Predictive gait simulations for investigation of musculoskeletal structure and locomotor function

Open Access
Celik, Huseyin
Graduate Program:
Mechanical Engineering
Doctor of Philosophy
Document Type:
Date of Defense:
June 13, 2014
Committee Members:
  • Stephen Jacob Piazza, Dissertation Advisor
  • Ashok D Belegundu, Committee Member
  • Joseph Paul Cusumano, Committee Member
  • H Joseph Sommer Iii, Committee Member
  • Predictive gait simulations
  • modeling
  • musculoskeletal structure
  • locomotor function
  • sprinting
  • plantarflexor moment arm
Human locomotion is often assumed to be governed by optimality principles. To the extent that this is true, it should be possible to reproduce various human gaits (walking, running, sprinting) with a predictive approach employing some sort of optimality criterion in an optimization framework. While there are many instances of humans using aperiodic gaits in everyday life and sporting activities, previous simulations of bipedal locomotion have focused almost exclusively on periodic gaits. The main purpose of this dissertation is to implement model-based optimal controls approaches to create novel bipedal gait simulations that are both periodic and aperiodic. Those simulations are used to investigate new optimality criteria for normal human walking and to characterize relationships between musculoskeletal architecture and human sprinting performance. In our first study, a novel computational model and a simulation framework were developed to create the first simulation of aperiodic sprinting from rest. The model used was a modified spring-loaded inverted pendulum (SLIP) biped driven by torque actuators at the hip and force actuators on retracting legs. The direct multiple shooting method was used to formulate the optimization problem in which the time to traverse 20 m from rest was minimized. The initial guess to the simulation was a “jogging” simulation obtained using a proportional-derivative feedback to control trunk attitude, swing leg angle, and leg retraction and extension. Although the model was very simple, it exhibited a number of features characteristic of human sprinters, such as forward trunk lean at the start, straightening of the trunk during acceleration, and a dive at the finish. In our second study, a muscle driven computational model was developed to create simulations of normal bipedal walking using the direct multiple shooting method and evaluation of optimality criteria. We implemented a set of optimality measures derived from muscle activation, mechanical energy expenditure, or metabolic energy expenditure to represent effort; and trunk angle as well as vertical ground reaction force (GRF). Initial guesses to the optimizations were generated using a feedforward control that relied on muscle reflex loops. The simulations converged to distinct gait cycles for different optimality criteria. The additional trunk angle and vertical GRF terms helped to alleviate some undesired behaviors observed in predictive simulations of normal walking such as spikes in GRF and excessive trunk excursion. In our third study, maximum speed sprinting simulations were created with a muscle-actuated bipedal model and the direct multiple shooting method. The simulation framework and model successfully reproduced salient features of human sprinting once maximum speed has been attained. We perturbed several musculoskeletal architecture parameters of the plantarflexors in isolation (maximum isometric force, optimal fiber length, tendon stiffness, and moment arm) to investigate how variations in musculotendon architecture affect maximum speed bipedal sprinting performance. We found that increases in each parameter analyzed in the study enhanced maximum speed bipedal sprinting performance. In our fourth study, we used the computational model and simulation framework developed in the third study to investigate how variations in the maximum isometric force parameter of each major muscle group affect sprinting performance. The maximum isometric force parameter of each musculotendon actuator in the model was perturbed in isolation. The results showed that increasing each muscle’s force-generating capacity enhanced sprinting performance, but hip flexors and quadriceps were found to have the most and least potential, respectively, to increase sprinting speed. The model employed mechanisms similar to those observed in human sprinters to attain higher speeds. Additional plantarflexor and hip flexor force increased speed primarily by enhancing stride length and stride frequency, respectively. In conclusion, this dissertation is the first study to create an aperiodic bipedal sprinting simulation from rest. We demonstrated that additional optimality criteria, vertical GRF and trunk angle, have the potential to eliminate some undesired behaviors and increase fidelity of predictive walking simulations. Contrary to the experimental findings showing that sprinters have smaller plantarflexor moment arms, we found that larger plantarflexor moment arms favor sprinting performance in the maximum speed sprinting phase. The results suggest that special attention should be given to strengthening hip flexor and plantarflexor muscles to increase maximum sprinting speed. The models and simulation frameworks described in this thesis can be used to simulate other bipedal gaits with only minor modifications.