10000小时人类数据，练出全球首个全身移动操作隐式世界动作模型

TL;DR — Based on the title alone (no body content was provided), this reports what it claims is the world's first "whole-body mobile manipulation implicit world-action model," trained on ~10,000 hours of human data — an embodied AI/robotics system for autonomous physical manipulation. It matters as a step toward general-purpose robots that plan and act in the real world.

• Claims a first-of-its-kind model for full-body mobile manipulation (locomotion + manipulation combined), positioning it as an embodied agent rather than a text-only model.
• Emphasizes a large-scale training corpus of ~10,000 hours of human demonstration data, suggesting an imitation/behavior-learning approach to acquiring physical skills.
• Describes an "implicit world action model," implying a learned internal world/dynamics representation coupled to action prediction for planning and control.
• Caveat: Only the title was available; no architecture details, benchmarks, or quantitative results were provided, so specifics above are inferred from the headline and cannot be verified.