MAXFAS:

A MOBILE ARM EXOSKELETON FOR FIREARM AIM STABILIZATION

by

Daniel M. Baechle

A thesis submitted to the Faculty of the University of Delaware in partial

fulfillment of the requirements for the degree of Master of Science in Mechanical
Engineering

Spring 2013

2013 Daniel M. Baechle

All Rights Reserved
 MAXFAS:

A MOBILE ARM EXOSKELETON FOR FIREARM AIM STABILIZATION

by

Daniel M. Baechle

Approved: __________________________________________________________
Sunil K. Agrawal, Ph.D.
Professor in charge of thesis on behalf of the Advisory Committee

Approved: __________________________________________________________
Suresh G. Advani, Ph.D.
Chair of the Department of Mechanical Engineering

Approved: __________________________________________________________
Babatunde A. Ogunnaike, Ph.D.
Dean of the College of Engineering

Approved: __________________________________________________________
James G. Richards, Ph.D.
Vice Provost for Graduate and Professional Education
 ACKNOWLEDGMENTS

I would like to express my gratitude to my advisor Dr. Sunil K. Agrawal for

his patience and expertise offered in helping me complete this project. His advice was

invaluable in helping me solve the robotics problems involved in this thesis, a field

which was entirely new to me when I started this project. He had faith in me at times

when I did not have faith in myself.

I would also like to thank my committee members, Dr. Ioannis Poulakakis and

Dr. Herbert Tanner for examining my thesis dissertation and providing their insights

during my dissertation defense.

I offer my sincere thanks to my team leader Eric Wetzel at the U.S. Army

Research Laboratory. His original ideas have motivated my work efforts throughout

the years, and his application of technical expertise to the benefit of the soldier

provides the inspiration I use in shaping my current and future career.

I cannot thank Dr. Ying Mao enough for his assistance throughout this project.

Without his advice, I could not have completed this project. Dr. Mao had nearly

completed his Ph. D when I was starting my thesis. He had no obligation to help me

after completing his Ph. D and starting a job far away, but continued to offer advice

whenever I asked.

Many of the students in the Mechanical Systems Laboratory assisted my efforts

and experiments, and I would be remiss not to thank them. In particular, I would like

to thank Xin Jin, Damiano Zanotto, Vineet Vashista, Paul Stegall, Joon Park and

Shazlin Sharaudin for their exemplary help.

iii
 Through decades of example, my parents Daniel A. Baechle and Kathryn A.

Murphy taught me the strong work ethic I required to complete my classes and thesis.

Indeed, without their encouragement, teachings, and support, I would not be in the

position I am in now. I will always be grateful for all they have done for me.

And finally, I cannot thank my beautiful wife Christy Clark Baechle enough. We

were dating when I began my studies for this degree. Her patience and devotion

through my long nights and absent days helped me realize that she was indeed the one

with whom I should spend the rest of my life. My greatest reward is the extra time I

will have to spend with my love, my life, my wife Christy.

iv
 TABLE OF CONTENTS

LIST OF TABLES ....................................................................................................... vii

LIST OF FIGURES ..................................................................................................... viii
ABSTRACT ................................................................................................................... x

Chapter

1 INTRODUCTION .............................................................................................. 1

1.1 Background: Exoskeletons ........................................................................ 1

1.2 Aim Stabilization ....................................................................................... 7
1.3 MAXFAS: Mobile Arm eXoskeleton for Firearm Aim Stabilization ....... 8

2 PRELIMINARY AIMING EXPERIMENT..................................................... 10

2.1 Experiment .............................................................................................. 10

2.2 Discussion & Conclusions ....................................................................... 15

3 MAXFAS ......................................................................................................... 17

3.1 Introduction ............................................................................................. 17

3.2 Design and System .................................................................................. 17

3.2.1 Exoskeleton Design ..................................................................... 17

3.2.2 Composite Braces ........................................................................ 18
3.2.3 Exoskeleton System..................................................................... 27
3.2.4 Control and Filter: Original Method............................................ 30
3.2.5 Control: Modified Method .......................................................... 37

4 HUMAN SUBJECTS EXPERIMENTS .......................................................... 39

4.1 Introduction ............................................................................................. 39

4.2 System ..................................................................................................... 39
4.3 Design of Experiments ............................................................................ 41

4.3.1 Data Analysis............................................................................... 44

4.4 Results ..................................................................................................... 44

4.5 Discussion................................................................................................ 50

v
 5 CONCLUSION ................................................................................................ 53

5.1.1 Suggestions for Future Work....................................................... 54

BIBLIOGRAPHY ........................................................................................................ 56

Appendix

A INFORMED CONSENT FORM ..................................................................... 60

B IRB APPROVAL LETTER ............................................................................. 64

vi
 LIST OF TABLES

Table 1: Preliminary experiment scenarios .................................................................. 12

Table 2: Ply properties used in computation ................................................................ 19

Table 3: Joint rotation rates .......................................................................................... 34

Table 4: Test trials ........................................................................................................ 35

Table 5: Denavit-Hartenberg table for MAXFAS ........................................................ 37

Table 6: Revised test trials ........................................................................................... 38

Table 7: Test group demographics ............................................................................... 44

vii
 LIST OF FIGURES

Figure 1: GE's Hardiman exoskeleton, c. 1968 [1]. ..................................................... 1

Figure 2: Raytheon's XOS 2 [5] ................................................................................... 2

Figure 3: Lockheed Martin's HULC [3] ....................................................................... 3

Figure 4: University of Delaware's ALEX (left) and ALEX II (right) devices

[7], [8] ........................................................................................................... 4

Figure 5: University of Delaware's CAREX [9]........................................................... 5

Figure 6: BONES [10] (left) and CADEN-7 [11] (right) ............................................. 5

Figure 7: SAM [12] (left) and RUPERT [13] (right) ................................................... 6

Figure 8: WOTAS device [15] ..................................................................................... 7

Figure 9: Device for preliminary experiments ........................................................... 11

Figure 10: Subject during preliminary experiment ...................................................... 11

Figure 11: Average standard deviation of laser point .................................................. 13

Figure 12: IR laser x-position for representative subject, scenarios 1 (top) and 5
(bottom)....................................................................................................... 14

Figure 13: Accelerometer data for representative subject, scenarios 1 (top) and 5
(bottom)....................................................................................................... 14

Figure 14: Gyroscope data for representative subject, scenarios 1 (top) and 5
(bottom)....................................................................................................... 15

Figure 15: An illustration of MAXFAS, cables shown in red for clarity ..................... 18

Figure 16: Early silicone molds and braces .................................................................. 21

Figure 17: Final molds.................................................................................................. 21

Figure 18: Molds ready for composite lay-up (left), composite plies (right) ............... 22

viii
Figure 19: Layup process and cable termination block for upper arm mold ................ 23

Figure 20: Routing tubes on upper arm mold ............................................................... 23

Figure 21: Vacuum bag process and materials ............................................................. 24

Figure 22: Outer (left) and inner (right) upper arm braces ........................................... 26

Figure 23: Forearm brace ............................................................................................. 27

Figure 24: The shoulder cuff and full exoskeleton ....................................................... 28

Figure 25: Shoulder cuff, tension sensors, and motors (left), and close-up of a
motor and cable reel (right)......................................................................... 29

Figure 26: Separation of voluntary and tremorous motion (top), and BMFLC
estimate of tremor signal (bottom) .............................................................. 32

Figure 27: Algorithm flow chart................................................................................... 35

Figure 28: Coordinate frames of the arm ..................................................................... 36

Figure 29: Revised algorithm flow chart ...................................................................... 38

Figure 30: Diagram of aim tracking setup .................................................................... 40

Figure 31: Target mounted 4 m away from shoulder cuff............................................ 41

Figure 32: Experimental protocol ................................................................................. 42

Figure 33: Subject performing the aiming task while wearing the exoskeleton .......... 43

Figure 34: Average distance to target center (Rc) for MAXFAS subjects ................... 45

Figure 35: Average Rc for MAXFAS subjects normalized to trial 1 ........................... 46

Figure 36: Average Rc for control group subjects ........................................................ 47

Figure 37: FFT analysis of joint speeds without and with motor control .................... 48

Figure 38: Average absolute rotation rates for MAXFAS subjects without and
with motor control....................................................................................... 49

Figure 39: Average absolute laser distance from center for MAXFAS subjects ......... 50

ix
 ABSTRACT

Accurately aiming and firing a pistol requires a steady hand. While many

devices can steady a shooters arm or hand by restricting movement or degrees of

freedom, few devices actively reduce involuntary tremors while allowing larger

voluntary aiming movements. This paper details the design, fabrication, and

experimental evaluation of an arm exoskeleton that can actively damp arm tremors

while allowing voluntary aiming movements. The MAXFAS (Mobile Arm

eXoskeleton for Firearm Aim Stabilization) device allows five degrees of freedom,

controlling four degrees of freedom using thin steel cables. The cable-driven

architecture allows the control motors to be mounted away from the arm, resulting in

device that adds very little weight and inertia to the arm. The cables attach to braces

that are mounted on the forearm and upper arm. Weight of MAXFAS is further

reduced compared to conventional arm exoskeletons through the careful design and

manufacture of the braces using stiff, lightweight carbon fiber composites. The initial

design called for tremorous movement to be filtered out from voluntary motion, and
an adaptive algorithm to provide a tremor-cancelling signal to the cable control

motors. Experiments described in this paper involved a control method which used the

motors to only actively allow voluntary motion, thus attempting to passively damp

tremorous motion. The device was evaluated on subjects performing a pistol aiming

task, using an airsoft pistol with a laser sight. Results indicate that this simpler control

mode was effective at reducing the amplitude of motion in all five degrees of freedom.

