ACE-Ego breaks the authoritative SOTA leaderboard and unlocks a new paradigm for human-centric large-scale embodied model training
Summary
ACE Robotics and CUHK MMLab have released ACE-Ego, a new embodied VLA model designed for human-centric robot training. The model sets new results on RoboCasa GR1 TableTop and RoboTwin 2.0 by combining first-person human video with multi-robot data. The article highlights four mechanisms that align action space, robot form factors, timing, and training objectives across heterogeneous data sources. It also shows how the model targets real retail workflows such as packing, sorting, and box handling, which points to broader commercialization potential in robotics.
Classifications
industries
No industries detected
applications
Engineering & Scientific
AskAI Classifications
Labels
No AI classifications detected