Analysing the Difficulty of Learning Goal-Scoring Behaviour for Robot Soccer

by Jeff Riley, Vic Ciesielski

A brief insight on a method for the analysis of the fitness landscape described by the problem of learning goal-scoring behaviour for robot soccer – a problem that is considered to be very difficult for evolutionary algorithms. The method described uses landscape measures to examine features of the fitness landscape such as the degree of correlation between, or randomness of, points on the landscape, the modality of the landscape, and the ruggedness of the landscape.