During a simulated shooting task, this control mode improved shooting performance

x
both while subjects were wearing the device, as well as 5 minutes after removing the

device. These results will be discussed, and future experiments will be suggested.

xi
 Chapter 1

INTRODUCTION

1.1 Background: Exoskeletons

Over the past 40 years, extensive research has taken robotic exoskeletons from

realm of science fiction to reality. The concept is simple a wearable device that can

aid human performance, be it strength, speed or agility. However, early robotic

technology was insufficient to make such a device feasible for real-world use. Early

devices such as GEs Hardiman (Figure 1), though designed to aid human

performance, were enormous, heavy, and inefficient [1]. Advances in computer

control, structural materials science, energy storage batteries and actuation

technologies have allowed exoskeletons to dramatically shrink in size while improving

effectiveness.

Figure 1: GE's Hardiman exoskeleton, c. 1968 [1].

1
 Raytheons full-body XOS 2 (Figure 2) weighs 195 lbs and increases the

wearers strength by a factor of 17, allowing a person to lift loads that would

otherwise be difficult or impossible unaided [2]. The XOS 2 is a tethered device, with

hydraulic power supplied by cables running to equipment that must remain nearby.

Lockheed Martins HULC device (Figure 3), which has evolved from work at

Berkleys exoskeleton group, is an untethered leg exoskeleton consisting of rigid

titanium links with joints at the hip, knee and ankle [3]. Hydraulics provide power

assistance at the hip and knee joints. The device allows the wearer to carry up to 200

lb loads for 20 km on a single battery charge, transferring the load from the back to the

ground through the titanium links. The HULC device is currently being field-tested by

the U.S. Army to determine if the device improves metabolic efficiency of the wearer

[4].

Figure 2: Raytheon's XOS 2 [5]

2
 Figure 3: Lockheed Martin's HULC [3]

Exoskeletons have also proven useful tools for human rehabilitation. Rather than

only increasing performance while the user is wearing the device, exoskeletons can be

used to re-train disabled people to properly use their appendages. The ALEX and

ALEX II devices at the University of Delawares Mechanical Systems Laboratory are

tethered leg exoskeletons for rehabilitation of stroke victims (Figure 4) [6], [7], [8].

The devices use an assist-as-needed control paradigm. A computer creates a desired

gait path for the wearer. If the wearers leg moves perfectly along the desired path, the

ALEX device applies no force to the leg. However, if the wearers leg strays from the

desired path, the device applies a force to guide the leg onto the desired path. The

amount of applied force is governed by a virtual force tunnel around the desired

path. Studies using the ALEX device have shown that the device and control method

are effective at modifying the gait of healthy subjects and stroke patients after training

in the devices [6] [7].

3
 Figure 4: University of Delaware's ALEX (left) and ALEX II (right) devices [7], [8]

The University of Delaware group has also developed a cable-driven arm

exoskeleton for rehabilitation purposes. The CAREX device uses a similar assist-as-

needed control method, using a virtual force tunnel to guide the wearers arm along a

path for a predetermined task (Figure 5) [9]. Training with the CAREX device has

been shown to improve a healthy subjects ability to follow an invisible path, even

after the force control is turned off. Critically, the mass of extra equipment on the arm

is only 1.32 kg, approximately an order of magnitude lower mass on the arm than

other arm exoskeletons for rehabilitation. These devices include BONES [10],

CADEN-7 [11], SAM [12], and RUPERT [13], seen in Figure 6 and Figure 7. These

rigid exoskeletons can compensate for the extra static weight on the arm using motors,

but the added inertia of the device caused by arm motion cannot be compensated.

Additionally, rigid-link exoskeletons require precise joint alignment for optimal

performance. Even with proper alignment, the rigid exoskeleton can cause discomfort

for the wearer. Unlike typical robotic joints, arm joints do not rotate perfectly about

4
one stationary point [14]. The rigid exoskeleton can be aligned to the arm joints in one

position, but the arm joint centers can become misaligned with the robot joints during

arm movement. Such misalignment can lead to poor performance, discomfort and

even skin sores for the wearer.

Figure 5: University of Delaware's CAREX [9]

Figure 6: BONES [10] (left) and CADEN-7 [11] (right)

5
 Figure 7: SAM [12] (left) and RUPERT [13] (right)

Rocon and Pons have developed an arm exoskeleton specifically for suppression

of pathological tremors in the upper arm [15]. Their WOTAS device consists of DC

motors mounted on a rigid orthotic structure on the arm, providing control of 3

degrees-of-freedom (DOF) at the elbow and wrist (Figure 8). Though the device has

rigid links and motors mounted directly on the arm, it adds only 850 g to the arm.

MEMS gyroscopes mounted on the arm measure tremorous movement in the wearers

arm. A computer algorithm separates tremorous movement from voluntary movement

in the gyroscope signal, and estimates the frequency of the tremor. The DC motors

then create anti-phase movement in the arm in order to cancel the measured tremor.

Results of a small trial found the WOTAS device able to reduce tremor power by an

average of 40% for wearers with essential Parkinsonian and other pathological tremor.

6
 Figure 8: WOTAS device [15]

1.2 Aim Stabilization

Tremors in the arm, whether pathological or postural (due to holding the arm

outstretched), have also been shown to negatively affect aiming tasks [16], [17], [18].

However, accuracy when aiming and shooting a firearm depends on many factors.

These factors can be divided into three primary groups: environmental, hardware, and

human factors. Numerous devices exist to mitigate the negative influence of

environmental factors such as available light, ambient noise, and wind. Hardware has

been made as accurate as possible using special barrels, ammunition, scopes and laser

sights. However, even the most accurate firearm in an ideal environment is subject to
the human factors that affect aim, which include fatigue [19], heart rate [20], shooting

experience [21], body sway [22], and arm tremors. Some static devices attempt to

stabilize the arm and reduce tremors by reducing range of motion or degrees of

freedom [23]. Other devices attempt to reduce arm tremors using passive elements

such as springs or dampers [24]. A 1991 US patent describes a device that is

7
essentially a gyroscope mounted to the back of the hand, intended to reduce tremors in

the hand [25]. Similarly, a rifle-mounted gyroscope to reduce small aiming errors was

awarded a US patent in 1992 [26]. Tactical Electronics currently offers a

gyroscopically stabilized rifle platform for the same purpose [27]. However, none of

these devices actively sense and cancel arm tremors, as in the WOTAS device.

1.3 MAXFAS: Mobile Arm eXoskeleton for Firearm Aim Stabilization

The device described in this work is a wearable arm exoskeleton which senses

and cancels tremorous motion in the arm of the wearer during an aiming and firing
task. Similar in functionality to the WOTAS device, adaptive filters separate

tremorous movement, allowing larger voluntary movements necessary for aiming. The

mobile arm exoskeleton for firearm aim stabilization (MAXFAS) device of this paper

is novel in several key ways. The MAXFAS device is cable-driven: motors are not

mounted on the arm, but above or behind the user. The cable-driven design allows for

a lighter exoskeleton, adding less than 280 g to the wearers arm. Carbon fiber

composite materials also make MAXFAS very low-weight, while maintaining the

structural rigidity required to control the arm. The MAXFAS device is an evolution of

the CAREX device. Unlike CAREX and WOTAS however, MAXFAS has no rigid

joint at the elbow, eliminating joint alignment issues. The MAXFAS device consists

of braces attached to the forearm and upper arm. MAXFAS allows five DOF:

flexion/extension, adduction/abduction, internal/external rotation of the shoulder,

flexion/extension of the elbow, and pronation/supination of the forearm. The device is

designed such that forearm rotation is somewhat constrained to minimize rotation

away from the desired aiming position. The wrist and hand are not controlled. A

preliminary experiment without the MAXFAS device was first performed to

8
determine the sensing requirements of the device, and if wrist control was necessary.

This preliminary experiment will be described in the following chapter. Design of the

MAXFAS device, control system, and adaptive filter is described in later chapters,

along with the design and results of human experiments.

9
 Chapter 2

PRELIMINARY AIMING EXPERIMENT

2.1 Experiment
Prior to designing the MAXFAS device, an experiment was performed to

evaluate key equipment and parameters to be used in the MAXFAS experiments. The

MAXFAS device required a sensor accurate enough to detect tremors in the arm.

Additionally, the MAXFAS experiments involved the subject aiming a toy pistol at a

target, and thus required the ability to track the subjects aim point on the target. The

preliminary experiment tested a VN-100r inertial measurement unit (IMU) from

Vectornav Technologies, LLC. The VN-100r incorporates a 3-axis accelerometer, 3-

axis gyroscope, and 3-axis magnetometer. The VN-100r also uses an on-board

processor to filter and compensate for drift in real-time. The VN-100r measures

36339 mm, and weighs only 13 g (not including power/data cable). For the

preliminary experiment, the VN-100r was attached to the handle of a toy pistol, as

seen in Figure 9. The barrel of the pistol was replaced with a 780 nm wavelength IR

laser. A visible red laser was mounted underneath the barrel, as seen in Figure 9. A

Vicon Bonita motion capture system, which operates at 780 nm, was used to track the

IR laser on a 3030 cm white target with black crosshairs, seen in Figure 10.

