A parameterized family of anatomically accurate human upper-body musculoskeletal models for dynamic simulation & control

Modeling human motion requires an accurate specification of musculoskeletal physiology, yet there exists no method to quantify the modeling accuracy required, or to predict the effect of modeling errors on subsequent analyses. Quantifying how inaccuracies in physiology, kinematics, or dynamics affect the study of human motion is a challenge that must be solved before we can construct robust generative models of human motor control at appropriate levels of detail. In this paper, we overcome two fundamental problems in characterizing the effect of model accuracy: the lack of ground truth about the arm's musculoskeletal kinematics, and the inability to systematically vary modeling accuracy. To do so, we developed a family of upper-body musculoskeletal models for a live human individual where modeled musculature was parameterized by decomposing volumetric muscles into fiber-groups of varying diameter and geometric complexity. The family of models thus obtained offer an unprecedented level of detail, and enable empirical comparisons of human motion analysis results across varying levels of anatomical accuracy and geometry. This sets the stage for large-scale studies of human motion that connect high level behavior to low level musculoskeletal dynamics, with applications in robotics, biomechanics, and human motor control.