Researchers led by Siyuan Wang have presented PROOF, a method that keeps a legged robot moving when its vision fails, in a paper on arXiv for the CoRL 2026 conference. Locomotion that relies on cameras can fail badly when the view is blocked, a surface reflects or the light changes, the authors write.
In a two-stage teacher-student setup, the teacher learns both movement guided by vision and a fallback that relies on the robot's sense of its own body. The student predicts features from the camera together with how uncertain they are, and suppresses them when the input looks unlike anything it was trained on.
On a Unitree Go2 with a RealSense D435 depth camera, the robot crossed large gaps, got onto raised platforms and walked along narrow beams in normal conditions, and kept moving stably under heavy occlusion and strong changes in lighting.
Background
Siyuan Wang
Robotics researcher
Researcher at the Southern University of Science and Technology and Pengcheng Lab.
More stories about Unitree Go2
Unitree Robotics
Robot maker
Unitree Robotics is a Hangzhou company that makes four-legged robot dogs and humanoid robots, including the G1, which research teams at Apple and other labs use.



