LEARNING CONVERGENCE IN THE CEREBELLAR MODEL ARTICULATION CONTROLLER

This paper presents a new way to look at the learning algorithm in the cerebellar model articulation controller (CMAC), proposed by Albus [1]. We obtain a proof that the CMAC learning always converges with arbitrary accuracy on any set of training data. We also propose an alternative way to implement CMAC based on the insights obtained in the process. We test the new implementation scheme with a computer simulation for learning the inverse dynamics of a two-link robot arm.