Inteligent control system taking account of system evaluation using weighted observational and explicit information

This study considers an intelligent control system to integrate flexibly its components by using weighted information where the system evaluation is reflected. Such system evaluates the information flowing through the components and converts them by weighting depending on the degree of importance. Integration of components based on the system evaluation enables a system consisting of them to realize various, flexible and adaptive control. In this study, the intelligent control method is applied to a swing up and stabilization control problem of a number of cart and pendulum systems on a restricted straight guide. To stabilize the pendulum in a restricted environment, each system should realize not only a swing-up and stabilization control of the pendulum, but also a position control of the cart to avoid collision or deadlock. The experiment using a real apparatus demonstrated that the controller learning light interaction acquires egoistic character, the controller learning heavy interaction behaves altruistically, and the controller equally considering self cart and another cart becomes cooperative. In other words, these autonomous decentralized controllers can acquire various characters and flexibility for cooperation.