Reflex: Visually Grounded Reactive Humanoid Control
A humanoid that catches thrown boxes using only its own RGB-D camera: learn whole-body catching with RL first, then train vision to infer the box's motion while the controller stays fixed.
Taoyang Jia*, Weikai Huang*, Linxin Song*, Jared Darlington, Jieyu Zhao, Yue Wang, Jiafei Duan, Zhongzheng Ren, Ranjay Krishna