You are here: Home / Projects / Past Projects

Past Projects

 

Minimizing the Effort of a 2D Model of Human Gait by Optimizing Joint Torque Controller Gains

Ben Baldwin

One of the steps toward understanding human gait is to create simplified models of the human body and then to simulate those models.  This project applies a random perturbation force that grows steadily over time to the torso of a torque driven 2D model to investigate how particular joint gains affect the overall model effort and walking time. Specifically, the goal is to determine whether or not there is an optimum set of joint torques that maintains a satisfactory walking time while simulataneously minimizing the overall model effort.  The contestation put forth is that gains will increase to a point, beyond which, additional increases will not yield better results.  This is investigated with a model that is driven by torques generated by a simple proportional-derivative controller.  The gains of the controller are the same for each joint (two gain model), allowing a three dimensional representation of the effort or walk time on one axis plotted against the proportional gain and the derivative gain on the other two axes.  These plots can be generated either by a grid method, which runs the model at predetermined intervals of proportional and derivative control, or by using a more sophisticated optimization routine such as particle swarm optimization (PSO).  PSO is also used to determine if using separate gains for each joint (six gain model) yields a lower effort than the two gain model.  The exploration of the six gain model precludes plotting the results.  Therefore, only the optimum gains found by the routine are shown as the output.


 

Gait Optimization using GPOPS-II

Milad Zarei

I used GPOPS-II [Matlab Optimal Control Software] to optimize squared torques of one DOF pendulum problem to evaluate GPOPS-II to be used in Human Motion & Control Lab. I am currently working to optimize 2D gait problem using GPOPS-II. The optimization problem that I am talking about is to find cyclic gait with a prescribed speed and cycle time, minimizing the integral of squared torques. The initial guess is static standing, which was easy to solve. The problem is very nonlinear because of foot-ground contact in the gait model.


 

Methods for Identification of Feedback Control During Standing 

Samin Askarian

The mechanism of human balance control could be studied by a direct approach (DA) in which a relationship between observed joint moments and potential feedback signals was identified. However, the human balance system operates in a closed loop and this would bias the estimated controller towards the inverse of the plant, i.e. inverse multibody dynamics. The aim of this work was to validate the direct approach method for identification of feedback control in human standing and to study the effect of platform perturbation amplitude on the accuracy of the DA identification technique. Furthermore, indirect approach (IA) was used for the same system to remove the systematic error in gain estimation. Test data were obtained from a simulation in which the plant was modeled as a double inverted pendulum, perturbed with horizontal accelerations at the base to mimic a test protocol for human standing balance.


 

Validation of an Accelerometry Based Method of Human Gait Analysis

Obinna Nwanna

Gait analysis is the quantification of locomotion. Understanding the science behind the way we move is of interest to a wide variety of fields. Medical professionals might use gait analysis to track the rehabilitation progress of a patient. An engineer may want to design wearable robotics to augment a human operator. Use cases even extend into the sport and entertainment industries. Typically, a gait analysis is preformed in a highly specialized laboratory containing cumbersome expensive equipment. The process is tedious and requires specially trained operators. Continued development of small and cheap inertial measurement units (IMUs) offer an alternative to current methods of gait analysis. These devices are portable and simple to use allowing gait analysis to be done outside the laboratory in real world environments. Unfortunately, while current IMU based gait analysis systems are able to quantify a subject’s joint kinematics they are unable to measure joint kinetics as could be done in a traditional gait laboratory. A novel musculoskeletal model-based movement analysis system using accelerometers has been developed that can calculate both joint kinematics and joint kinetics. The aim of this master’s thesis is to validate this accelerometry based gait analysis against the industry standard optical motion capture gait analysis.


 

Optimal Linear Feedback Control of Walking

Raviraj Nataraj

This project involves the development of optimal feedback controllers that efficiently reject perturbations about desired state and control trajectories during walking. We initially identify time-dependent state feedback gains optimized for linear quadratic regulator (LQR) control of joint torques according to performance objectives of tracking versus effort. We then characterize controller performance against various perturbations. Long-term objectives of this work include: (1) developing controllers for muscle-based actuation, (2) development of linear quadratic Gaussian (LQG) control for the case of incomplete and noisy sensing, and (3) designing control templates for operating powered prosthetic and orthotic devices that improve gait performance and efficiency.


