Computing with Actions: the case of driving a car in a simulated car race

Describing and modeling a complex system is very hard and time consuming, so, we propose to use a linguistic approach introduced by Zadeh. By taking a linguistic approach we can use linguistic variables and approximate rules to characterize approximately phenomenons which are too complex or too ill-defined to be done in numerical terms. Starting from the results obtained in our previous works, in this paper we propose to decompose complex behaviors in simple behaviors, and context dependent rules for the behaviors coordination, which are described by set of fuzzy rules. The use of fuzzy rules for behavior coordination and behavior description has allowed us to have a shared framework that makes easy to develop and to compute with linguistic descriptions of actions. We propose to use a mixture of competitive and cooperative coevolutionary algorithms to manage the big search space and tackle a complex and decomposable problem. We use the Fuzzy-IEEE 2007 Car Racing Competition as an example of a problem that allows to show the benefits of our approach. In the example presented the results obtained are better and the flexibility is increased with respect to previous works. Keywords— Computing with Words, Computing with Actions, Fuzzy Linguistic Descriptions, Fuzzy Controllers, Co-evolutionary Algorithms, Behavior coordination.