Humanoid robot localisation using stereo vision

This paper reports on the application of stereo vision localisation estimation techniques to a humanoid robot. Stereo vision has been used to localise wheeled robots with some success. Humanoid mounted stereo vision systems produce very jerky video sequences when the robot is walking. This can make the localisation task more problematic, especially when combined with the noisy nature of stereo depth, the 6DOF of camera motion, the difficulty to produce an accurate odometry estimate, and the desire to use dense stereo depth maps. Here, we apply established 2D localisation techniques to a humanoid robot using dense stereo vision depths maps and robot odometry.