Study of Bi-criterion Upper Body Posture Prediction Using Pareto Optimal Sets

This study involves further development of a direct approach to optimization-based posture prediction by using multi-objective optimization (MOO). Human performance measures representing joint displacement and delta potential energy are aggregated to predict more realistically, how virtual humans move. It is found that potential energy does not govern independently human posture. Rather, it must be coupled with another objective to avoid non-unique solutions and to improve realism. In any case, it is more suitable when reaching behind the avatar. Thus, we refine the idea of task-based posture prediction, concluding that performance measures should depend not only on the task being completed but also on where the task is completed relative to the human. Pareto optimal sets are depicted using the weighted sum and weighted min-max methods for MOO. By leveraging a special form of Pareto optimal set, insight is gained concerning how the functions should be combined. We find that the two MOO methods perform equally well, and the general form of the sets is independent of the target (to be touched with the finger) location.