A navigation system for autonomous mobile robots working in indoor and outdoor environments is presented. We have designed the system to be sufficiently robust and practical to perform real world tasks. For the purpose, we have developed it as an integrated system of a reactive controller and a deliberate controller.
The reactive controller consists of several local behaviors, each of which has its own goal and tries to achieve the own goal without concerning other behaviors and even the global goal. Global and complex behaviors of the system are emerged from cooperation of those simple local behaviors. By restricting the cooperation occurs only in a specified module called blender, we have made the reactive controller more extensible. At the same time, we have developed the controller not to suffer from the well-known problems like information loss while keeping extensibility by making the output of each behavior be fuzzy action command.
The reactive controller is only good at local navigation like obstacle avoidance. It is the deliberate controller``s responsibility to make the navigation system accomplish high-level tasks like office delivery. However, the deliberate controller does not actually control the robot but only provides guidance to the reactive controller. The deliberate controller consists of a task description, a planner and a state estimator. When human operator gives a task, the task is described in the deliberate controller as a state machine. Especially, in case of indoor navigation like office delivery, the task including an environmental model is described in an enhanced topological map, which has topological framework and some additional features such as augmented metric information. This dissertation presents a detailed description of the enhanced topological map and associated planner and state estimator. The planner determines an optimal sequence of actions while the state estimator finds out where the robot is.
The integration of t...