Dynamic walking pattern generation for a humanoid robot based on optimal gradient method

Abstract

Describes a method in which walking patterns based on the ZMP (zero moment point) are generated by a simple search algorithm; OGM (optimal gradient method). This approach has advantages in algorithmic simplicity and height generality. First, the designer gives prescribed time trajectories of both feet, hands and a reference of ZMP. Second, initial trajectories of a trunk are determined on the basis of a static walk. Finally OGM optimizes the horizontal motion of a trunk to reduce the deviation of the calculated ZMP from its reference. The same algorithms can be applied to the problems of yaw moment compensation around ZMP and shock compensation between soles and the ground by changing the objective function of OGM. The proposed method could generate various dynamic walking patterns such as steps forward, backward and aside or turning, and stable walks on the life-size humanoid 'H5' were realized by applying them to it.