Humanoid Gait Stabilization based on Omnidirectional Visual Gyroscope

Omnidirectional vision has already been used for attitude and heading estimation, thanks to its wide field of view. We adapt some existing methods in order to obtain a simple vision system capable to estimate pitch, roll and yaw angles for a real time application, using the parallel lines of the environment. We test this visual gyroscope and visual compass in some common environments to show that the proposed sensor continues to work even if the number of detected lines is small. We mount the catadioptric sensor in place of the head of a humanoid robot and use the measured angles to stabilize the walk, allowing the robot to keep a vertical position also if the terrain changes its slope. In this way we also stabilize the robot’s perception, reducing camera oscillations.