You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I work in an intersection of AI, robotics, and space exploration. I build systems for you ヾ(^▽^*)) so feel free to ask me if you have any question!
Currently I am a PhD candidate at the University of Hong Kong.
⚡ Recent works: Train a generalized policy / video generation model for robotics manipulation.
GR1-Training is the current SOTA policy on CALVIN benchmark for multi-task robotics manipulation (ABC-D split).
MimicTest integrates some of my policies, including Florence Policy developed on Microsoft's Florence2 VLM, which is trained with VQA, OCR, detection and segmentation tasks on 900M images.
Unlike OpenVLA and RT2, Florence2 is much smaller with 0.23B (Florence-2-base) or 0.7B (Florence-2-large) parameters.
Florence policy directly generates continuous actions with a linear action head / a diffusion transformer action head.
I am contributing to the HuggingFace LeRobot toolbox, including integration of the Florence policy.
I am participating the 1x challenge to train video generation model for robotics manipulation.