Intelligent Control of Autonomous Soccer Robots Compensating for Missing Information

Abstract

For autonomous mobile robots, visual information is used to recognize the environment. Although the acquisition of visual information is often disturbed in the real environment, it is necessary for a robot to act appropriately even if information is missing. We compensate for missing information for autonomous mobile robots by using short-term memory (STM) to make robots act appropriately. This method involves short-term memory and action selectors. Short-term memory is constructed based on the model of human memory and the forgetting curve used in cognitive science. These action selectors use compensated-for information and determine suitable action. One action selector consists of a neural network whose connection weights are learned by a genetic algorithm. Another selector is designed based on the designer's knowledge. These action selectors are switched based on reliability index of information. RoboCup Middle Size League soccer robots are used for demonstration. The experimental and simulation results show its effectiveness.