Optimization Model of the Predictive Head Orientation for Humanoid Robots
Most studies on humanoid robot locomotion focus on the dynamic stability of the robot and movements of the limbs. The role of the head during locomotion is one aspect that has received relatively little focus. In humans, the head plays a very important role of stabilizing in space and anticipating walking direction. In order to generate realistic humanoid motion, this is a vital aspect that must be taken into account. In this study, we propose a generalized framework for modeling human like head-body behavior during locomotion. We developed a dynamic model of the head and body orientation in which the head orientation is actively controlled and the trunk is coupled to it by a rotational spring-damper like term. Least-squares optimization was used to determine the parameters of this model from recorded human experimental data. The results were evaluated against novel recordings and provided a good prediction of human behavior. Our approach is an effective way to create human-like motions for anthropomorphic agents such as humanoid robots, and provides a good mathematical basis for extension to other aspects of human movements.