Model-Based Optimization for the Analysis of Human Movement and the Design of Rehabilitation Devices

Human motions result from a complex and well-coordinated interaction between the body segments. Walking and the sit-to-stand transfer are amongst the most challenging human motion in terms of coordination and internal loads, respectively. We propose model-based nonlinear optimal control methods to reconstruct and synthesize these motions while considering the dynamics of the motion over the whole time horizon. The redundant and highly nonlinear character of the computed motions encourages to discretize the optimization problem according to direct multiple-shooting methods. The goal is to identify principles which enable us to describe the patterns of these motions. We approach human walking from the perspective of unimpaired subjects and subjects walking with unilateral transfemoral prostheses. Their walking motion is reconstructed from motion capture data using subject-specific threedimensional multibody models. The motion of the models is fitted to the recorded data for a whole stride in a least-squares sense in multi-stage optimal control problems. Analyzing the reconstructed motion for the individual foot placement of the subjects suggests that it relates with the Capturability concept: foot locations are chosen by the subjects which enable a balance between the inherently conflicting goals of effortless progression and quick response to perturbations. In addition, the modulation of the ground collision impact forces at heel strike is found to play a major role in the step-by-step stability strategy. Based on these findings, we propose Capturability as a complementary criterion to the established clinical stability assessment methods. The sit-to-stand motion is particularly demanding for humans with mobility impairments, due to the high joint loads required to lift the body into the standing pose. We synthesize optimal sit-to-stand by solving two-stage optimal control problems. We presume that the sit-to-stand motion is substantially characterized by a preparation phase prior to the actual lift-off. Full body models are established with dynamic model parameters which specifically represent elderly humans from different levels of mobility. For impaired subjects, mobility support is assumed to be provided by generic support actions. The optimization computations result in different patterns which include significant arm motion in both phases. Therefore, the results support our approach to choose a full body representation of the human as well as to consider two stages in the optimal control problem. The computation of optimal assisted sit-to-stand motions of impaired humans offers the opportunity to optimize design parameters for mobility assistance devices providing adequate support. Based on the support actions for the sit-to-stand motions computed for two different levels of impairment, optimal mechanical design parameters for two different sit-to-stand assistance devices are generated. Our approach to separate the human-device interaction at their interface ensures that the optimal support provided to the human by the device is not compromised by any dynamic coupling between them. Solving large-scale nonlinear optimal control problems with multiple stages, we obtain design parameters for the devices which are optimal in terms of the workspace and the mechanical effort required