Self-organizing CMAC neural networks and adaptive dynamic control

A self-organizing CMAC neural network mechanism and an CMAC based adaptive control scheme are presented. Two main efforts have been made in this study. One is on the self-organizing mechanism of CMAC neural network. The CMAC basis functions with a stair-waveform are introduced. A data clustering technique is used in reducing the memory size significantly and a structural adaptation technique is developed in order to accommodate new data sets. Another effort is on the unsupervised learning scheme, which is based on a Lyapunov index function. Adaptive dynamic control is implemented by means of the self-organizing CMAC neural network, and it can identify the unmodelled dynamics of a plant and ensures asymptotic system stability in a Lyapunov sense. The adaptive control system has been applied in the locomotion control of a bipedal walking robot successfully in simulation.