Amap releases the first full-body foundation model λ0 based on open-world behavior, advancing robots from "local learning" to "full-body motion alignment"

What happened: Alibaba's Amap released λ0, described as the first full-body foundation model based on open-world behavior, which lets robots learn whole-body coordination from human action data alone. Key numbers: Data collection efficiency about 3.4x mainstream methods; single demonstration collection time cut from about 100 Why it matters: The model reports zero-shot transfer to real robots across 7 whole-body coordination and long-range mobile manipulation tasks, and performance improves as human whole-body data scales. Code and data samples are open-sourced on GitHub and HuggingFace.

Alibaba's Amap released λ0, described as the first full-body foundation model based on open-world behavior, which lets robots learn whole-body coordination from human action data alone.

The model reports zero-shot transfer to real robots across 7 whole-body coordination and long-range mobile manipulation tasks, and performance improves as human whole-body data scales. Code and data samples are open-sourced on GitHub and HuggingFace.