Researchers have developed SONIC, a large-scale foundation model for humanoid robot control that significantly surpasses previous models in size and capability. By treating motion tracking as a scalable task and leveraging extensive motion-capture data, SONIC learns human motion priors without manual reward engineering. This model demonstrates utility in real-time kinematic planning for tasks like navigation and supports unified token spaces for VR teleoperation and vision-language-action models, enabling complex loco-manipulation. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Demonstrates that scaling models and data can lead to more generalist and robust humanoid robot control, potentially accelerating autonomous applications.
RANK_REASON The cluster contains a research paper detailing a new AI model and its capabilities. [lever_c_demoted from research: ic=1 ai=1.0]