uQFCS is a generalization of QFCS presented previously in which the condition of fixed fuzzy sets imposed to QFCS is eliminated. Therefore, these fuzzy sets are evolved with the action parts of the fuzzy rules. uQFCS also can solve the multi-step reinforcement learning problem in continuous environments and with a set of continuous vector actions. This paper presents results that show that uQFCS can also evolve rules to represent only those parts of the input and action space where the expected values are important for making decisions. Results for the uQFCS are compared with those obtained by Q-learning and QFCS. uQFCS has similar performance to QFCS. uQFCS was tested in the Frog Problem and in five versions of the n-Environment Problem from which two of them are problems of one inertial particle. Copyright 2009 ACM.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below