Mobile robot navigation by user-friendly goal specification

摘要

In this article, we discuss mobile robot navigation in an unknown environment. Different from the conventional formulation of the problem, we assume that the position of the goal is not given by its coordinates defined with the environment but by a verbal description of the route to it, such as "turn right at the 2nd corner, and...". We call distinctive places such as intersections, which are used to verbally describe a route, "landmarks" in a broad sense. Since it is difficult to give the robot landmark models that guarantee correct landmark recognition, we consider positional relations among landmarks given by the description, of the road to disambiguate the result of each landmark recognition. In addition, we try to make the robot learn the correct landmark models from the results of the previous navigation.