Genetic Design of Fuzzy Controllers: The Cart and Jointed-Pole Problem

This paper considers the application of genetic algorithms to the automatic generation of fuzzy process controllers. In contrast to prior genetic-fuzzy systems which require that every input combination be enumerated, we employ a novel encoding scheme which maintains only those rules necessary to control the target system. The key to this method is to represent each fuzzy system as an unordered list of an arbitrary number of rules. This compact rule base methodology, introduced in [1] to generate a controller for the cart and single-pole problem, is applied to the more complex cart and jointed-pole problem. By using a compact rule base representation, successful systems evolve quickly, increasing the likelihood of success when applied to complex problems.†