Fast and Accurate Foot and Muscle Models for the Prediction of Human Walking using Optimal Control

• Applicant: Dr. Matthew Millard
• Subject Area: Automation, Mechatronics, Control Systems, Intelligent Technical Systems, Robotics
• Term: from 2016 to 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 299133258

Studying how people use their musculoskeletal system to create movement is very challenging: measuring muscle force in-vivo is not possible without invasive surgery. The barriers to studying the human musculoskeletal system make it challenging to perform basic research, design prosthetics, and plan surgeries. Physics-based modeling provides a powerful lens for observing the musculoskeletal system without the need for surgery. The mathematics of optimal control can be used to predict the movements of a musculoskeletal model. One way to predict human walking patterns is to use an optimal control method to search for the muscle force waveforms, which when applied to the model and simulated, result in the most efficient walking movement. Given a realistic musculoskeletal model and a physiologically accurate cost function, the optimal control method can predict the motion a human subject would take. Unfortunately current models are not well suited for fast and accurate motion prediction: models of foot-ground contact are computationally expensive, and the nonlinear characteristics of musculotendon dynamics make it challenging to find an initial feasible solution. To address these problems we will first develop a foot-ground contact model that is accurate and well suited for optimal control. Most of the foot-ground contact models in the literature are compliant. Simulating the foot as a compliant element is computationally expensive due to the wide ranging stiffness of the foot pads: 20 N/mm at initial contact and to 1600 N/mm at 1 body weight of load. Since the foot pads are so stiff we will model the foot using smooth rigid parametric geometry that we will fit to experimental data. Second, we will make a Hill-type musculotendon model which can have all of its nonlinearities removed and smoothly re-introduced. The blendable Hill-type musculotendon model will allow us to more easily find an initial feasible solution, and iteratively find more realistic solutions. Finally, we will use optimal control to solve for a physiologically efficient gait of a fitted musculoskeletal model. We will solve for an efficient walking gait for this model using a multiphase optimal control method. To support the research of others we will contribute implementations of our work to OpenSim and RBDL, open-source codes which are used by thousands of biomechanics and robotics researchers world-wide.Predicting the motions and musculotendon forces of a musculoskeletal model is a long standing problem in biomechanics. Today it is challenging to predict human motion, and only a few research groups around the world have accomplished this goal. In 3 years it will be possible for researchers, prosthetic designers, and surgical teams to gain new insight into the human musculoskeletal system in a timely and accurate manner.

• DFG Programme: Research Grants
• Co-Investigator: Professorin Dr.-Ing. Katja Mombaur