Fuzzy Fitted Q

In recent years a number of different algorithms have been applied to learning robot control [1], [2], [3]. One family of methods, Fitted value iteration methods [4], are particularly promising because of their data efficiency, a valuable capability in real world problems [5]. This paper proposes a variation of Fitted value iteration that uses Q-learning with fuzzy logic based function approximation [6]. We demonstrate the viability of this algorithm in simulation and in real world experiments.