Goal-directed, dynamic animation of human walking

Abstract

This paper presents a hybrid approach to the animation of human locomotion which combines goal-directed and dynamic motion control. Knowledge about a locomotion cycle is incorporated into a hierarchical control process. The desired locomotion is conveniently specified at the top level as a task (e.g. walk at speed v), which is then decomposed by application of the concepts of step symmetry and state-phase-timings. As a result of this decomposition, the forces and torques that drive the dynamic model of the legs are determined by numerical approximation techniques. Rather than relying on a general dynamic model, the equations of motion of the legs are tailored to locomotion and analytically constrained to allow for only a specific range of movements. The dynamics of the legs produce a generic, natural locomotion pattern which is visually upgraded by some kinematic "cosmetics" derived from such principles as virtual leg and determinants of gait. A system has been implemented based on these principles and has shown that when a few parameters, such as velocity, step length and step frequency are specified, a wide variety of human walks can be generated in almost real-time.