Centered learning model in omni-directional controller of neural oscillator based biped locomotion

Human being has different walking behavior in each walking direction and walking speed. This paper proposed learning model of omni directional bio-inspired humanoid locomotion which is generating different walking behavior in different walking provision. Step length in sagittal and coronal direction, and degree of turning are considered parameters in walking provision. In proposed omni-directional walking model, interconnection structures which are composed by 16 neurons where 1 leg is represented by 4 joints and 1 joint is represented by 2 motor neurons are optimized by using NSGA-II in order to acquire walking behavior in certain walking provision. Several optimized interconnection data with different walking provisions are generated in optimization processes. Learning models are proposed for solving non-linearity of relationship between walking input and walking output representing the synaptic weight of interconnection structure, where one learning model representing one walking parameter. Furthermore, by using optimized model, walking behavior can be generated with unscaled walking provision. Smooth walking transition with low error of desired walking provision was proved based on several experiments in physical computer simulation.