Gaode releases λ0 humanoid-robot foundation model
Alibaba subsidiary Gaode released λ0, a whole-body foundation model designed to let robots learn coordinated movement, posture and bimanual tasks from human action data. The model research was jointly completed by Gaode and Alibaba's ATH business group. Gaode reports that λ0 achieved state-of-the-art results on the SIMPLE simulation benchmark and real-robot evaluations, with data-collection efficiency about 3.4 times that of the mainstream method and a 22.5% higher real-robot task success rate than the strongest comparison baseline. Its pipeline uses embodiment-free human-data collection, software-hardware alignment and cross-embodiment transfer. The release includes a technical report, open-source hardware configuration code and a HumanVerse-500 dataset sample; the underlying dataset contains 500 hours of open-world, embodiment-free human whole-body data.
λ0 provides an integrated pipeline for collecting human whole-body demonstrations, aligning them with robot observations and actions, and transferring the results to humanoid robots. The release also makes the technical report, hardware configuration code and a dataset sample available for research.