10
 Vicon IR
camera

IR laser

Red laser

IMU
sensor

Figure 9: Device for preliminary experiments

Figure 10: Subject during preliminary experiment

Each subject was instructed to hold the toy pistol and aim the red laser at the

target (approximately 2.5 meters from the subject) for 30 seconds in 7 different

11
scenarios. The scenarios are enumerated in Table 1, and were designed to test the

effect of static arm and wrist stabilization and constraint on arm tremor and aim. A

simple wrist brace was used to constrain the dominant wrist in scenarios 2, 3 and 6.

Subjects wrested the forearm of their dominant arm on a 140 cm tall stand in scenarios

5 and 6.

Table 1: Preliminary experiment scenarios

Scenario Arms Wrist Brace Arm Stand

1 1 N N
2 1 Y N
3 2 N N
4 2 Y N
5 1 N Y
6 1 Y Y

Ten right-handed subjects were tested: 9 males and 1 female. Only one subject
had any shooting experience. Figure 11 presents the standard deviation of the IR laser
x- and y-positions, averaged across all 10 subjects. As the experiment is primarily

concerned with differences in tremors between the scenarios, only standard deviation

of the aim point is presented. Accelerometer and gyroscope data from the VN-100r

was recorded to directly measure the subjects arm tremors. A representative plot of

IR laser x-position vs. time for a single subject in trials 1 and 5 can be seen in Figure

12. A representative plot of accelerometer data vs. time for a single subject in trials 1

and 5 can be seen in Figure 13. A representative plot of gyroscope data vs. time for a

12
single subject in trials 1 and 5 can be seen in Figure 14. Figure 12 illustrates the static

signal recorded by the Vicon system while the laser sat motionless and untouched,

indicating that the Vicon system can record the laser point with sub-millimeter

accuracy. Figure 13 and Figure 14 also illustrate the static signal from the

accelerometer and gyroscope, respectively, which was recorded while the VN-100r sat

motionless and untouched.

Figure 11: Average standard deviation of laser point

13
Figure 12: IR laser x-position for representative subject, scenarios 1 (top) and 5
(bottom)

Figure 13: Accelerometer data for representative subject, scenarios 1 (top) and 5
(bottom)

14
Figure 14: Gyroscope data for representative subject, scenarios 1 (top) and 5 (bottom)

2.2 Discussion & Conclusions

Comparing scenarios 1 to 2 and 5 to 6 in Figure 11, it is evident that the

standard deviation of the IR laser aim point did not significantly decrease when the

subjects were wearing the wrist brace. This result indicates that wrist control is not a

critical factor in improving aim, and will not be necessary in the MAXFAS device.
Comparing scenarios 1 to 5 and 2 to 6 in Figure 11, it is evident that the standard

deviation of the IR laser aim point significantly decreased when the subjects rested

their dominant arm on the stand. This result can also be seen in Figure 12. This result

indicates that a device that can steady the arm during aiming will significantly

improve aim. The accelerometer data in Figure 13 indicates that the accelerometers in

the VN-100r are not sensitive enough to detect significant changes in arm tremor.

However, the gyroscope data in Figure 14 indicates that the gyroscopes are sensitive

15
enough to detect significant changes in arm tremor. As such, the gyroscopes on the

VN-100r will be used to track arm tremors in the MAXFAS device.

16
 Chapter 3

MAXFAS

3.1 Introduction
MAXFAS is a unique arm exoskeleton designed specifically to improve aim

while adding little mass and volume to the arm. The basic concept of a cable-driven

arm exoskeleton was adapted from the CAREX device. However, whereas CAREX

used force control to guide the arm along a specified path in space, MAXFAS was

designed to allow a wider variety of large movements of the arm while actively

damping only the small tremors in the arm. This chapter details the design and

fabrication of the MAXFAS device, considering the functional requirements and

physical constraints.

3.2 Design and System

3.2.1 Exoskeleton Design

An illustration of the MAXFAS device can be seen in Figure 15. The

exoskeleton consists of three braces: one on the anterior (inner) forearm, and one each

on the inner and outer upper arm. Six cables are used to control the arm: two cables

terminating at each brace. Thus, four cables govern the flexion/extension,

adduction/abduction, and internal/external rotation of the upper arm. Two cables

terminating on the forearm brace near the wrist govern flexion/extension of the

forearm. These two cables are routed through the outer upper arm brace, near the

17
elbow. Note that although forearm pronation/supination cannot be explicitly controlled

in this configuration, the location of the cable termination points on the forearm brace

and routing points of these cables on the upper arm brace constrain rotation away from

the proper aiming position when the cables are tensioned. Additionally, the forearm

brace itself constrains forearm rotation away from the proper aiming position. Cable

termination and routing points were chosen based on the CAREX design.

Figure 15: An illustration of MAXFAS, cables shown in red for clarity

3.2.2 Composite Braces

Each brace consists of a 1 mm thick carbon fiber laminate shell, custom

manufactured to minimize weight while providing the structural stiffness necessary to

transmit control to the arm using cables. The shell consists of 8 plies of carbon fiber

prepreg (Cytec Cycom IM7-381). Each ply is unidirectional, having carbon fibers

running in only one direction. Ply orientation design analysis was performed using a

custom program in MATLAB. The program uses the principles of classical laminate

theory to determine the mechanical properties of any sequence of ply orientations. The

18
program was used to minimize coupling stiffnesses that can lead to undesirable

behavior under load. Manufacturer-provided ply properties used in the program can be

found in Table 2. Poissons ratio was estimated. A ply orientation of [0/0/90/+45/-

45/90/0/0] degrees was chosen, which has the following stiffness matrices:

9.73 1.19 0
A = 1.19 5.81 0 10 7 N/M (1)
0 0 1.38

0 0 6.12

B= 0 0 6.12 10 2 N (2)
6.12 6.12 0

12.22 0.25 0
D = 0.25 2.22 0 NM (3)
0 0 0.41

Table 2: Ply properties used in computation

Longitudinal Transverse Shear Poisson

Stiffness E1 Stiffness E2 Modulus G12 Ratio 12
(GPa) (GPa) (GPa)
165 8.8 4.3 0.27

It is generally desirable to minimize B13 and B23, which couple in-plane stresses

with laminate twisting. As these terms are non-zero for this laminate, some warpage

may occur during the cure cycle [28]. Perhaps more importantly, the D13 and D23

terms, which couple bending and twisting, are 0 here. As these braces will be

19
subjected primarily to out-of-plane loads that may cause bending stresses, it is

important that the laminate not twist when subjected to such loads. Extra 0o plies were

added to the outer layers to increase stiffness along the length of the braces. Calculated

theoretical longitudinal laminate stiffness (along the length of the braces) is Ex = 94.28

GPa. The two 90o plies help provide some stiffness in the laminate transverse

direction, yielding a calculated theoretical transverse laminate stiffness of Ey = 55.95

GPa. The overall antisymmetric ply orientation was chosen as it can result in a lower

stress concentration factor than a crossply laminate (consisting of only 0o and 90o

plies) [28]. A low stress concentration factor is very desirable, as holes will need to be

drilled into the braces to mount hardware.

The prepreg plies must be laid onto a mold with the desired brace shape.

Initially, silicone molds (Figure 16) were cast of the orthotic braces used in the

CAREX device. However, early composite braces cast from these molds (Figure 16)

fit only a few people. Whereas the softer plastic CAREX braces could flex to

accommodate larger arms, the composite braces had little flex. To accommodate a

wider variety of healthy users (particularly those having soldier body types), new

molds were designed in Solidworks. The dimensions were chosen based on a small

survey of healthy military-aged males. The molds, seen in Figure 17, also feature flat

tops for mounting of the VN-100r sensors. The new molds were milled out of high-

temperature epoxy, and designed to create near-net shape braces. As such, 9 mm thick

aluminum plates were affixed to the bottom of the molds. These plates allowed excess

prepreg to drape better over the edges of the mold, rather than make a nearly 90o angle

at the edge of the mold. Sharp turns near the edges of the molds can lead to poor

consolidation of the composite laminate during cure in these areas.

20
Figure 16: Early silicone molds and braces

Figure 17: Final molds

21
 The molds were treated with two release compounds: Honey Wax (Specialty

Products, Co.) and then Universal Mold Release spray (Smooth-On, Inc.). The molds

were then placed on an aluminum caul plate with brown Teflon-coated release sheet

(RE234 TFNP50, Airtech International, Inc.) between the molds and plate, as seen in

Figure 18. The prepreg plies were laid onto the molds in the afore-mentioned ply

sequence. A small aluminum cable termination block, seen in Figure 19, was placed

on the upper arm brace after the -45o layer, 25 mm from the distal end of the brace.

The remaining 90o and 0o plies were cut slightly to accommodate the protruding cable

termination tab. Two 3.1 mm diameter Delrin tubes were affixed to the upper arm

brace using small strips of prepreg on the outermost layer of the brace (Figure 20).

These tubes were intended to help route the upper arm cables along the braces and

provide more protection to the cables. The forearm brace had no integrated cable

termination block or routing cables.

