Ep#72: SONIC: Supersizing Motion Tracking for Natural Humanoid Whole-Body Control

SONIC (Supersizing Motion Tracking for Natural Humanoid Whole Body Control) introduces a universal humanoid controller designed to enable general-purpose robotic movement. Research scientist Zhengyi Luo explains that the system scales motion tracking by utilizing 700 hours of motion capture data, increased compute, and larger model sizes to achieve stable, high-quality imitation. By mapping human motion directly to humanoid joint angles, the controller supports diverse interfaces, including VR-based whole-body teleoperation, joystick-driven kinematic planning, and multimodal music-to-motion control. While the system demonstrates significant improvements in tracking precision and versatility, challenges remain in handling long-tail behaviors such as navigating stairs, climbing, and managing hardware variations in real-world environments. This approach bridges the gap between simulated humanoids and real-world deployment, providing a foundational framework for autonomous local manipulation and complex human-centric tasks.