Topic 02 of 17

Vision-language-action in robotics

A vision-language-action model is a single policy that takes camera images and a natural-language instruction and outputs robot actions directly, usually by fine-tuning a pretrained vision-language model so it inherits open-world visual and linguistic understanding. VLAs replaced the older pipeline of separate perception, planning and control stages for many manipulation tasks, and they are the fastest-growing area in robotics research.

arXiv query: cat:cs.RO AND (abs:"vision-language-action" OR abs:"VLA")