Figure 18: Molds ready for composite lay-up (left), composite plies (right)

22
Figure 19: Layup process and cable termination block for upper arm mold

Figure 20: Routing tubes on upper arm mold

23
 Once the prepreg layup process was complete, a blue release ply (1005081

Airtech) was placed over the uncured braces (Figure 21). The release ply was carefully

cut near the cable routing tubes and termination blocks. These cuts help avoid

wrinkles, which can transfer onto the finished braces, and to help the release ply drape

into sharp corners near around the routing tubes. If the release ply bridges over these

corners, the cured braces can have incomplete consolidation and dry carbon fibers

(without epoxy holding them together) in these areas. A white nylon breather ply (RC-

3000-10 Airtech) was then placed over the blue release ply to absorb excess epoxy,

and improve pressure distribution and consolidation. Nylon vacuum bag (WN 1500

Airtech) was placed over the white breather ply and sealed with black Tacky Tape

(Schnee-Morehead, Inc.).

Figure 21: Vacuum bag process and materials

24
 The braces were then cured per Cytecs specifications for the prepreg: 710 mm

Hg vacuum was applied to the layup for approximately 15 minutes. The plate was then

placed in a room-temperature oven and temperature was ramped to 127oC (260oF) at

2.8 oC/min (5oF/min). Temperature was held at 127oC for 100 minutes, then the oven

was turned off and the layup allowed to cool down to room temperature gradually over

the course of several hours. Vacuum was applied to the layup during the entire ramp-

up, hold, and ramp-down cure process. Temperature was monitored using a

thermocouple stuck into the Tacky Tape.

After cure, the braces were released from the molds and trimmed to final

dimensions of approximately 9532160 mm (upper, not including cable mount

point) and 7522190 (forearm). Holes were drilled in the forearm and outer upper

arm braces to attach the VN-100r sensors, cable routing (upper arm brace) and

mounting (forearm) brackets. The bracket mounting holes were drilled 12 mm from

the edges of the braces and 16 mm apart, as Chou indicated that edge effects in

composites extend a distance of 2 of the laminate thickness [28]. After drilling, a 3

mm thick foam (Plastazote) was adhesively bonded to the inside of each brace. The

foam is often used in orthoses. Two brackets were then attached to each brace using

two brass 8-32 screws per bracket. The use of steel hardware was minimized for

weight considerations as well as to avoid interference with the magnetometers in the

VN-100r sensors that would be mounted on the braces. The cable routing/mounting

brackets were custom designed and cut out of 6 mm thick aluminum plate. They are

designed to align the cables that control the forearm with the long axes of the forearm

and upper arm, with adjustability to accommodate a variety of arm sizes. A bronze 18

mm long -20 carriage bolt is connected to each forearm mounting bracket using an

25
aluminum nut. A hole drilled in the end of each bolt provides the termination point for

each of the two cables that govern the forearm motion. A 16 mm long piece of -20

threaded Teflon rod with a small lengthwise hole is mounted in each upper arm

routing bracket using two aluminum nuts. Each VN-100r sensor was attached to the

flat surface of the forearm brace and outer upper arm brace using two 4-40 brass

screws and small-profile brass nuts. The braces are attached to the arm using 25 mm

wide Velcro straps, which are in turn attached to the braces using adhesive-backed

Velcro strips. The completed braces with all hardware and cables can be seen in

Figure 22 and Figure 23.

Figure 22: Outer (left) and inner (right) upper arm braces

26
 Figure 23: Forearm brace

3.2.3 Exoskeleton System

The control cables consist of pre-stretched nylon-coated wire rope,

approximately 0.9 mm in diameter with a maximum load of 440 N (100 lbs). All six

cables are routed through two Teflon blocks on the shoulder cuff, as seen in Figure 24.

The shoulder routing blocks are adjustable along the circumference of the shoulder

cuff, but remain in place for purposes of this paper. The shoulder cuff is attached to

the same frame as the motors, and can be adjusted to each subjects shoulder height.

As the shoulder cuff is rigidly attached to a frame, it should help reduce body sway,

which has been shown to affect aiming performance [29], [30].

27
 Figure 24: The shoulder cuff and full exoskeleton

Each cable is connected to a Kollmorgen motor. Initially, high-

resolution/accuracy AKM43L motors and AKD-P01206 drives were purchased for the

project. Unfortunately due to facilities issues, these motors and drives could not be

installed. Older Goldline XT motors driven by Kollmorgen ServoStar CD amplifiers

in velocity mode were used instead. The motors are mounted on an aluminum frame

above the user. Each cable winds onto a 7.24 cm diameter Delrin reel (Figure 25),

which is custom designed to prevent the cable from wrapping onto itself. Each cable

also passes through a thin rubber sheet to keep it from unwinding from the reel under

low tension. A tension sensor (Futek LSB200) with a 220 N (50 lb) limit is mounted

in line with each cable between the shoulder cuff and the rubber sheet, as seen in

28
Figure 25. Care was taken to route the wire for each tension sensor above the sensor,

so the weight of the wire does not affect the tension reading. Each tension sensor is

paired with and calibrated for an amplifier (Futek CSG100). The tension sensors are

attached to the cable with a small plastic clip, which will separate at 90 N (20 lbs) for

safety. The Teflon routing points and nylon coating on the cables help reduce friction,

as accurate tension sensing is critical to the control of the device.

Figure 25: Shoulder cuff, tension sensors, and motors (left), and close-up of a motor
and cable reel (right).

The VN-100r gyroscopes are used to directly sense arm tremors and to

calculate joint rotation angular rates. The magnetometers are used to calculate joint

rotation angles. The IMUs sample rate is 100 Hz. A National Instruments PXIe

29
system with FPGA running LabVIEW 2012 is used to interface with the sensors and

motors, as well as provide real-time control of the exoskeleton. The real-time control

loop runs at 200 Hz on a quad-core PXIe-8133 controller.

3.2.4 Control and Filter: Original Method

MAXFAS is intended only to reduce small tremors while allowing large

voluntary movement. However, as the motors are operating in velocity mode, the large

voluntary movement of the arm must be explicitly allowed: a zero-volt signal to the

motors does not allow any movement. The gyroscopes of the IMUs mounted on the
arm braces sense both small tremors and large voluntary movements. Two second-

order point-by-point Butterworth filters are employed to separate voluntary and

tremorous motion. A 2 Hz cutoff low-pass second-order Butterworth filter isolates

voluntary motion to be explicitly allowed by the motors. In parallel, a 4-15 Hz

bandpass filter separates the tremorous movement (Figure 26). Previous studies

indicate tremors to occur primarily in the 2-15 Hz range, and more specifically the 8-

12 Hz range [15], [31]. Movement below 2 Hz is considered voluntary. Early

experimentation in MATLAB indicated a lower limit of 4 Hz would provide better

cancelling behavior for this system. The isolated 4-15 Hz tremor signal is then sent to

a bandlimited multiple Fourier linear combiner (BMFLC). The BMFLC algorithm

estimates the tremor signal as a sum of a finite number of sine and cosine signals, with

different frequencies within a predefined band fh fl. Each sine and cosine is assigned

a weight ar and br respectively, which are updated on each loop iteration to reduce

error between the estimated tremor signal and the actual tremor signal. The algorithm

can be stated as follows (equations 4-7 from [32]). At each time step k, yk forms an

estimate of tremor signal sk using the equation

30
 L
r r
y k = a r sin 2 f l + k + br cos 2 f l + k (4)
r =0 G G

where L=(fhfl)G. For r weights ar and br, equation 1 can be written using reference

input vector xk, weight vector wk, and error as

1
sin 2 +
, 1
= 1
(5)
cos 2 +
, + 1 2

= (6)

+1 = + 2 (7)

Error is calculated as the difference between signal sk and the signal estimate.

Weights ar and br (in weight vector wk) are updated on each loop iteration (time step)

to include the error, reference input vector xk, and adaptive gain . Since the tremor

estimation signal is comprised of sines and cosines, a prediction of the future tremor

signal can be made. In each iteration, the algorithms main loop calculates the next

data point of the tremor signal, negating the amplitude to create a tremor-cancelling

signal. An illustration of the algorithms tremor estimation capability can be seen in

Figure 26. The data in Figure 26 was gathered from the VN-100r during the

preliminary experiment of Chapter 2. The gyroscope data was processed through the

zero-phase filter and BMFLC offline in MATLAB. The result plotted in blue in Figure

26 is the BMFLC prediction of the future tremor at each time step.

31
 30
Raw gyro data
Tremor
20 Voluntary motion

Angular rate (rad/s)

10

0
5 7 9 11 13
-10

-20

-30
Time (s)

4
Tremor signal
3 BMFLC estimate
Signalestimate
2
Angular rate (rad/s)

1
0
10 10.5 11 11.5 12
-1
-2
-3
-4
Time (s)

Figure 26: Separation of voluntary and tremorous motion (top), and BMFLC estimate
of tremor signal (bottom)

Several variables can be chosen to maximize tremor estimation and thus

cancelling. These variables include adaptive gain parameter , number of frequency

band divisions G, and frequency band fh fl. The adaptive gain parameter affects the

algorithms ability to estimate the tremor signal, and is inversely related to the

algorithm convergence time. A value of = 0.0825 was chosen through experiments

in MATLAB as a compromise between convergence time and tremor estimation

efficacy. Similarly, experimentation in MATLAB revealed G = 5 to provide the best

tremor estimation for this system.

