Chapter 6 BEHAVIOR LEARNING AND BEHAVIOR FUSION IN AUTONOMOUS NAVIGATION OF MOBILE ROBOTS

Behavior-based robotics has demonstrated its success in autonomous navigation of mobile robots. It has produced intelligent autonomous systems for a wide variety of areas, including factory automation, military applications and space exploration. Behavior-based approach decomposes a navigation system into task-specific behaviors, each of which has a direct connection between sensors and actuators. Behaviors operate in parallel and they may produce conflicting actions while attempting to control the same actuator simultaneously. In order to resolve the conflicts, an arbitration mechanism is required to integrate the independent behaviors seamlessly. In the previous research, this was accomplished by the high-priority-take-all command selection scheme or by the winner-take-all approach which votes for the actions of individual behaviors and selects the winner as the final control command. The former may cause failure in navigation as important navigational information may be lost when it completely switches the control to a single behavior. The disadvantage with the latter is that it uses a constant weight for each behavior. This is in contradiction with the fact that the desirability of each behavior may vary from situation to situation, e.g., the weight o