Alibaba’s Amap unveils humanoid model that learns whole-body tasks from human movements

Alibaba’s Amap and Alibaba Token Hub have introduced λ0, a foundation model designed to teach humanoid robots coordinated movement and manipulation using human demonstrations. The system learns from first-person video synchronized with body and hand movements, allowing robots to combine walking, posture adjustments and two-handed operations in longer tasks.

The researchers built HumanVerse, a 500-hour dataset of people performing activities in real-world environments. Their approach reduces reliance on collecting demonstrations through robot teleoperation. In the team’s evaluations, λ0 achieved a 62.5% success rate across real-world locomotion-and-manipulation tasks. The project has released data-collection hardware configuration code and sample data for researchers. [Alibaba λ0 project]