32
 The limits of the frequency band can also be expressed using a center frequency

f0, as

fh = f0 + fs (8)

and

fl = f0 fs (9)

Previous research [32] and experimentation in MATLAB indicate 9 Hz as a good

initial value for f0, and fs = 1 Hz for good tremor estimation. However, in practice,

different subjects may have different dominant tremor frequencies ( [15], Ch. 2). To

accommodate such differences, a fast Fourier transform (FFT) can be used in real-time

to measure the exoskeleton wearers dominant tremor frequency, which then takes the

place of the default 9 Hz value for f0.

In order to calculate joint rotation rates, gyroscope data from the IMUs must be

transformed. This transformation involves a least-squares solution for the forearm

rotation rates, since [RF] is a 32 rotation matrix in the following equation.

q 2 4
q 5
[RU ] q 3 + [RF ] = [Rs 2 ] 5 ,
(10)
q 4 q 6 6

where

q 2 1
[RU ] q 3 = [Rs1 ] 2
(11)
q 4 3

33
 In Eqs. 10 and 11, 1 3 and 4 6 represent gyroscope readings from the

VN-100r sensors on the upper arm and forearm, respectively. The terms [RU], [RF],

[Rs2], and [Rs1] represent rotation matrices for the upper arm, forearm, forearm sensor,

and upper arm sensor, respectively. The terms 2 6 are the joint angle rotation rates

listed in Table 2 below. From Eqs. 10 and 11, it is evident that calculated elbow

flexion/extension and forearm pronation/supination rates 5 and 6 , respectively, will

be affected by gyroscope signals 1 3 from the upper arm sensor. Thus, the filter

and BMFLC algorithm should be applied to the joint rotation rates after

transformation of the gyroscope data. Additionally, the BMFLC can be applied to

different joint rotation rates to investigate the effect of cancelling different tremor

components. Two scenarios to test different applications of the BMFLC have been

programmed into the algorithm, and are listed in Table 3. The general control scheme

order can be seen in Figure 27 below.

Table 3: Joint rotation rates

Joint rotation rate Anatomical movement

2 Upper arm flex/extend
3 Upper arm adduct/abduct
4 Upper arm pronate/supinate
5 Elbow flex/extend
6 Forearm pronate/supinate

34
 Table 4: Test trials

Trial Condition BMFLC Filtered signals

1 No MAXFAS N/A
2 MAXFAS, no control N/A
3 Filter A 3 5
4 Filter B 2 3 5
5 No MAXFAS N/A

26 Bandpass
16
16 & BMFLC
Tension &
filter
Gyros Transform Jacobian safe load Motors
Lowpass check
filter

Figure 27: Algorithm flow chart

The terms 1 6 in Figure 27 are the cable speeds. Once the BMFLC algorithm
has built an estimation of the tremor signal, the 4-20 Hz cancelling signal is added to

the < 2 Hz voluntary 26 signal. The combined joint rotation rate signals are then

transformed to cable speeds using the Jacobian:

{}l = [J ]{q} (7)

li
[J ] = , i = 1 6, j = 2 6 (8)
q j

The coordinate frames of the arm and the Denavit-Hartenberg parameters used to

calculate the Jacobian can be found in Figure 28 and Table 4, respectively. Given the

programmed cable speeds, the required motor speeds are calculated using the cable

35
reel diameter. The cables can only pull, not push, and thus must remain in tension at

all times. In order to ensure that the cables are not slack, an algorithm monitors each

cable tension sensor. If the cable is taut, the tremor-cancelling motor speed signal is

passed from the algorithm to the motors. If the cable is slack, a small constant signal is

sent to the motor to wind up the cable. Additionally, the algorithm ensures that the

motors do not pull too hard on the arm. If the tension in any cable exceeds 130 N (30

lbs.) for any reason, a small constant signal is sent to the motor to slack the cable.

Z0

X0 Y0
X2, Z3

X3, Z1 X1, Z2

X4, X5
Z4

Z5

Y6 X6
Z6

Figure 28: Coordinate frames of the arm

36
 Table 5: Denavit-Hartenberg table for MAXFAS

Link ai i di qi Human Joint Motion

1 0 /2 0 /2 Fictitious joint
2 0 /2 0 q 2 + /2 Shoulder flexion/extension
3 0 /2 0 q 3 + /2 Shoulder adduction/abduction
4 0 /2 d4 q 4 + /2 Shoulder internal/external rotation
5 0 /2 0 q5 Elbow flexion/extension
6 0 0 d6 q6 Forearm pronation/supination

3.2.5 Control: Modified Method

Initial evaluation of the system revealed that the available Goldline XT motors

did not have sufficient control bandwidth to produce the tremor-cancelling signals sent

by the algorithm. Therefore, the control scheme had to be modified. Two new control

scenarios were devised. The first scenario, a 2 Hz cutoff low-pass second-order

Butterworth filter operates on gyroscope signals 1 6 to isolate voluntary motion to

be explicitly allowed by the motors. The voluntary motion signals are then

transformed into joint rotation rates and cable speeds as described in section 3.2.4

above. Since the cables are only allowing the < 2 Hz large voluntary motion, higher-

frequency tremorous motion may be reduced. The second scenario is essentially the

same, but adds an aim check step. In this scenario, if the standard deviation of the

calculated joint rotation rates falls below a minimum value of 0.15 rad/s (over any 0.5

s period), the algorithm sends a 0 volt signal to all of the motors. This signal

essentially locks the cables in place, constraining the arm (although not completely).

This lock signal is maintained as long as the tension in the cables is at least 18 N (4

lbs), and the standard deviation of the joint rotation rates remains below the specified

minimum value. To exit this lock mode and return to normal motion, the wearer

37
moves their arm slightly to raise the joint rotation rates above the minimum threshold

value. The updated algorithm flow chart can be seen in Figure 29, and the updated test

trials can be seen in Table 5 below.

Mode A

16 26
16
Tension &
Lowpass
Gyros Transform Jacobian safe load Motors
filter
check

Mode B
16 26
16
Tension &
Lowpass Aim
Gyros Transform Jacobian safe load Motors
filter check
check

Figure 29: Revised algorithm flow chart

Table 6: Revised test trials

Trial Condition
1 No MAXFAS
2 MAXFAS, no control
3 Mode A
4 Mode B (lock mode)
5 No MAXFAS

38
 Chapter 4

HUMAN SUBJECTS EXPERIMENTS

4.1 Introduction
This chapter details the system and procedure used to evaluate the effectiveness

of MAXFAS on human subjects during a pistol aiming task. Experiments were

designed to evaluate the MAXFAS devices effect while wearing the device as well as

any learned effects after wearing the device. The MAXFAS devices effects on arm

tremor will also be presented and discussed. These experiments are intended to

simulate as closely as possible aiming and firing of a real pistol, without the safety

concerns associated with firing a real pistol. Use of a simulator pistol greatly

simplifies the experiments, allowing them to be performed without eye and ear

protection, and in a laboratory rather than at a firing range. Variables associated with

firing a real pistol, which are not evaluated in these simulator experiments and may

affect live-fire results, will be discussed in section 4.5.

4.2 System
Subjects in these experiments used an airsoft pistol to evaluate aim. The

KJWorks M9 PTP was chosen for its similarity in weight and functionality to the

Armys standard issue Beretta M9. The M9 PTP uses a CO2 cylinder to propel an

airsoft pellet and recoil the slide. The pistol weighs 1210 g unloaded, including clip,

CO2 gas cylinder, and two rail-mounted lasers (Figure 30). The subject uses a red laser

to aim the pistol, while a 780 nm infra-red laser is used by a motion capture system to

39
track the aim point on the target. The motion capture system consists of 8 Vicon T40-

S cameras and a MX Giganet hub. Early experiments revealed that this system can

track the laser aim point with sub-millimeter resolution at 100 Hz. The system can also

track small reflective spherical markers with the same resolution. Two such markers

were placed on the pistol slide (Figure 30) to allow time correlation of each shot.

Here, a shot means that the user pulled the trigger causing recoil but not firing a

projectile, as the gun remained empty of projectiles at all times. The shot is counted as

the IR laser point on target in the last time frame before the slide began to move

backwards in recoil.

Vicon IR
cameras
Reflective markers

y
x Red laser

IR laser

Target

Figure 30: Diagram of aim tracking setup

The target was a white 3030 cm square with black crosshairs and reflective

markers on each corner, seen in Figure 31. The target was mounted with its center 144

cm off the ground, 4 m away from the exoskeleton shoulder cuff. The target is

40
mounted so that when the subject is standing with their shoulder in the shoulder cuff,

the target is essentially centered laterally with the subject. The target was not moved to
accommodate geometry differences between subjects.

