Convolution-sum-based generation of walking patterns for uneven terrains

Abstract

In generating walking patterns for humanoid robots, a Center-of-Mass trajectory is usually derived from the desired Zero-Moment-Point (ZMP) trajectory. One way to accomplish this is the use of the preview-control method, which tracks the desired ZMP trajectory while minimizing the jerk. Another method, which is more computationally efficient, is based on the convolution-sum method. Although this method is simple to implement, the resulting motion could be jerky. In this paper, we utilize the convolution-sum method to generate walking patterns for slopes and stairs walking while minimizing jerky motions. Furthermore, we extend the method to generate walking patterns for non-uniform terrain walking. This is accomplished by defining certain coordinate frames and maintaining the right-foot posture necessary for achieving robust walking. Computer simulations utilizing Webots were performed to validate the proposed convolution-sum method for the generation of walking patterns for a HOAP-2 humanoid robot.