Nvidia has developed a new foundation model called SONIC that enables humanoid robots to learn a wide range of movements, including walking, running, and object manipulation, from real-time human demonstrations. This open-source model utilizes over 100 million motion-capture frames to train a single controller, a significant departure from previous methods that required separate controllers for each skill. While still a research system, SONIC aims to provide whole-body control and adaptability for robots operating in unpredictable environments, with Nvidia planning to release checkpoints for broader developer use. AI
IMPACT Could accelerate the development and deployment of more versatile humanoid robots.
RANK_REASON Research system release from a major AI lab. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
