A Model for Human Postural Regulation

摘要

We present a computational model of a quietly standing human. The model is composed of a system of 3 rigid links. There are two joints, one at the ankle and one at the hip. This is consistent with considerable experimental data which indicates that humans keep their knee angle nearly constant when there are small perturbations to their posture. Our model also includes the delays associated with sensing and actuation by the human neuromuscular system. An optimal control problem is designed with a cost function that contains a cost for state errors that is quartic or some higher even power and a cost for control that is quadratic. This performance measure tolerates larger state errors than the usual quadratic criterion and reduces the energy expended by the controller, a very plausible goal for a neuromuscular controller. We have solved the resulting constrained nonlinear optimal control problem. Simulations of the optimal control demonstrate better agreement with experimental studies of posture than other analytical models. Furthermore, we propose an implementation of the optimal control using only neurons and muscles. Straightforward learning/adaptation schemes would then make the proposed optimal controller a feasible solution to the real posture regulation problem.