Perception issues in mobile robotics

Abstract

This paper presents some recent advances in perception for mobile robotics The main objective is, by fusing multisensor information, to obtain sufficiently robust perception capabilities for the successful completion of typical mobile robotics tasks, including human-machine interaction. The use of these techniques is tailored at service robotics applications, such as cleaning of large industrial areas, inspection, security, transportation, costumer service, etc. The article focuses in two specific topics: a) simultaneous localization and mapping of large indoor industrial settings, and b) localization and tracking of human faces in image sequences. We tackle the simultaneous localization and mapping problem (SLAM) from an stochastic estimation perspective, and propose techniques to solve the effects of partial observability inherent with having two coupled inference problems. With respect to the localization and tracking of human faces (and other objects), we have developed an articulated head mounted vision system capable of tracking an individual under the most severe changing illumination conditions. With this platform, we have tested a mirage of estimation techniques for tracking color histograms.