An optimal control approach to reconstruct human gait dynamics from kinematic data

A common approach to record full-body human movement data is by using marker based motion capture systems. To obtain dynamic gait data such as joint torques and ground reaction forces additional measurement devices have to be employed that pose restrictions on where feet have to be placed during the recording. In this paper we use articulated rigid multibody models and optimal control methods to recover dynamic gait data solely from kinematic data. Our approach is independent from the used marker set and creates the rigid multibody model and computes all controls for the model such that when applied to the model, it closely reproduces the originally recorded motion. To achieve this there are two steps involved: i) create a subject-specific rigid multibody model of the recorded person and used marker set and compute the joint kinematics using inverse kinematics. ii) reconstruct the gait dynamics by solving an optimal control problem. For step i) we created a parameterize human model HEIMAN and a graphical user interface PUPPETEER that facilitates creation of the subject specific model and the motion capture mapping. For ii) we use MUSCOD-II, which implements the direct multiple-shooting method. We apply our method on 15 emotional human walking motions to compare joint angle and torque patterns of different emotions.