Transcript Slide - Human Motion and Control Laboratory

Virtual muscle and reflex controllers
are capable of describing human gait
and responses to perturbation
Sandra K. Hnat
Antonie J. van den Bogert
Cleveland State University
Mechanical Engineering Department
Camp Dynamic Walking
June 5, 2016
INTRODUCTION
Current Active Prostheses
 Various control strategies proposed, but
able-bodied locomotion and humanlike
control is rarely achieved
 Modelling human neurological control is
challenging
Short answer: we don’t know
Long answer: we kind of know
Virtual Muscles and Reflex Control[1]
 Promising results in both simulation and hardware
 How well reflex controllers explain human control?
[1] Geyer, H. and Herr, H. (2010). A muscle-reflex model that encodes principles of legged mechanics and produces human walking dynamics and muscle activities.
INTRODUCTION
Objectives
1
Use optimization to tune
the parameters of the
VMR system to produce
realistic joint torques
from experiments
2
Evaluate the performance
of the VMR in describing
variations within and
between gait cycles
[2] Winter, D. A. Biomechanics of Human Movement, John Wiley & Sons Ltd., New York, NY, 1979.
METHODS
1 Experiment
 10 subjects walked on treadmill for
8 minutes while being longitudinally
perturbed[3]
 Joint angles and torques calculated
through inverse dynamics
2 Muscle Model
 Inputs: joint angles and neural excitations
 Outputs: joint torque
 Muscle contraction and activation modelled as two IDEs
and simulated through fixed-step Rosenbrock solver[4]
[3] Moore, J. K., Hnat, S. K., and van den Bogert, A.J. (2015). An elaborate data set on human gait and the effect of mechanical perturbations.
[4] van den Bogert, A. J., Blana, D., and Heinrich, D. (2011). Implicit methods for efficient musculoskeletal simulation and optimal control.
METHODS
3 Reflex Controller
 Neural excitations generated by the
autonomous muscle-reflex model of
Geyer et al.[1]
 Considering lower limb prosthesis
with three muscles (simulation only)
4 Optimization
 Particle Swarm Optimization[5]
 Minimize the cost function:
𝐶 = 𝑊1
1
𝑁
𝜏𝑉𝑀𝑅 − 𝜏𝑒𝑥𝑝
2
+ 𝑊2
1
𝑁
𝑎2
[1] Geyer, H. and Herr, H. (2010). A muscle-reflex model that encodes principles of legged mechanics and produces human walking dynamics and muscle activities.
[5] Simon, D. Evolutionary Optimization Algorithms, John Wiley & Sons Ltd., New York, NY, 2013.
RESULTS
Muscle Forces
 Produces variations in peak
moment between gait cycles
L.Gastroc
L.Soleus
L.TibialisAnt
 Matches the amplitude and
timing of ankle push-off
 Mimics the shape of the
experimental torque during
swing phase
Force (N)
1000
500
0
1
2
3
4
Time (s)
5
6
HERE IS MY POSTER
ACKNOWLEDGMENTS
 Advisor: Dr. Antonie van den Bogert
 Lab Members: Anne Koelewijn, Milad Zarei, Ravi
Nataraj, Huawei Wang, Farbod Rohani, Ben Baldwin,
Brad Humphries, and Nicole Strah
 Cleveland State University, Parker Hannifin Laboratory
for Human Motion and Control
 Parker Hannifin Graduate Research Fellowship Program
 Supported by the National Science Foundation
under Grant No. 1344954.