Episode Details

A neural dynamics world model paired with model-free RL for quadruped and humanoid locomotion in IsaacLab, enabling long-horizon autoregressive prediction with policies trained in imagined rollouts that outperform model-based baselines on sim-to-real transfer.