NAVIGATION OF MOBILE ROBOTS USING A FUZZY BEHAVIORIST APPROACH AND CUSTOM-DESIGNED FUZZY INTERFERENCING BOARDS

Two types of computer boards incorporating recently developed VLSI fuzzy inferencing chips have been developed to support the addition of qualitative reasoning capabilities to the real-time control of robotic systems. The design and operation of these boards are first reviewed and their use, in conjunction with our proposed Fuzzy Behaviorist approach, is discussed. This approach uses superposition of elemental sensor-based behaviors expressed in the Fuzzy Sets theoretic framework, to emulate ''human-like'' reasoning in robotic systems. In this paper, we discuss the implementation formalism which we developed to embody into fuzzy rule bases the decision-making modules of approximate sensor-equipped robots, and present an application to the development of qualitative reasoning schemes for autonomous robot navigation in a priori unknown environments. Several experiments involving mobile robots navigating in indoor and/or outdoor environments on the basis of very sparse and inaccurate sensor data are presented to illustrate the rule base development and augmentation process. The experiments progress from a small omnidirectional platform navigating autonomously in laboratory-type environments, to an actual car operating in parking lots under driver's aid control mode. These experiments illustrate that with the proposed Fuzzy Behaviorist approach and its associated implementation formalism, the inferencing rule base is easily developed and tested, while the progressive addition of behaviors allows the system to resolve situations of increasing complexity. Results from these experiments also indicate the efficiency of the proposed approach and of the resulting inferencing schemes, which can consist of as little as half a dozen behaviors, each consisting of about three qualitative rules, for typical navigation tasks.