Mobile robot mapping with geometrically inconsistent measurements in dynamic environments

Given a set of sensor data acquired by a mobile robot, we address a problem of evaluating degree of geometrical consistency between every datapair. Recently, such problems of consistency evaluation (or consistency matrix construction) have been increasingly important for mobile robot mapping systems to handle a geometrically inconsistent set of sensor data. In this paper, we focus on the application to 2D laser SLAM, and consider the consistency evaluation problem on a set of local maps acquired in dynamic environments. A computational difficulty arises from the need to perform a map-matching process for every mappair. To deal with the multiple map-matching processes, allocation of the limited computational resources among individual processes is an important planning problem. We show such a framework of multiple mappairs matching can be obtained as an extension of our previously published framework of single mappair matching, called incremental RANSAC. The methods have been applied to our SLAM system and experimentally evaluated in dynamic environments.