%0 Generic
%A Kopitzsch, Ruth Malin
%C Heidelberg
%D 2020
%F heidok:27748
%R 10.11588/heidok.00027748
%T Analysis of Human Push Recovery Motions Based on Optimization
%U https://archiv.ub.uni-heidelberg.de/volltextserver/27748/
%X The ability to cope with large perturbations is essential to avoid falling for humans as well as for humanoid robots. Every day millions of people are affected by injuries due to falling. This is a huge problem not only for the individuum but also for the society as it costs the health care systems billions of euros. Also in the field of humanoid robots fall avoidance is very important as it protects robots against breakage.   In this thesis, the problem of fall avoidance is addressed using a combination of optimization, human-modeling and recorded push recovery motions. The aim is to identify the principles that lead to human-like push recovery motions. The human is modeled by rigid segments combined by joints leading to an underactuated multi-body representation. These models are included in multiple stage optimal control problems to reconstruct and sythesize human push recovery motions considering the dynamics of a human over the whole time horizon. Due to the high nonlinearity, the optimization problem is solved based on a direct multiple shooting method.  To analyze the human push recovery motions, dynamically-consistent motions for the model that closely track experimental data are produced. The joint angles and joint torques for the human model controlled by joint torque derivatives are compared for perturbed and unperturbed motions from two subjects. The results verify the assumption that the heavier the perturbation is and the higher it is applied at the upper body, the larger are the resulting joint torques. We show that including optimally chosen spring-damper elements in the joints can reduce the active joint torques significantly. We further exploit our motion reconstruction approach to determine the states that are most affected during a perturbation. Relevant parameters such as the orientation and position of the head and body, joint angles and torques of the perturbed motions are analyzed for deviations to the unperturbed motions at the point in time when the push occurs. Identifying the point in time when the model states of the perturbed motions differ from the unperturbed motions, the reaction times are determined.  To better understand human push recovery motions, we also investigate in a motion sythesis approach. This approach enables a control hypothesis, in the form of a specific objective function, to be formed. The minimization of effort combined with a periodicity formulation results in human-like motions and the influence of the push strength is analyzed. Formulating the objective function as a weighted linear combination of possible optimality criteria provides the possibility to analyze different optimality criteria and their resulting motion. The difficulty is, that for a given motion, it is not known, which criteria lead to that specific motion. In this thesis, the results for different basal objective functions are analyzed. These studies prepare to determine the optimal weights of the criteria by including the presented motion generation formulation in an inverse optimal control problem. Having analyzed general weights that lead to a good approximation of the human recovery motions, the resulting objective function can be used to generate push recovery motions also for humanoid robots or assistive devices such as exoskeletons.  To show another application in the improvement of technical assistive devices, we include two combined human exoskeleton models of different weights in our calculations. This allows us to analyze the joint torques for these models including the exoskeletons and compare the results to a human model. As the resulting joint torques are quite large, we also formulate combined human exoskeleton models with passive spring-damper elements that act in parallel to the active torques. This compliant formulation leads to a significant reduction of the active joint torque needed for the recovery motion. The reduction of the active joint torques allows the reduction of energy needed for the recovery motion or can enable the recovery from stronger perturbations.