BeingBeyond Launches Being-M0.7 Model for Humanoid Robots

BeingBeyond, founded by Peking University associate professor Lu Zongqing, has launched the Being-M0.7 model. This model enables humanoid robots to walk and manipulate objects simultaneously, pretrained on over 10,000 hours of human video and motion data. The model addresses messy human data through three stages: first, mapping video and motion into a shared space; second, adapting that to the robot's camera view; and third, adding an action expert to produce whole-body commands. It utilizes frozen image features and shares a compact motion representation between humans and robots.