Humanoid walking improves with adaptive sensor noise handling
CAP: Continuously Adaptive Perception-Blind Humanoid Locomotion via Learned Denoising
Robotics
Summary
Walking robots use special cameras to see obstacles ahead, but these cameras sometimes fail or give bad information. The authors created a method called CAP that can clean up noisy camera data and combine it with the robot’s own body sensors, so the robot can keep walking smoothly even when vision is spotty. They trained this system to handle various levels of camera noise, allowing the robot to adapt continuously rather than switching between on and off modes. Tests in simulation and on a real robot showed the approach helps the robot walk better when the camera data is partly broken or missing.
What this means in practice
- •For robotics engineers: Build walking robots that can adapt to unreliable sensor data for safer navigation over uneven or cluttered terrain.
- •For industrial automation teams: Deploy legged robots in warehouses or factories where sensor occlusions and noise commonly occur without losing locomotion reliability.
Authors
Hongjin Chen, Zijun Xu, Shihao Ma, Yi Zhao, Xilai Liu, Ke Ma, Wei Zhang, Chunyang Xie, Pengfei Li, Jieru Zhao, Wenchao Ding
Abstract
Humanoid locomotion across complex terrain demands forward-looking exteroception to anticipate obstacles, yet this signal is unreliable in real-world deployment, failing partially and intermittently. Existing perceptive policies often assume that depth observations remain clean and in-distribution, while recent attempts to unify perceptive and blind control typically route or switch between separate sub-policies, leaving recoverable information in partially corrupted depth unexploited. We instead propose CAP, a single-stage humanoid locomotion policy that recovers this signal with a perceptive world-model encoder trained as a learned denoiser to reconstruct clean depth from a corrupted input, together with a co-active proprioceptive variational encoder that supplies depth-free body-state information. A coupled training recipe pairs a depth-noise curriculum on the world-model input with world-model feature dropout on the policy-facing latent, exposing the policy to failures across the entire perception-quality spectrum. In simulation, CAP matches or improves upon perceptive baselines when depth remains informative, and degrades more smoothly than a binary-switching baseline as perception worsens. On the Unitree G1, controlled trials and indoor-outdoor deployments demonstrate perception-robust locomotion under intermittent occlusion, real-sensor corruption, and outdoor depth artifacts.