Learning Wayfinding Heuristics Based on Local Information of Object Maps

In recent years, the problem of inferring and utilizing semantic information has gained considerable interest within the mobile robotics community. In this paper we focus on the problem of how to utilize the local semantic information of objects in a map to solve a navigation task more efficiently. In particular, we consider a wayfinding task and choose a supermarket environment as an example domain. We present an approach allowing a mobile robot to efficiently find the location of a target product in an unknown supermarket and to guide the search by heuristic rules based on information about the objects in the vicinity of the robot. The wayfinding strategy is learned from data by observing optimal search paths in a training set of supermarkets and then applied and evaluated in a previously unseen supermarket. We evaluate different search strategies and also give a comparison to the performance of humans in a real market. Our results demonstrate, that the learned wayfinding heuristics yield significantly shorter search paths than a standard search technique.