A new method of distance estimation for robot localization in real environment based on manifold learning

Abstract

A new distance estimation method for robot autonomous localization from high-dimensional camera images is proposed based on 4 popular manifold learning algorithms. The camera images are supposed to embed in a high-dimensional manifold, and then the dimension is reduced to estimate the corresponding coordinate of the robot. Two experiments show that the distance is estimated regardless of the illumination, motion noise and environment geometric features. Experiment results with 3 image sets acquiring from the real environment verify the feasibility and effectiveness of the scheme and algorithms proposed in this paper.