SONIC: a robot whole-body control model at half GPT-1 scale
Original title:What can half of GPT-1 do? We trained a 42M transformer called SONIC to control the body of a humano...
AI summary
SONIC is a 42-million-parameter Transformer model (only half the scale of GPT-1), trained on more than 100 million motion-capture frames and more than 500,000 parallel robots in NVIDIA Isaac Lab, replacing handcrafted reward functions with dense frame-level supervision. After 3 days of training, it transfers zero-shot to a real G1 robot, achieving a 100% success rate on 50 motion sequences. A single policy supports VR teleoperation, video motion capture, text instructions, music response, and VLA model control. The project is fully open-source.
Key Takeaways
- SONIC is a 42-million-parameter Transformer model (only half the scale of GPT-1),trained on more…
- After 3 days of training,it transfers zero-shot to a real G1 robot,achieving a 100% success rate…
- A single policy supports VR teleoperation,video motion capture,text instructions,music…
- The project is fully open-source.
Quick Facts
- Published
- Updated
- Source
- X:Jim Fan (@DrJimFan)
- Category
- Other
42M小模型实现人形机器人全身控制,零样本迁移真实硬件且完全开源,开发者可复现
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: SONIC: a robot whole-body control model at half GPT-1 scale. February 25, 2026. https://new.nat6.net/en/news/cmo22oj0y0109slba6wj0c0mk/