An approach for robust mapping, detection, tracking and classification in dynamic environments

Understanding its environment remains a difficult problem for a mobile robot. Several intricate problems (localization, mapping, detection, tracking, classification) have indeed to be solved concurrently. However, most perception algorithms solve these issues independently leading to limited performances in highly changing environments. We present in this paper an original approach where the mapping, the tracking, the detection and the classification problems are addressed concurrently and where the perceptual knowledge of the robot is described using four recursively estimated discrete probability mass functions. Our first experiments based on simulated and real range data show that our approach is able to cope with complex outdoor situations.