Mobile robots are rapidly increasing the application area for robotics in today``s world compared with a couple of decades ago. One of most difficult challenges in mobile robotics is real-world navigation. A real world can be changed suddenly and this change makes robot relinquish planning actions in advance. In order to overcome such a change behavior-based navigation was introduced, which decomposed the problem of autonomous control by task rather than by function. Each behavior, special-purpose task-achieving module, handles sensors and actuators directly. Since a behavior had minimal memory and a lack of relationship among them, it had a difficulty in planning deliberate actions and in communicating with humans.
To overcome such problems, we propose a new control strategy combining both the merits of behavior-based and planner-based approaches. The architecture has three major portions: Behaviors, Planner, and Coordinator. The Planner plays two important roles as a flexible human interface and the planner itself. It can interpret a sentence coming through multimodal human interfaces like voice and build topological map corresponding to the sentence. The Coordinator served as an interface between Behaviors and Planner and guided Behaviors to accomplish meaningful tasks according to guidelines from the Planner and the Position estimator which estimated robot``s posture in the topological map. In order to manage Behaviors the Coordinator made a wake-up list which contained those behaviors which should be activated. Since Behaviors are independent processes in the view of the operating system, they are scheduled by an operating system and only activated behaviors are able to produce their control outputs. These control outputs can try to control the same actuator simultaneously because they are made without consideration for the other behaviors. In order to resolve this conflict we designed a two-stage blender which considered how to surmount the drawbacks of ...