On February 12, 2026, AutoNavi, a subsidiary of Alibaba Group, proudly announced the launch of two groundbreaking foundation models: the Embodied Manipulation Foundation Model, ABot-M0, and the Embodied Navigation Foundation Model, ABot-N0.
ABot-M0, a versatile foundation model for embodied manipulation, has undergone a comprehensive transformation in three key areas: data integration, algorithmic innovation, and spatial awareness. By amalgamating over 6 million real-world manipulation trajectories, it has established a universal robot dataset. This enables cross-platform motion prediction through an advanced motion manifold learning algorithm. In the rigorous Libero-Plus benchmark test, ABot-M0 demonstrated an impressive 80.5% task success rate, outperforming industry solutions by nearly 30%.
ABot-N0 stands as the world's pioneering full-stack navigation foundation model, seamlessly integrating five navigation tasks: point navigation, target navigation, instruction following, point-of-interest navigation, and person following. Trained on an extensive dataset comprising approximately 8,000 high-fidelity 3D scenes and nearly 17 million expert examples, it leverages a hierarchical 'brain-action' architecture. This innovative approach has set new global benchmarks, achieving remarkable success rates in seven benchmark tests. Notably, ABot-N0 enhanced the SocNav closed-loop simulation success rate by 40.5% and the HM3D-OVON evaluation success rate by 8.8%, marking significant advancements in the field.