Horizon Robotics open-sources HoloMotion-1, a 400-million-parameter robot cerebellum large model capable of dancing, fitness, and box carrying
AI summary
Horizon Robotics' robot lab released HoloMotion-1, its open-source whole-body control model for humanoid robots. It is a 400-million-parameter "cerebellum" large model that, through MoE sparse activation and KV-cache inference mechanisms, achieves real-time inference of about 300 FPS on-device. The model is trained on motion data from multiple sources including internet videos, optical motion capture, and VR teleoperation, and has successfully demonstrated zero-shot transfer of complex motions such as dancing, crawling, fitness, and box carrying on real robots. The related code and technical report have been made public.
Key Takeaways
- Horizon Robotics' robot lab released HoloMotion-1,its open-source whole-body control model for…
- It is a 400-million-parameter "cerebellum" large model that,through MoE sparse activation and…
- The model is trained on motion data from multiple sources including internet videos,optical motion…
- The related code and technical report have been made public.
Quick Facts
- Published
- Updated
- Source
- IT之家·人工智能
- Category
- Model Releases
- AI score
- 74
地平线开源的这个4亿参数模型,把机器人“小脑”拉到一个小模型级别,300FPS实时跑在端侧,零样本跳舞搬箱子,搞具身智能的该跟进看看了。
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: Horizon Robotics open-sources HoloMotion-1, a 400-million-parameter robot cerebellum large model capable of dancing, fitness, and box carrying. May 19, 2026. https://new.nat6.net/en/news/cmpce9a1i01kislaea9ioa7tn/