A multi-joint model of quiet, upright stance accounts for the “uncontrolled manifold” structure of joint variance

The upright body in quiet stance is usually modeled as a single-link inverted pendulum. This agrees with most of the relevant sensory organs being at the far end of the pendulum, i.e., the eyes and the vestibular system in the head. Movement of the body in quiet stance has often been explained in terms of the “ankle strategy,” where most movement is generated by the ankle musculature, while more proximal muscle groups are only rarely activated for faster movements or in response to perturbations, for instance, by flexing at the hips in what has been called the “hip strategy.” Recent empirical evidence, however, shows that instead of being negligible in quiet stance, the movement in the knee and hip joints is even larger on average than the movement in the ankle joints (J Neurophysiol 97:3024-3035, 2007). Moreover, there is a strong pattern of covariation between movements in the ankle, knee and hip joints in a way that most of the observed movements leave the anterior–posterior position of the whole-body center of mass (CoM) invariant, i.e., only change the configuration of the different body parts around the CoM, instead of moving the body as a whole. It is unknown, however, where this covariation between joint angles during quiet stance originates from. In this paper, we aim to answer this question using a comprehensive model of the biomechanical, muscular and neural dynamics of a quietly standing human. We explore four different possible feedback laws for the control of this multi-link pendulum in upright stance that map sensory data to motor commands. We perform simulation studies to compare the generated inter-joint covariance patterns with experimental data. We find that control laws that actively coordinate muscle activation between the different joints generate correct variance patterns, while control laws that control each joint separately do not. Different specific forms of this coordination are compatible with the data.