact. Consequently, they struggl…