Evaluation of genetic-fuzzy systems in the configuration space

We propose an approach to ground the design of learning systems on the analysis of the configuration space of the learning device (e.g., a robot) and on the interpretation of input data. In this paper, we focus on Learning Fuzzy Classifier Systems adopted to evolve behavioral controllers for autonomous robots. We show how it is possible to define some indexes to evaluate objectively both the learning process and the evolved system, thus supporting their designing with engineering principles