Evolving prototype control rules for a dynamic system

摘要

The genetic algorithm is used for the learning of prototype control rules for a dynamic system. Prototype control rules are point based, but only a limited number of points in the state space with associated control actions are learned. The nearest-neighbour algorithm is used to decide which of the rules to fire in any situation. The example of a simulated cart-pole balancing problem is used to demonstrate the advantages of this approach over other rule-learning methods.