A Hierarchical Approach to Sonar-Based Landmark Detection in Mobile Robots

Abstract. Landmark recognition in autonomous robots entails extracting symbolic information from sensory data, and is a crucial phase in bridging the gap between the symbolic and the sub-symbolic worlds. In this paper we propose a technique which identi es landmarks by matching a set of templates with a local occupancy grid built by sonar measures; another coarse-grained occupancy grid is used to decide when and where template matching is useful. Our approach is robust since the occupancy grid is a probabilistic model which represents the uncertainty on sensory data; hence, it may work well even on incomplete or noisy landmark shapes. It is fast since, during template matching, the second-level occupancy grid enables the search area to be restricted; thus, the other activities of the robot are not slowed down. The paper reports the results of experimental tests executed on a Pioneer mobile robot equipped with an array of sonars, a compass, an odometer and a camera.