← October 24, 2026 briefing · added in the 20:00 KST update
Amap and Alibaba open-source 'Lambda-0,' a whole-body humanoid model trained only on human egocentric video, applied zero-shot to Unitree G1 (report)
Verified (fact)
According to Pandaily, Amap and Alibaba's ATH group open-sourced 'Lambda-0,' a whole-body humanoid foundation model. Its training data consists solely of 500 hours of first-person video recorded by humans, with no robot data. The model was reportedly transferred to the Unitree G1 robot without additional training.
Needs fact-check (claim)
The developers said it transfers zero-shot to the Unitree G1 (according to the report). Detailed figures such as success rates are not in the excerpt.
Why it matters (interpretation)
The biggest bottleneck in robot learning is expensive robot demonstration data. If whole-body control works from human first-person video alone, data collection costs could drop sharply.
Confidence medium · official source pending
Sources
More about Amap
-
Amap quadruped robot 'Tutu' begins security patrols at Alibaba's Xixi campus, auto-filing inspection photos to work tickets (report)
Amap's quadruped robot Tutu is handling security inspections at Alibaba's Xixi campus in Hangzhou. It plans routes with the ABot stack,...