Real time obstacle detection and navigation planning for a humanoid robot in an indoor environment

This paper describes a stereo vision guided humanoid that can perform real time obstacle detection, path analysis and navigation planning on a carpet floor. The real time obstacle detection algorithm uses the two dimensional disparity gradients approach to detect the obstacle on the floor without constructing the floor plane and affine adjustment even though the baseline of the camera system is not parallel with the floor. The size of the obstacle that can be detected is as small as 30 mm in diameter. The navigation planning is performed by using a distance transform method in which a shortest path is found and then the number of footsteps required to reach the destination is calculated.