COMBINING OBJECT-ORIENTED VISION AND BEHAVIOR-BASED ROBOT CONTROL

Abstract

Behavior-based system architectures are now generally accepted as an effective basis for high-performance mobile robots. As one realization out of this class of architectures the concept of situation-oriented behavior-based navigation has been proposed by Wershofen and Graefe [9]. Its main characteristics are active perception of the robot’s dynamically changing environment, recognition and evaluation of its current situation, and dynamic selection of behaviors appropriate for the actual situation. A key question in all these architectures is how to choose an appropriate behavior from the repertoire of possible behaviors to change the state of the robot according to its goals. In contrast to other approaches we believe that this behavior selection and subsequent execution should at each moment depend on the dynamically changing situation the robot is finding itself in. This situation results in a complex way from a wide range of facts, assumptions and constraints and may be modified by the robot interacting with the environment (a more precise definition will be given in this paper). A prime prerequisite for recognizing complex situations in real time is a powerful perception system delivering rich and timely information about the environment. Object-oriented vision as proposed by Graefe [4] has proved to be capable of delivering such information in real-world applications, for instance, in