Amap, Alibaba Group's mapping and local services unit, and Alibaba's ATH business group have released Lambda-0, a whole-body humanoid foundation model trained on human first-person data rather than robot teleoperation recordings. The information was published on October 24, 2026, in a Pandaily report, covered by AI Affairs. The core idea is to change the source of learning: instead of teaching the robot only from demonstrations made on the robot itself, the team uses recordings of people doing everyday tasks.
According to the report, the team collected 500 hours of video, body movement and hand-action data from people wearing consumer-grade devices in everyday spaces, with no robot present during collection. Training took place in three stages: first hand interactions, then whole-body coordination, and finally humanoid control. To handle the difference between the human body and the robot body, the method uses viewpoint matching and an adjustment from broad movements to finer ones, including errors in where the robot's hands make contact with objects.
The figures released are the team's own and have not yet been independently reproduced, and should be read with that caveat. On seven whole-body and longer mobile-manipulation tasks, the team reports a 62.5% success rate when transferring Lambda-0 to a Unitree G1 without additional robot-specific training, 22.5 percentage points above the strongest baseline. In simulation, the team reports about 90% on the SIMPLE benchmark. According to the report, the code is available on GitHub under an Apache 2.0 licence, alongside a paper on arXiv.
The practical interest lies in tasks that combine locomotion and manipulation, such as crossing a room while carrying an object, which require walking and handling at the same time. If the approach holds up outside the lab, learning from human video could reduce the costly, time-consuming collection of demonstrations on physical robots. For now, however, what is confirmed is the reported release and the results presented by the team, not an independent validation nor immediate use in homes or factories.
Image credit: Unitree G1 at a trade show, photo by Sayanesy, CC0, via Wikimedia Commons, reproduced in the Humanoid Review index. The image shows the G1, the platform used in the reported tests, not the Lambda-0 model itself.
Sources: AI Affairs, com base em reportagem da Pandaily e Humanoid Review (índice de robôs humanoides, foto do Unitree G1).
Transparency: this article was produced with the help of artificial intelligence, based on cross-checking public posts on X and sources on the internet. Check the original sources to confirm the details before sharing.
Muse
:quality(80)/p7i.vogel.de/wcms/36/c0/36c0bf49d7c99b559e9977fde4376217/a264826-largev1.png)
