Amap unveils λ0 full-body foundation model, reporting a 22.5% gain in real-world task success

Amap unveils λ0 full-body foundation model, reporting a 22.5% gain in real-world task success

N
News Editor
2026-10-23 17:43:08
Amap, an Alibaba Group company, has introduced λ0, which it described as the first full-body foundation model built on open-world behavior. The model is designed to let robots learn coordinated movement, posture control, and bimanual manipulation using only human action data. According to the announcement, λ0 delivered data collection efficiency about 3.4 times that of mainstream methods in the SIMPLE simulation benchmark and across multiple real-robot evaluations. In those real-world tests, task success rates improved by 22.5% versus the strongest baseline used for comparison. Amap said λ0 uses a technical route centered on embodiment-free full-body data collection and software-hardware coordinated human alignment. The company said this approach is aimed at challenges in open environments, including collection efficiency, full-body coordination, and transfer across different robot embodiments. The company also said collectors can gather demonstrations in everyday settings with consumer-grade devices, cutting the time needed for a single representative task demonstration from about 100 seconds to 29 seconds. Alongside the model release, Amap introduced the HumanVerse-500 dataset, which contains 500 hours of open-world embodiment-free full-body data, opened related hardware configuration code on GitHub, and published HumanVerse data samples on HuggingFace.

Amap, part of Alibaba Group, has released λ0, which it said is the first full-body foundation model based on open-world behavior. The model uses human action data alone to teach robots full-body coordination across locomotion, posture, and two-handed manipulation.

Benchmark and real-robot results

In the SIMPLE simulation benchmark and in multiple real-robot evaluations, λ0 achieved data collection efficiency about 3.4 times higher than mainstream methods, according to the announcement. Amap also said real-world task success rates were 22.5% higher than the strongest comparison baseline.

Data collection approach

The company said λ0 introduces a technical path built around embodiment-free full-body data collection and software-hardware coordinated human alignment. It framed that approach as a response to open-scene challenges tied to collection efficiency, full-body coordination, and transfer across different embodiments.

Amap said collectors only need to wear consumer-grade devices to gather data while moving through everyday environments. For representative tasks, the time required to capture a single demonstration was reduced from about 100 seconds to 29 seconds.

Dataset and code release

Amap also built HumanVerse-500, a dataset containing 500 hours of open-world embodiment-free full-body data. It has released related hardware configuration code on GitHub and published HumanVerse data samples on HuggingFace.

The item was published by Techub and attributed to QbitAI.

This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.
10400

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.

Amap unveils λ0 full-body foundation model, reporting a 22.5% gain in real-world task success | Bit.Fan