Genetic Programming for Strategy Learning in Soccer Playing Agents : A KDD-Based Architecture

A KDD-based architecture should serve as a good framework to learn an improved strategy of ball control for intelligent soccer playing agents. Current work on using genetic algorithms to improve large scale data mining has been successful and provides an architecture for implementing future systems. This architecture is well suited for genetic programming and it is proposed that it can be extended to such applications. By learning “real-world” strategies, the performance of robot agents can be improved and become more similar to that of their biological counterparts. Layered learningis used to learn high-level behaviors that encompass previously learned low-level behaviors. Three phases of implementation are presented, with the goal of reducing the difference between soccer agent learning and human soccer learning.