Learning and variable structure techniques in the control of a mechanical biped

The gait control of a mechanical biped is considered. Such a problem constitutes one of the most challenging tests for the control algorithms applied to robotic systems because, unlike the industrial manipulator, bipeds can fall. To avoid such a situation, provided suitable reference trajectories have been chosen a priori, a variable structure control algorithm using direct evaluation of the coordinates of the center of gravity is adopted. In order to smooth control actions and improve performance, an iterative learning procedure which progressively substitutes for the variable structure control algorithm is introduced