FiberTune: Preserving Action-Fiber Visual Residuals in Vision-Language-Action Fine-Tuning 事件

FiberTune: Preserving Action-Fiber Visual Residuals in Vision-Language-Action Fine-Tuning arXiv:2606.08653v1 Announce Type: new Abstract: Action-supervised fine-tuning of vision-language-action (VLA) policies fits demonstrations effectively but constrains only the directions that change predicted actions, leaving visual structure consistent across action-equivalent states free to collapse. We formalize this as residual visual collapse along local action fibers and propose FiberTune, a training-