Amap, a subsidiary of Alibaba Group, has introduced λ0, a groundbreaking whole-body foundation model rooted in open-world behaviors. This innovative model empowers robots to master coordinated whole-body movements by leveraging solely human action data, thereby transcending the constraints of localized motion learning. In rigorous evaluations, λ0 has demonstrated state-of-the-art (SOTA) performance, achieving a motion capture efficiency that is roughly 3.4 times higher than that of mainstream methods, along with a real-machine task success rate that surpasses the strongest comparative baseline by 22.5%.

To tackle the challenge posed by existing embodied AI technologies, which often fall short in capturing and learning whole-body coordinated data for robots, λ0 pioneers a novel technical pathway. This pathway features agentless whole-body data capture and a software-hardware collaborative human-machine alignment strategy. The model constructs the HumanVerse-500 dataset, encompassing 500 hours of open-world, agentless whole-body data. Through multi-stage training and targeted alignment mechanisms, λ0 enables zero-shot transfer to real machines across seven whole-body coordination and long-range mobile manipulation tasks.

Furthermore, scaling up the volume of human whole-body data significantly enhances the performance of real robots in task execution. λ0 establishes a comprehensive pipeline for applying human whole-body experience to humanoid robots, laying a solid foundation for robots to undertake coordinated whole-body tasks in a variety of open scenarios. Currently, λ0 has made relevant hardware configuration codes open-source and released HumanVerse data samples, inviting developers to delve into and experience its capabilities.