Using sketch-maps for robot navigation: Interpretation and matching

We present a study on sketch-map interpretation and sketch to robot map matching, where maps have nonuniform scale, different shapes or can be incomplete. For humans, sketch-maps are an intuitive way to communicate navigation information, which makes it interesting to use sketch-maps for human robot interaction; e.g., in emergency scenarios. To interpret the sketch-map, we propose to use a Voronoi diagram that is obtained from the distance image on which a thinning parameter is used to remove spurious branches. The diagram is extracted as a graph and an efficient error-tolerant graph matching algorithm is used to find correspondences, while keeping time and memory complexity low. A comparison against common algorithms for graph extraction shows that our method leads to twice as many good matches. For simple maps, our method gives 95% good matches even for heavily distorted sketches, and for a more complex real-world map, up to 58%. This paper is a first step toward using unconstrained sketch-maps in robot navigation.