Reactive Robot Navigation: A Purposive Approach

The term navigation refers to the capability of a system to move autonomously in its environment, by using its own sensors. The more specific term visual navigation is used for the process of motion control based on the analysis of data gathered by visual sensors. The field of visual navigation is of particular importance mainly because of the rich perceptual input provided by vision. The problem of visual navigation has been traditionally treated without taking very much into account the environment of the robot, its body and the characteristics of the behavior that the robot is about to exhibit. Typically, monocular or stereoscopic visual systems are assumed and the effort is then focused in constructing a general representation of the environment that may thereafter serve the solution of any visionrelated problem. During the last decade, a new vision paradigm has attracted the interest of the computational vision research community. In this paradigm, called active and purposive vision [1], vision is more readily understood in the context of the visual behaviors in which the system is engaged. Consequently, vision tries to explore those aspects of the world that are important to the system at a given point in time, instead of aiming at a general representation of the environment which, besides being extremely difficult to extract, it is probably also not needed. In this paper, we describe a new method for visual robot navigation based on the principles of purposive vision. Thus, the aim of this research is not what vision can offer towards building a generalpurpose world representation, but how the visual system of a robot can be designed in order to assist the robot in exhibiting particular behaviors. We assume a robot that can translate in the forward direction and rotate (pan) around its vertical axis. We aim at developing a vision based reactive navigation capability that enables a robot to navigate in indoor environments (long corridors, narrow passages), avoiding collisions with walls and obstacles. The term reactive is used to express lack of a particular destination that could be set by using maps of the environment, landmark recognition etc. Free space is defined based on the motor capabilities of the robot: since the robot moves on a plane, all 3D structures that do not belong to this plane can be potentially harmful if the robot crashes on them and are therefore considered as obstacles. Since the robot is about to “live'' in indoor environments, it is expected to be able to handle situations where long corridors and narrow passages are encountered. It can be shown that difficulties arise when only central vision is used (i.e. a camera or a fixating stereo configuration at the direction of translation). Moreover, the use of cameras with wide field of view give rise to depth dependent geometric distortions that are difficult to correct. Instead, the proposed method employs a forward-looking camera for central vision and two sidelooking cameras for sensing the periphery of the visual field (see for example the configuration in Fig. 3). By using such a camera configuration, the robot is able to perceive walls and obstacles that are immediately close to it. Moreover, the target behavior may be implemented by indirectly comparing structure information acquired by the left and right cameras instead of computing precise structure information. This approach is motivated by experiments [2] that study the navigational behavior of honeybees who possess eyes that are pointing laterally (at about 180 degrees). In these experiments, bees were trained to navigate along corridors towards a source of food. The bees were observed to navigate in the middle of the corridor. The behavior is based [2] on velocity information computed at the left and right eyes of the bee. In simple terms, if the bee is in the center of the corridor, it perceives the world as “leaving'' its optical field with the same velocity in both eyes, while if the bee is closer to one of the sides of the corridor, it perceives it as moving faster.