Nvidia's SONIC Revolutionizes Humanoid Robotics with Single Foundation Model
Nvidia has made a significant breakthrough in humanoid robotics with its SONIC (supersizing motion tracking for natural humanoid control) system, allowing robots to be taught to walk, run, crawl, dance, and manipulate objects using a single foundation model.
The open-source lightweight foundation model is designed to give humanoid robots whole-body control, enabling them to coordinate their joints, maintain balance, and adapt movements in real-time. SONIC was trained on over 100 million motion-capture frames, representing around 700 hours of human movement.
Nvidia's Yuke Zhu, director and distinguished research scientist, explained that whole-body control requires every joint to coordinate while maintaining balance, handling contacts, and adapting to changing motion goals in real-time. SONIC has been demonstrated across three dimensions: model size, training data, and compute.
The system has shown impressive results in tests, with the robot performing a range of tasks such as picking up and placing a drill in a box, dropping a soda can into the trash, and handling objects like a carrot, sponge, and apple. The team also demonstrated SONIC's ability to perform kung-fu and crawl on the floor, copying a human demonstrator in real-time.