Training a vision-language-action policy in your browser…
Charlot Lab · Vision-Language-Action
The VLA, seen
A real vision-language-action policy, drawn in full. It reads a camera frame, is told which block to reach, and drives an arm's joints, three streams, converging. Every node is real and its glow is the live activation. Tell it which block; watch it look, fuse, and move.
vision · the camera → convolutions
language · the instruction → an embedding
action · fuse → the arm's joint moves
drag scene · orbit
instruction:
what the policy sees