Figure 31: Target mounted 4 m away from shoulder cuff

4.3 Design of Experiments

Twenty right-handed subjects were tested for the purposes of this paper. Fifteen

subjects wore the exoskeleton during the experiments and a control group of 5 subjects
did not wear the exoskeleton. Before the experiment, each subject read and signed an

informed consent form, which can be found in Appendix A. Red and IR laser

alignment was verified on the target at range before each subject. The lasers were

aligned only with each other, and not the iron sights. Each subject was informed of

general operation of the pistol, and instructed to aim with two hands. Subjects were

instructed to aim using the red laser only, and not the iron sights of the pistol. Each

41
subject was then allowed three practice shots to familiarize themselves with the target

and operation of the pistol. Since the pistol was unloaded, the slide lock was taped

down to keep the pistol from locking open after each shot. All shots were in single

action trigger mode. After the three practice shots, each subject performed five trials

consisting of 15 shots each. Experimental protocol for the five trials can be seen in

Figure 32.
Order randomized

5 2 2 5
No No Control Control No
min min min min
Exo Control Mode A Mode B Exo
rest rest rest rest

Trial 1 Trial 2 Trial 3 Trial 4 Trial 5

5 2 2 5
No No No No No
min min min min
Exo Exo Exo Exo Exo
rest rest rest rest

Control group

Figure 32: Experimental protocol

In each trial, subjects were instructed when to start and stop, so they did not have

to keep track of the number of shots. Subjects were instructed to allow at least 1 s
between shots. Early evaluation revealed that a single CO2 cartridge could provide

recoil for up to 80 shots. The CO2 cartridge was replaced after trial 3 (45 shots) for
each subject. Decrease in recoil force was not perceptible after 45 shots. Experimental

scenarios are described in section 3.2.5 and Table 5 above. Trial 1 consisted of

shooting without the exoskeleton. Subjects then performed 3 trials while wearing the

exoskeleton, consisting of two trials with motor control and one trial without motor

control. A picture of a subject aiming the pistol while wearing the exoskeleton can be

42
seen in Figure 32. Experimental scenarios for trials 2-4 were randomized for subjects

who wore the exoskeleton, so as to minimize effects of fatigue, learning, and any

imperceptible changes in recoil force after replacing the CO2 cartridge. Finally,

subjects performed trial 5, consisting of shooting without the exoskeleton. Subjects in

the control group performed all 5 trials using the same timing, but without ever

wearing the exoskeleton. All subjects in all trials stood with their shoulder in the

shoulder cuff while shooting. Upper arm and forearm length was recorded for each

subject, and entered into the Labview algorithm for subjects who wore the

exoskeleton. Subject height, age, sex, and shooting experience were also recorded. As

caffeine may increase arm tremors [33], subjects were asked to abstain from any

sources of caffeine before the experiment. All subjects reported no caffeine intake

within a 5 hour period before the experiment.

Figure 33: Subject performing the aiming task while wearing the exoskeleton

43
 The location of the IR laser point on the target was recorded for all subjects

throughout each trial. Unfiltered joint rotation rates were also recorded while subjects

were wearing the exoskeleton.

4.3.1 Data Analysis

Statistical analysis was performed using SPSS (IBM Corp.). Data were averaged

as described below and Friedmans test was used to check for significant differences

in each metric among trials. If a metric was found to change significantly among trials,
pairwise comparisons were then performed using Wilcoxon signed rank tests.

Significant differences described below refer to results of the Wilcoxon tests. Holm-

Bonferroni correction was used to control family-wise error rate.

4.4 Results
Average subject age and height for the control group and MAXFAS group (who

wore the exoskeleton) can be found in Table 7. Here, shooters refers to the number

of people in each group who indicated they had some shooting experience.

Table 7: Test group demographics

Average Average
Group age height Females Shooters
(years) (cm)
Control 29.85.8 173.611.4 2 1
MAXFAS 29.15.8 176.76.7 3 5

44
 Target center location was calculated for each trial using the four reflective

markers on the target. Absolute distance from target center (radius Rc) was calculated

for each shot. Rc was averaged over the 15 shots in each trial, then averaged across all

subjects in each trial. Average Rc of all MAXFAS subjects in each trial can be found

in Figure 34. Additionally, each subjects average Rc in trials 2-5 was normalized

against their trial 1 Rc (pre-exoskeleton) value. Figure 35 thus presents the average

shooting performance for trials 2-5 normalized to each subjects initial performance.

18.0
MAXFAS group
16.0
14.0
12.0
Average Rc (mm)

10.0
8.0
6.0
4.0
2.0
0.0
Trial 1: Exo w/o Control Control Trial 5:
Pre-exo Control Mode A Mode B Post-exo

Figure 34: Average distance to target center (Rc) with standard error
for MAXFAS subjects

45
 140
MAXFAS group
120
Baseline: Pre-exo

Average Rc % vs. Baseline

100

80

60

40

20

0
Exo w/o Control Control Trial 5:
Control Mode A Mode B Post-exo

Figure 35: Average Rc for MAXFAS subjects normalized to trial 1, with standard error

Average Rc decreased significantly while subjects wore the exoskeleton using

both control mode A and control mode B, compared to trial 1 (pre-exoskeleton).

Compared to trial 1, 14 out of 15 subjects had a lower average Rc using control mode

B. Average Rc was not significantly lower than trial 1 while subjects wore the braces

without motor control. This result indicates that the improvement in shooting

performance is not merely attributable to the braces, but that the motor control plays

an important role in reducing average Rc. Compared to the trial 1 average, the
reduction in average Rc with control mode B was greater and more significant than the

reduction in average Rc with control mode A. Additionally, the average Rc for trial 5

(5 minutes after removing the exoskeleton) was somewhat smaller than that of trial 1.

Compared to trial 1, 12 out of 15 subjects had a lower average Rc in trial 5. However,

this reduction is not statistically significant using the corrected Wilcoxon test

46
described in section 4.3.1 above. Average Rc of all control subjects in trials 1 and 5

can be seen in Figure 36. The shooting performance of the control group did not

improve from trial 1 to trial 5, on average.

18.0
Control group
16.0
14.0
12.0
Average Rc (mm)

10.0
8.0
6.0
4.0
2.0
0.0
Trial 1 Trial 5

Figure 36: Average Rc with standard error, for control group subjects

Calculated joint rotation rates were analyzed 0.5-1 s before each shot. This short

timeframe was chosen to avoid the inclusion of recoil movement for subjects shooting

at 1-2 s intervals. Absolute rotation rates were averaged across MAXFAS subjects

without and with motor control during this pre-shot time frame. These averaged

absolute rotation rates can be seen in Figure 37below. It is apparent that the addition

of motor control significantly reduced arm movement in this critical pre-shot

timeframe. Also, motor control mode B was more effective at reducing arm tremor

47
than control mode A. As such, Figure 38 and Figure 39 below will focus on results for

control mode B.

0.020
Exo w/o control
Average absolute rotation rate (rad/s) 0.018 Control mode A
Control mode B
0.016
0.014
0.012
0.010
0.008
0.006
0.004
0.002
0.000

Figure 37: Average absolute rotation rates for MAXFAS subjects without and with
motor control (standard error shown)

The fast Fourier transform (FFT) was used on the joint rotation rates to

determine the dominant frequency of movement for the same 0.5-1 s pre-shot

timeframe as above. The frequency having the maximum amplitude was then averaged

across subjects in two scenarios: subjects wearing the exoskeleton with no control, and

with control mode B. The results of this FFT analysis can be seen in Figure 38 below.

48
 12.0
Exo w/o control
Control mode B
10.0

Dominant frequency (Hz)

8.0

6.0

4.0

2.0

0.0

Figure 38: FFT analysis of joint speeds without and with motor control mode B
(standard error shown)

Shoulder flexion/extension, adduction/abduction, and internal/external rotation

dominant frequencies did not vary much with and without motor control, all staying

around 5.90.7 Hz. Elbow flexion/extension dominant frequency did change

significantly with the addition of motor control mode B, from 5.50.6 to 3.70.6 Hz.

At 9.90.9 Hz, forearm rotation dominant frequency did not change significantly with

the addition of motor control, but was on average significantly higher than the other

measured joint rotation rate frequencies. All numbers given above and in Figure 38 are

expressed with standard error.

Absolute laser x- and y-distance from target center was also averaged across

MAXFAS subjects in the scenarios mentioned above, as well as trials 1 and 5. These

values were measured in the 0.5-1 s before each shot, and the averages are plotted in

49
Figure 39 below. Average absolute laser x- and y- distance to target center did not

change significantly across the trials. However, absolute x-distance to center was

significantly smaller than absolute y-distance to target center for all MAXFAS

subjects across all trials.

9.0
MAXFAS group X-dir
8.0 Y-dir
Average absolute distance (mm)

7.0
6.0
5.0
4.0
3.0
2.0
1.0
0.0
Trial 1: Exo w/o Control Trial 5:
Pre-exo Control mode B Post-exo

Figure 39: Average absolute laser distance from center for MAXFAS subjects

4.5 Discussion
It is clear that shooting performance improved while subjects wore the

exoskeleton. This improvement cannot be attributed to the use of the passive,

uncontrolled braces alone. Figure 35 illustrates that the addition of motor control,

particularly the lock mode of control mode B, significantly improved shooting

performance compared to the pre-exoskeleton trial as well as wearing the exoskeleton

without motor control. Control mode B resulted in an over 50% reduction in average

50
Rc for some subjects, compared to the pre-exoskeleton trial 1. These results are very

unlikely to be due to learning (acclimation to the pistol or task), since the order of

trials involving wearing the exoskeleton was randomized. Average shooting

performance appeared to be improved 5 minutes after subjects removed the

exoskeleton. However, this improvement over trial 1 was slight, and not quite

statistically significant using the statistical analysis described in section 4.3.1.

However, it should be noted that the control group did not improve on average from

trial 1 to trial 5. Indeed, the standard deviation for the control groups average Rc was

actually worse in trial 5, which may indicate some fatigue effects. Compared to the

MAXFAS groups trial 5 average performance, this result may also indicate that the

MAXFAS device may combat fatigue during aiming. It should be noted however that

the control group only consisted of 5 people, whereas the MAXFAS group consisted

of 15 people.