 

Inertial Compensation in a Moving Force Platform

Sandra Hnat, Rutwick Bakku

When load cells are located directly underneath a force plate, moving the platform will introduce inertial artifacts in the force measurements of the load cell.  To compensate for these errors, we've developed a simple, accelerometer-based technique that assumes a linear relationship between force and acceleration. Artifacts due to inertia and gravity are estimated from accelerometer signals and subtracted from  measured forces and quantified by the reduction in the root-mean-square (RMS). The method was tested experimentally on a 2 degree-of-freedom (DOF) instrumented force treadmill capable of mediolateral translation and sagittal pitch.  Mass coefficients from one trial of random movements was used to compensate for inertial errors of another trial containing different random movements. The compensation was evaluated in five experimental conditions, including platform motions induced by actuators, by motor vibration, and by human ground reaction forces. In the test that included all sources of platform motion, the root-mean-square (RMS) errors were 39.0 N and 15.3 Nm in force and moment, before compensation, and 1.6 N and 1.1 Nm, after compensation. A sensitivity analysis was performed to determine the effect on estimating joint moments during human gait. Joint moment errors in hip, knee, and ankle were initially 53.80 Nm, 32.69 Nm, and 19.10 Nm, and reduced to 1.67 Nm, 1.37 Nm, and 1.13 Nm with our method. It was concluded that the compensation method can reduce the inertial and gravitational artifacts to an acceptable level for human gait analysis.


 

A Neuromuscular Reflex Controller for Prostheses and Exoskeletons

Sandra Hnat

Recent powered lower-limb prosthetic and orthotic (P/O) devices aim to restore legged mobility for persons with an amputation or spinal cord injury. Though various control strategies have been proposed for these devices, specifically finite-state impedance controllers, natural gait mechanics are not usually achieved. The goal of this project was to invent a biologically-inspired controller for powered P/O devices. We hypothesize that a more muscle-like actuation system, including spinal reflexes and vestibular feedback, can achieve able-bodied walking and also respond to outside perturbations. The outputs of the Virtual Muscle Reflex (VMR) controller are joint torque commands, sent to the electric motors of a P/O device. We identified the controller parameters through optimizations using human experimental data of perturbed walking, in which we minimized the error between the torque produced by our controller and the standard torque trajectories observed in the able-bodied experiments.

In simulations, we then compare the VMR controller to a four-phase impedance controller. For both controllers the coefficient of determination R^2 and root-mean-square (RMS) error were calculated as a function of the gait cycle. When simulating the hip, knee, and ankle joints, the RMS error and R^2 across all joints and all trials is 15.65 Nm and 0.28 for the impedance controller, respectively, and for the VMR controller, these values are 15.15 Nm and 0.29, respectively. With similar performance, it was concluded that the VMR controller can reproduce characteristics of human walking in response to perturbations as effectively as an impedance controller.


 

Evaluation of a Virtual Muscle Model using a Powered Exoskeleton

Sandra Hnat, Huawei Wang, and Raviraj Nataraj

Phase-based impedance controllers are the current standard for controlling a powered orthosis. We hypothesize that the spring-damper characteristics of an impedance controller can be achieved with virtual muscles. Here, we design a muscle controller that produces a joint torque signal prescribed to the electric motors of a powered exoskeleton. The controller was modeled and built in MATLAB\textsuperscript{TM} Simulink\textsuperscript{TM} for use on the Parker Hannifin Indego\textsuperscript{\textregistered} exoskeleton. To test the force-length and force-velocity relationships of our muscles, we program the Indego to perform standard isokinetic and isometric knee rehabilitation exercises, in which the joint angle, velocity, and torque is measured by a Biodex-2 isokinetic dynamometer. In the isometric results, root-mean-square (RMS) error between the measured and commanded extension and flexion torques are 3.28 Nm and 1.25 Nm, respectively. In the isokinetic trials, we receive RMS error between the measured and commanded extension and flexion torques of 0.73 Nm and 0.24 Nm. The muscle model demonstrates typical force and length dependencies as shown in human muscles undergoing the same exercises. Therefore, we conclude a virtual muscle model is capable of the same stabilizing properties as observed in an impedance controller.