Experimental verification of a walking robot self - localization system with the kinect sensor

In this paper we investigate methods for self-localisation of a walking robot with the Kinect 3D active range sensor. The Iterative Closest Point (ICP) algorithm is considered as the basis for the computation of the robot rotation and translation between two viewpoints. As an alternative, a feature-based method for matching of 3D range data is considered, using the Normal Aligned Radial Feature (NARF) descriptors. Then, it is shown that NARFs can be used to compute a good initial estimate for the ICP algorithm, resulting in convergent estimation of the sensor egomotion. Experimental results are provided.