Robot Navigation with Schematic Maps

We outline an approach to high-level interaction with autonomous robots by means of schematic maps. Schematic maps are knowledge representation structures to encode qualitative spatial information about physical environments. We present a scenario in which robots rely on high-level knowledge from perception and instruction to perform navigation tasks in physical environments. We discuss the general problem of formally representing a physical environment for acting in it. We propose a hybrid approach to knowledge and perception driven navigation. 1 A robot navigation scenario We describe a scenario consisting of an autonomous mobile robot and a structured spatial environment it lives in. The robot is equipped with rudimentary sensory abilities to recognize the presence as well as certain distinguishing features of obstacles that may obstruct the robotÕs way during navigation. The robotÕs task is to move to a given location in the environment. This task Ð that appears so easy to humans Ð is a rather difficult task for autonomous robots. First of all, the robot must determine where to go to reach the target location; thus it needs knowledge about space. Next, the robot must determine what actions to take in order to move where it is supposed to go; thus it needs knowledge about the relation between motor actions and movements and about the relation between movements and spatial locations. In theory, we could provide the robot with detailed information about the spatial structure of its environment including precise distance and orientation information as well as information about its own location in the environment. The robot then could compute a route through unobstructed space from its current location to the target location. Consequently, some route following procedure could traverse this route. In practice, however, this approach does not work. What are the problems? First, it is very hard to provide the robot with detailed knowledge about its spatial environment in such a way that this knowledge actually agrees with the encountered situation in the environment at a given time in all relevant aspects. Even if it agrees, it is impossible to get * An extended version of this paper is published in [Freksa et al. 2000].