Figure 37 indicates that a reduction in amplitude of arm shaking coincided with

improvement in shooting performance. This reduction was evident in all joint rotation

rates, but shoulder flexion/extension, elbow flexion/extension, and curiously, shoulder

internal/external rotation were most reduced during the motor control mode B trial. It

is expected that a two-hand grip on the pistol will provide more lateral aim stability

than vertical stability, and thus the motor control will have less of an effect on

shoulder adduction/abduction. Indeed, shoulder adduction/abduction was least reduced

by motor control. However, the large reduction in shoulder internal/external rotation

using motor control is unexpected. Perhaps this DOF contributes more to aim stability

than previously expected.

51
 Figure 39 reinforces the hypothesis that vertical aim stability is weaker than

lateral aim stability using a two-handed pistol grip. Y-distance to target was much

larger than x-distance to target even with motor control, indicating that further

improvements in aim may be achievable by better vertically stabilizing the arm during

aim and shooting. Curiously, average x- and y-distance to target changed little with the

addition of motor control. Considering that the average Rc and arm tremor (as

measured directly by the gyroscopes) were both reduced while wearing the MAXFAS

device, the laser average x-and y-distance to target would be expected to be reduced

during the 0.5-1 s period before each shot. Why this trend is not borne out in the

current results is unclear, but this metric could be investigated in larger future trials in

order to verify the expected trend.

52
 Chapter 5

CONCLUSION

This work detailed the scientific challenges of designing, manufacturing, and

testing an arm exoskeleton for pistol aim stabilization. The device adds very little

weight to the arm, reducing extra arm inertia that cannot be compensated by motors.

MAXFAS owes its light weight to a cable-driven architecture, as well as custom-

fabricated carbon fiber braces that attach to the arm. These braces were carefully

designed and manufactured to provide the necessary stiffness to transmit motor control

to the arm while maintaining a very low weight (section 3.2.2). A tremor-cancelling

algorithm was designed and optimized for this application (section 3.2.4). While the

aforementioned algorithm could not be implemented due to equipment issues, a

simpler control method was tested in experiments using human subjects.

Experiments were designed simulate real shooting, using a pistol with recoil,

realistic weight and trigger action (section 4.2). Experiments were also designed to

eliminate many of the confounding factors involved in aiming and shooting, including
body sway, acclimation to the pistol, caffeine intake, and familiarity with iron sights

(section 4.3). Experiments were conducted on subjects wearing the MAXFAS device,

as well as a small control group who never wore the device. The experiments

demonstrated an improvement in shooting performance while subjects wore the

device, as well as improved shooting 5 minutes after removing the device. Amplitude

of arm shaking, as measured by gyroscopes on the arm, was also reduced while the

MAXFAS devices motors applied control to the arm (section 4.4). These experiments

53
indicated correlation between reduced arm tremors and improved shooting and aiming

performance, which is echoed by previous literature.

The implications of this work extend beyond steadying aim. This device could

be used to reduce tremor for a number of applications. Those suffering from

Parkinsons disease and other debilitating tremors could benefit from active tremor

cancelling. MAXFAS could also be used in training for a number of sports and

hobbies such as billiards, golf, darts, and archery.

5.1.1 Suggestions for Future Work

This project completed the initial experiments demonstrating that a cable-driven

arm exoskeleton could be used to improve pistol shooting performance in a simulated

shooting and aiming task. The control mode was limited by the use of outdated

motors. First and foremost, the original tremor-cancelling design described in section

3.2.4 should be tested on human subjects using the motors originally purchased for

this device. The tremor cancelling algorithm could be combined with the control

method demonstrated in this paper, perhaps resulting in further improvement in

shooting performance for wearers of the device.

Some slight modifications can be made to the braces based on observations

during human subject experiments carried out for this paper. The routing brackets on

the upper arm brace should be redesigned to be further away from the arm radially.

Extreme care had to be taken while attaching the braces to the upper arm to ensure that

these brackets did not pinch the skin of the arm.

The upper arm cables should be routed through the embedded tubes in the upper

arm braces. This will require modification of the Jacobian, and may cause excess

frictional wear on the cables and Delrin tubes. However, the tubes will likely provide

54
an optimal path for routing of the upper arm cables and allow for easier attachment

and removal of the braces on the upper arm.

The cable that terminates on the bottom of the wrist (ulna side) was observed to

occasionally contact the elbow when some subjects bent their elbow. While this

contact is not a great concern as most subjects lock their elbows when aiming, this

cable should have an extra routing point on the forearm brace to avoid such contact

during arm motion from the rest position to the aiming position.

The safety clips attached in line with the load sensors and cables occasionally

detached below their 89 N (20 lb) limit. Large aiming movements thus had to be

performed slowly to avoid detaching the clips. The clips should be replaced by clips

that detach at 133-178 N (30-40 lbs) to allow more natural movement without the fear

of prematurely detaching the safety clips. Parameters within the control algorithm

could also be modified to allow faster movement.

Finally, experiments should be performed on a larger group of trained soldiers

using a real pistol and aiming with the iron sights rather than a laser. Such experiments

should also include a large control group. The experiments should involve longer

periods of shooting while wearing the exoskeleton, as well as evaluation at later than 5

minutes after removing the device to further evaluate any possible training capability

of MAXFAS. Results were good for simulated shooting, but shooting a real pistol

using iron sights to aim is very different from the experiments carried out in this work.

55
 BIBLIOGRAPHY

[1] B. J. Makinson, "Research & Development Prototype for Machine Augmentation

of Human Strength and Endurance," General Electric Company, Schenectady,
NY, 1971.

[2] D. Nosowitz, "Real-Life Iron Man Exoskeleton Gets a Slimmer, More Powerful
Sequel," Popular Science, 2010. [Online]. Available:
[Link]
sequel. [Accessed 28 2 2013].

[3] Lockheed Martin, "HULC," [Online]. Available:

[Link] [Accessed 28 2 2013].

[4] Lockheed Martin, "Lockheed Martins HULC Robotic Exoskeleton Enters

Biomechanical Testing at U.S. Army Natick Soldier Systems Center," 2011.
[Online]. Available: [Link]
releases/2011/june/[Link]. [Accessed 28 2 2013].

[5] Raytheon, "Raytheon unveils lighter, faster, stronger second generation

exoskeleton robotic suit," PR Newswire, 2010. [Online]. Available:
[Link] [Accessed 28 2 2013].

[6] S. H. Kim and e. al., "Robot-assisted modifications of gait in healthy individuals,"

Esperimental Brain Research, vol. 202, no. 4, pp. 809-824, 2010.

[7] S. K. Banala, S. H. Kim, S. K. Agrawal and J. P. Scholz, "Robot Assisted Gait

Training With Active Leg Exoskeleton (ALEX)," IEEE Transactions on Neural
Systems and Rehabilitation Engineering, vol. 17, no. 1, pp. 2-8, 2009.

[8] K. N. Winfree, P. Stegall and S. K. Agrawal, "Design of a Minimally

Constraining, Passively Supported Gait Training Exoskeleton: ALEX II," in IEEE
International Conference on Rehabilitation Robotics, Zurich, Switzerland, 2011.

[9] Y. Mao and S. K. Agrawal, "Transition from Mechanical Arm to Human Arm
with CAREX: a Cable Driven ARm EXoskeleton (CAREX) for Neural

56
 Rehabilitation," in IEEE International Conference on Robotics and Automation,
St. Paul, MN, 2012.

[10] J. Klein and e. al., "Biomimetic Orthosis for the Neurorehabilitation of the Elbow
and Shoulder (BONES)," in 2nd Biennial IEEE/RAS-EMBS International
Conference on Biomedical Robotics and Biomechatronics, Scottsdale, AZ, 2008.

[11] J. C. Perry, J. Rosen and S. Burns, "Upper-Limb Powered Exoskeleton Design,"

IEEE/ASME Transactions on Mechatronics, vol. 12, no. 4, pp. 408-417, 2007.

[12] P. Letier, M. Avraam, S. Veillerette, M. Horodinca, M. D. Bartolomei, A. Schiele

and A. Preumont, "SAM : A 7-DOF Portable Arm Exoskeleton with Local Joint
Control," in IEEE/RSJ International Conference on Intelligent Robots and
Systems, Nice, France, 2008.

[13] S. Balasubramanian, R. Wei, M. Perez, B. Shepard, E. Koeneman, J. Koeneman

and J. He, "RUPERT: An Exoskeleton Robot for Assisting Rehabilitation of Arm
Functions," in Virtual Rehabilitation, Vancouver, Canada, 2008.

[14] G. Wua, F. C. T. v. d. Helm, H. E. J. Veeger, M. Makhsous, P. V. Roy, C.

Anglin, J. Nagels, A. R. Karduna, K. McQuade, X. Wangk, F. W. Wernerl and B.
Buchholz, "ISB recommendation on denitions of joint coordinate systems of
various joints for the reporting of human joint motionPart II: shoulder, elbow,
wrist and hand," Journal of Biomechanics, vol. 38, no. 5, pp. 981-992, 2005.

[15] E. Rocon and J. L. Pons, Exoskeletons in Rehabilitation Robotics: Tremor

Suppression, Berlin: Springer, 2011.

[16] W. Tang, W. Zhang, C. Huang, M. Young and I. Hwang, "Postural tremor and
control of the upper limb in air pistol shooters," Journal of Sports Sciences, vol.
24, no. 14, pp. 1579-1587, 2006.

[17] M. Lakie, "The influence of muscle tremor on shooting performance,"

Experimental Physiology, vol. 95, no. 3, pp. 441-450, 2009.

