Skip to Main Content
Autonomous robots act according to the information they acquire from the environment. However, in real environments, the acquisition of the information regarding an object is often disturbed. This paper deals with the problem of developing an intelligent control for autonomous mobile robots to make them able to adapt to dynamic environment even when the uncertainty of information exists. RoboCup soccer robot is chosen as a demonstration target. In the method we propose, robots extrapolate missing environmental information with short-term memory. However, the acquired or extrapolated information is not always certain. In order to reduce the influences caused by misunderstandings and errors of information, it might be desirable that the robots use qualitative information for the situation where the reliability of information is low. Therefore, in our method, the each robot has an integrator which switches three kinds of action selectors, "quantitative", "qualitative" and "base" according to the reliability index of information. Each action selector operates on some action modules. The action selector selects the action module according to the environmental information. The integrator which is constructed by neural network selects the most suitable action selector based on "reliability indexes" of the ball position and the self-position estimated with short-term memory. The usefulness of this control system was shown through simulations.