New milestone: DreamZero based on a world-model backbone achieves zero-shot open-world robot control
Original title:New milestone: we trained a robot foundation model on a world model backbone, and enabled zero-shot,...
AI summary
The team releases DreamZero, the first World Action Model (WAM) based on a world-model backbone. Breaking with the traditional Vision-Language-Action paradigm, the model uses a pixel-level world model to achieve zero-shot open-world prompting and can perform new tasks it was not trained on. The study finds that WAM relies on diverse data rather than repeated demonstrations, and uses pixels as a universal bridge across embodiments to enable robot2robot and human2robot knowledge transfer. It can adapt to entirely new hardware with only 55 trajectories (about 30 minutes of teleoperation), validating the feasibility of world models as the next-generation foundation for Physical AI.
Key Takeaways
- The team releases DreamZero,the first World Action Model (WAM) based on a world-model backbone.
- Breaking with the traditional Vision-Language-Action paradigm,the model uses a pixel-level world…
- The study finds that WAM relies on diverse data rather than repeated demonstrations,and uses…
- It can adapt to entirely new hardware with only 55 trajectories (about 30 minutes of…
Quick Facts
- Published
- Updated
- Source
- X:Jim Fan (@DrJimFan)
- Category
- Other
世界模型成为物理AI新底座,机器人零样本泛化能力逼近GPT-2时刻
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: New milestone: DreamZero based on a world-model backbone achieves zero-shot open-world robot control. February 5, 2026. https://new.nat6.net/en/news/cmo22oj0y010cslba8d36s9r4/