[18] B. Pellegrini and F. Schena, "Characterization of arm-gun movement during air

pistol aiming phase," Journal of Sports Medicine and Physical Fitness, vol. 45,
no. 4, pp. 467-475, 2005.

[19] J. E. Frberg, C. Karlsson, L. Levi and L. Lidberg, "Circadian Rhythms of

57
 Catecholamine Excretion, Shooting Range Performance and Self-ratings of
Fatigue During Sleep Deprivation," Biological Psychology, vol. 2, no. 3, pp. 175-
188, 1975.

[20] W. J. Tharion, W. R. Santee and R. F. Wallace, "The Influence of Heart Rate,

Rectal Temperature and Arm-Hand Steadiness on Rifle Marksmanship During
and After Field Marching in MOPP 0 and MOPP I," U.S. Army Research
Laboratory, Aberdeen Proving Ground, MD, 1992.

[21] R. S. Goontilleke, E. R. Hoffmann and W. C. Lau, "Pistol Shooting Accuracy as

Dependent on Experience, Eyes Being Opened and Available Viewing Time,"
Applied Ergonomics, vol. 40, no. 3, pp. 500-508, 2009.

[22] K. A. Ball, R. J. Best and T. V. Wrigley, "Body Sway, Aim Point fluctuation and
Performance in Rifle Shooters: Inter- and Intra-individual Analysis," Journal of
Sports Sciences, vol. 21, no. 7, pp. 559-566, 2003.

[23] A. A. Ward, H. H. McFann and J. A. Hammes, "A Comparative Test of Accuracy

and Speed of Fire with the Improved Loop Sling, with the Combat Rifle Sling
and Without a Sling," George Washington University, Alexandria, VA, 1954.

[24] J. Kotovsky and M. J. Rosen, "A Wearable Tremor-suppression Orthosis,"

Journal of Rehabilitation Research and Development, vol. 35, no. 4, pp. 373-387,
1998.

[25] W. D. Hall, "Hand-held gyroscopic device". United States Patent 5058571, 1991.

[26] S. Allen, "Stabilizing device for a gun". United States Patent 5113745, 1992.

[27] Tactical Electronics, "Gyro Stabilized Platform," 2013. [Online]. Available:

[Link]
cameras/88/gyro-stabilized-platform. [Accessed 28 2 2013].

[28] T. W. Chou, Microstructural Design of Fiber Composites, Cambridge, England:

Cambridge University Press, 1992.

[29] K. A. Ball, R. J. Best and T. V. Wrigley, "Body Sway, Aim Point fluctuation and
Performance in Rifle Shooters: Inter- and Intra-individual Analysis," Journal of
Sports Sciences, vol. 21, no. 7, pp. 559-566, 2003.

58
[30] P. Era, N. Konttinen, P. Mehto, P. Saarela and H. Lyytinen, "Postural Stability
and Skilled Performance - A Study on Top-Level and Naive Rifle Shooters,"
Journal of Biomechanics, vol. 29, no. 3, pp. 301-306, 1996.

[31] Y. Mao and S. K. Agrawal, "A Cable Driven Upper Arm Exoskeleton for Upper
Extremity Rehabilitation," in Proceedings of IEEE ICRA, Shanghai, China, 2011.

[32] K. C. Veluvolu, U. X. Tan, W. T. Latt, C. Y. Shee and W. T. Ang, "Bandlimited

Multiple Fourier Linear Combiner for Real-time Tremor Compensation," in
Proceedings of the 29th Annual International Conference of the IEEE EMBS,
Lyon, France, 2007.

59
 Appendix A

INFORMED CONSENT FORM

Study Title: Evaluation of a Wearable Arm Exoskeleton for Aim

Improvement of Healthy Individuals

Principal Investigators: Sunil Agrawal, PhD

Additional Investigators: Dan Baechle

You are invited to participate in a

research study for damping of arm
movements. You are a healthy adult
between the ages of 18 and 40 and will
participate in one experiment session at
the University of Delaware.
The experiment session will take place
in the Mechanical Systems Lab, Spencer
Laboratory at the University of Delaware
and will take no more than 1 hour.
Participation in this study is voluntary and participants will receive no
compensation. You may withdraw from this study at any time without any
consequences. If you have severe respiratory problems such as chronic
obstructive pulmonary disease (COPD), heart disease, loss of sensation,
uncontrolled blood pressure, a seizure disorder, severe arthritis, arm surgery,
or other arm orthopedic conditions that limit your activity level, you should not
participate in this study. Participants who have upper arm or forearm lengths
that are longer or shorter than the exoskeleton will be excluded.

EXPERIMENTAL EVALUATION:
You will aim and dry-fire (no ammunition or projectile) a toy pistol 15
times while standing, taking your time to aim between each trigger pull. The
toy pistol will be equipped with a red laser to assist your aim at a target
approximately 10 feet away. The toy pistol will also be equipped with an
invisible laser, which will be tracked by a special camera system in order to
evaluate your aim. Next, the exoskeleton will be fitted to your arm. The
exoskeleton will apply small forces to your arm while you repeat the
aiming/dry-fire task. You will have a brief rest, and repeat the aiming/dry-fire

60
task again. This will be repeated several times. The exoskeleton will then be
removed, and after a brief rest you will repeat the aiming/dry-fire task once
more without the exoskeleton. The entire test session should take no more
than one hour including setup. The goal is to evaluate any effect on aim that
the exoskeleton might have.

RISKS AND BENEFITS:

There is a risk of eye damage if you point the lasers directly at your eye, or
anyone elses eyes. Even though the laser power is less than 5 mW, (a
standard laser pointer), you should never point the lasers at your eyes or
anyones eyes, as with any laser. You will only point the lasers at the target, or
the ground directly between you and the target. Failure to follow this rule will
result in termination of your involvement in the experiment.

There are slight risks of injury from arm movement with the cable-actuated
exoskeleton attached to the arm. These risks include joint and muscle
soreness and skin irritation. These risks are minimized in several ways. The
motor controller of the device is designed to gently modify the arm movement
by applying only small forces. The motors are set to limit the amount of applied
force. If forces exceed these values, the motors are automatically shut off. In
addition, software defined stops disable the motors when the subject's arm
approaches anatomical limits. Both you and the experimenter will have a
switch that can be pushed to immediately shut down the motors. Nonetheless,
it will be important for you to inform the investigators immediately if you
perceive uncomfortable forces being applied to your arm. If you experience
any discomfort, the motors will be immediately stopped and appropriate
adjustments will be made to reduce the discomfort. Automatic shutoffs and
shutdown switches of the exoskeleton will be tested to perform as designed
before the experiment.

Pressure from the links of the exoskeleton that modify the arm movement
can occur due to faulty alignment, which may result in irritation and redness of
the skin. Aligning the braces or the support cuffs for each participant
individually reduces these risks. However, it is important that you inform the
investigators if you are experiencing any unusual pressure from the braces or
support cuffs while wearing these so that proper adjustments can be made.

In the event that you are injured or experience acute medical emergency
during the study, you will be provided with first aid by the researchers. If you
seek or need additional medical care (including care from paramedics), it will
be at your own expense.

61
 Your participation may provide valuable information that will help in the
design and application of new technology although this may be of no direct
benefit to you. This information may help us to improve aim and training for
soldiers, competitive shooters, or any other sport or task that requires a steady
arm.

STATEMENT OF FINANCIAL INTEREST

Dr. Sunil Agrawal, the Principal Investigator for this study, and other inventors
have applied for a U.S. patent protection for the arm device. If the patent is
granted, Dr. Agrawal, the team of inventors and the University of Delaware
would have a significant financial interest in any commercial development of
the device.

CONFIDENTIALITY:
Personal information and the associated case number will be stored in an
encrypted and password protected file. Data will be associated directly with
the case number alone, not the personal information. Only the researchers will
have access to this information. Your individual evaluation results will not be
shared with anyone outside the laboratory. Neither your name nor any
identifying information will be used in any publication or presentation resulting
from this study, unless you provide us consent to use your photographs and/or
videos for presentation in seminars and technical papers. The data collected
about your aiming performance during these studies will be saved on long-
term storage media such as CDs or DVDs, without information that can
directly identify you. The media will be stored in the investigators laboratory in
Spencer Laboratory. Following completion of this project, the data will be
stored in a secured file cabinet in the investigators laboratory if the
information is deemed to continue to be useful to explore future experimental
questions.

Subjects STATEMENT
I have read this consent form and have discussed the procedure described
above with the investigator(s). I have been given the opportunity to ask
questions, which have been answered to my satisfaction. I understand that
any further questions that I might have will be answered verbally, or if I prefer,
with a written statement.
In the event that I am injured or experience acute medical emergency
during the study, I will be provided with first aid. If I seek additional medical
care, it will be at my own expense.

62
 I have been fully informed of the above-described procedure with its
possible risks and benefits, and I hereby consent to the procedures. I have
received a copy of this consent form.

I give permission to use my pictures/videos for presentation in

seminars and/or technical papers. (Please initial)

Subjects signature Date

.
Subjects Name (please print)

Investigators Signature Date

If you have any questions concerning your rights as a research participant,

you may contact the University of Delaware Human Subjects Review Board
(302) 831-2137. Questions regarding the arm exoskeleton, or anything related
to the study may be addressed to Dr. Sunil Agrawal (302) 831-8049.

63
 Appendix B

IRB APPROVAL LETTER

64
65