BlinkMimic

Whole-body motion tracking for the Unitree G1 in about 3 minutes on one RTX 4090. BlinkMimic trains one policy for all 40 LAFAN1 clips (2.45 h of motion) to an evaluation normalized return of 0.82 in 2.9 min of training. It learns one differentiable world model and trains the policy by backpropagation through imagined rollouts, on mjlab / MuJoCo Warp.