Study on using wavelet neural networks for the gait control of a biped robot

According to the deficiency in the structure as well as the capability of information storage of the neuron models used in traditional neuron networks, an intelligent neuron assemblage model with generalized wavelet basis function network as its stimulating function is proposed in this paper. The wavelet neural network not only converges much faster with better nonlinear approaching capability, but also is characterized by its ability in variable-scale adaptive structure adjustment and the generalized information storage, etc., which guarantee the network reflects the biological prototype much more faithfully. Static and pseudo dynamic learnings are demonstrated to prove the validity of the proposed mechanism.