Loading sim environment…
The shared brain attempts tasks solo. Clip the wins and the fails — share them anywhere.
Success
10%
Trained on
20 demos
Eval rollouts
20
Per task
Sim only — no physical robot. ACT (lerobot 0.x) fine-tuned on mimbot pick_place — scripted seed demos; 300 train steps. Success measured in browser-wasm @mujoco/mujoco 3.9.0 over 20 held-out rollouts (MuJoCo 3.9.0). Not robot-ready; a pre-training supplement, not real-teleop data.
The brain trains once enough demos are collected. Be the first to contribute — your runs seed Brain v1.
Sim only — no physical hardware. Autoplay replays a fine-tuned open model (ACT) in the browser; training is weekly-batched, not real-time.