ACE Robotics, CUHK Open-Source ACE-Ego
ACE ROBOTICS and CUHK have open-sourced ACE-Ego, a unified Vision-Language-Action (VLA) embodied AI model that enables robots to learn from egocentric human videos. The model utilizes camera-space actions and morphology conditioning to translate human movements into robot trajectories, achieving state-of-the-art benchmark performance.
While hardware gets most of the spotlight, the real bottleneck in general-purpose robotics is the scarcity of high-quality training data; ACE-Ego's approach of converting abundant human egocentric videos into robotic trajectories bypasses this constraint and could accelerate cross-embodiment deployment.
* By bridging the morphological gap between humans and robots, the model leverages massive existing human action datasets instead of relying solely on expensive, platform-specific robot demonstrations.
* A reliability-aware training objective ensures that the model filters out the inherent noise when mapping human camera-space movements to robot kinematics.
* Open-sourcing the pretraining framework and weights (ACE-Ego-0) will allow the research community to test and scale bimanual manipulation and multi-embodiment performance across diverse hardware configurations.
DISCOVERED
1h ago
2026-06-27
PUBLISHED
2h ago
2026-06-27
RELEVANCE
AUTHOR
Paull626