Introducing A-0: a complete robot learning system built on 16k hours of human demonstrations, with zero hours of robot data collected.
It spans data processing, pre-training, post-training and the inference harness, and it's embodiment-agnostic: the same weights run long-horizon, precise manipulation on four different robots. All the data is recorded with YUBI, our handheld gripper.
▶️ More demos and details:
The paper is out today, and we're releasing the code, weights and dataset: