A neuro-fuzzy system architecture for behavior-based control of a mobile robot in unknown environments

A neuro-fuzzy system architecture for behavior-based control of a mobile robot in unknown environments is presented. A neural network is used to understand environments. Its inputs are a heading angle between the robot and a specified target, and range information acquired by an array of ultrasonic sensors. The output from the neural network is a trained reference motion direction for robot navigation. The methodology of the behavior-based control approach proposed in this paper is: (1) to analyze and to decompose a complex task based on stimulus-response behavior; and (2) to quantitatively formulate each type of behavior with a simple feature by fuzzy sets and fuzzy rules as well as to coordinate conflicts and competition among multiple types of behavior by fuzzy reasoning. An advantage is that building fuzzy sets and rules for each simple-featured type of behavior is much easier than for a complex task. Based upon a reference motion direction and distances between the robot and obstacles, different types of behavior are fused by fuzzy logic to control the velocities of the two rear wheels of the robot. Simulation experiments show that the proposed neuro-fuzzy system can improve navigation performance in complex and unknown environments. In addition, this architecture is suitable for multisensor fusion and integration. (C) 1997 Elsevier Science B.V.