No abstract

Using optimization to predict human posture provides a unique means of studying how and why people move. In a formulation where joint angles are determined in order to minimize a human performance measure subject to various constraints, the general question of when to model components as objective functions and when to model them as constraints has not been addressed thoroughly.We suggest that human performance measures, which act as objective functions, model what drives human posture, whereas constraints provide boundary conditions that restrict the scope of the model. This applied research study tests this hypothesis and concurrently evaluates how vision affects the prediction and assessment of upper-body posture.Singleobjective and multi-objective optimization formulations for posture prediction are used with a 35 degree-offreedom upper-body model of a virtual human called Santos TM . Vision is modeled as an objective function or as a constraint, and these two cases are tested in the context of standing reaching-tasks as well as reachingtasks within a cab environment. Results are evaluated qualitatively in terms of predicted postures, and quantitatively in terms of values for various performance measures. We find that the proposed hypothesis is accurate. We also find that vision alone does not govern human posture and that the selection of specific performance measures and constraints is task based. Some scenarios require one to see a target, while others necessitate only trying to see a target. The function of restricting the scope of the model is only relevant with difficult tasks, where constraints are likely to be active. Consequently, performance when using vision as a constraint and as an objective is similar for targets that are relatively easy to see.The proposed vision constraint provides the capability to model tasks that require vision. It also allows one to conduct what-if studies and evaluate the human-performance consequences of forcing a subject to see a target.