Beruflich Dokumente
Kultur Dokumente
47
48
ABSTRACT
This research paper investigates what it takes to design an
affordable autonomous ground vehicle platform from the
ground up. Using a handful of readily available parts, a go-kart
sized vehicle was built and made functional for approximately
$400. Using an independent direct drive rear wheel system, the
vehicle can achieve a fixed axis rotation. Using vision
algorithms loaded onto a Raspberry Pi, the robot is able to
detect a red target and send commands to the Arduino. The
Arduino controls the motion logic and allows the vehicle to
follow the target. While the vision logic run on the Raspberry
Pi is not the main focus of this research paper, developing it
was important to make sure the platform built could achieve the
necessary mobility. The vehicle can also be driven manually
using a hand held controller. The vehicle platform was
completed, along with wiring of components, and it achieved
motion on a flat smooth surface. This platform will allow for
expansion of the project into more complex tasks and can be
upgraded as resources allow.
INTRODUCTION
Recent years have seen an explosion of research into
autonomous vehicles [17]. This project aims to design and
build a platform that uses visual data to navigate its
surroundings. Using simple and affordable microcontroller
boards, a vehicle was created that was able to track and follow
a specified moving target using visual algorithms to supply
motion logic. Other research teams have successfully designed
systems that can, using a stationary camera, track specified
shapes and colors. And still others have mounted multisensory
arrays onto automobiles and achieved controlled environment
autonomous travel such as Googles driverless car [8]. As far as
commercial success a few civilian vehicles have driver assistive
features, but no fully autonomous automobile is available to the
consumer. So far a few states have legalized driverless vehicles
on state roads, and others have legislation in the works.
Through this research, technologies can be developed that can
utilize this new market space and radically change the
transportation industry.
This research paper will focus on creating a robust
mechanical platform that will have the capabilities to simulate a
go-kart sized vehicle. It will be designed so that it has a high
49
enough that it did not flex and cause problems with image
capture.
50
MICROCONTROLLER PROGRAMMING
While versatile and simple to use, microcontrollers like the
Arduino and applications processors like the Raspberry Pi are
limited in their processing speed, especially if parallel
processing threads are required. For this reason, an Arduino
Uno was chosen to send pulse width modulation (PWM)
signals to the motor controllers and the Raspberry Pi for
processing the camera image in order to compartmentalize the
computing tasks. The Pi was chosen for its camera module,
which connects directly to the board with a dedicated camera
port. Both the Pi and camera are protected by placing them in a
commercially available enclosure.
The Pi collects visual information from the camera,
processes it, and sends serial data commands to the Arduino via
USB cable as illustrated in Figure 6. The role of the Arduino is
to read inputs from the controller and Pi, and write appropriate
PWM commands to the motor controllers as described in
Figure 7. The PWM signals to the Talon motor controllers
consist of 1 to 2 ms pulses periodically with a 20 ms period.
Full counter-clockwise motion is achieved with a 1 ms pulse,
where 1.5 ms is neutral and 2 ms full clockwise rotation. The
software limits PWM output between 1.1 and 1.8 ms in order to
operate the tethered vehicle at reasonable speeds.
51
Acquire
image
Process
image
Yes
Kill?
Portable
Battery
No
Raspberry Pi
USB
Controller
Yes
USB
No
Manual?
DO6
Arduino
Talon SR
Write PWM proportional
to potentiometer values
DO9
Yes
Talon SR
12 V
Battery
Value = 1?
No
CIM Motors
CIM Motors
IMAGE PROCESSING
The eventual goal of this project is to equip the
autonomous vehicle with sophisticated image processing
52
REFERENCES
[1] M. A. Zakaria, H. Zamzuri, R. Mamat, and S. A. Mazlan,
A Path Tracking Algorithm Using Future Prediction Control
with Spike Detection for an Autonomous Vehicle Robot,
International Journal of Advanced Robotic Systems, vol. 10, pp.
19, Aug. 2013.
[2] R. Domnguez, J. Alonso, E. Onieva, and C. Gonzlez, A
transferable belief model applied to LIDAR perception for
autonomous
vehicles,
Integrated
Computer-Aided
Engineering, vol. 20, no. 3, pp. 289302, Sep. 2013.
[3] Pan Zhao, Jiajia Chen, Yan Song, Xiang Tao, Tiejuan Xu,
and Tao Mei, Design of a Control System for an Autonomous
53
if(digitalRead(5) == LOW) {
newVal = 0;
while((buffer=Serial.read())>=0)
newVal = buffer;
if(newVal > 0)
val = newVal;
if(val == 1) {
digitalWrite(6,HIGH);
delayMicroseconds(1500);
digitalWrite(6,LOW);
delayMicroseconds(20000-1500);
} else {
if(val < 64) { // Move left
time = map(val,64,0,
baseSpeed,maxSpeed);
digitalWrite(6,HIGH);
digitalWrite(9,HIGH);
delayMicroseconds(1500baseSpeed);
digitalWrite(6,LOW);
delayMicroseconds(baseSpeed +
time);
digitalWrite(9,LOW);
delayMicroseconds(20000
1500 - time);
} else { // Move right
time = map(val, 64, 128, 0,
maxSpeed-baseSpeed);
digitalWrite(6,HIGH);
digitalWrite(9,HIGH
delayMicroseconds(1500-time);
digitalWrite(6,LOW);
delayMicroseconds(time +
baseSpeed);
digitalWrite(9,LOW);
delayMicroseconds(20000
1500 - time);
}
}
} else { // Manual mode
left = map(analogRead(1),0,1023,
1500,1500+maxSpeed);
right = map(analogRead(2),0,1023,
1500,1500+maxSpeed);
digitalWrite(6,HIGH);
digitalWrite(9,HIGH);
delayMicroseconds(1500 - left);
digitalWrite(6,LOW);
delayMicroseconds(left + right);
digitalWrite(9,LOW);
delayMicroseconds(20000 - 1500
right);
}
}
}
54
ABSTRACT
In this study the Surface Response to Excitation method
(SURE) is employed for remote performance monitoring of
manufacturing operations. Basic manufacturing operations
including cutting and drilling are studied and the ability of the
method to evaluate the dimensional accuracy of each process is
investigated.
The SURE method uses piezoelectric elements to monitor
the changes to the condition of structure. One of the piezo
elements excites the high frequency surface guided waves and
the second piezo senses the received waves in another point. In
this study, a laser-scanning vibrometer (LSV) is used to
measure surface vibrations. The advantage of LSV over the
piezo sensor is that the measurements are not limited to a
certain point, and LSV can scan surface vibrations from any
point on the surface of the structure. Each operation was
performed in a certain number of steps and during each step
laser scanning LSV made measurements from the
predetermined scan-points on a grid.
The normalized sums of square differences (NSSDs) of the
frequency spectrums in every step of the operations are
calculated using the SURE algorithm. Results revealed that the
NSSD values increase at each step of operation. Therefore,
dimensional accuracy of the work piece could be related to
NSSD values to monitor the quality of the performance of
process. This behavior was observed not only in scan points
adjacent to the position of operation, but also it was observed to
some extent in all of the scan points. Also the contour map plots
of NSSD values clearly demonstrated the creation and
development of the operation in the correct position.
The proposed method has an advantage over conventional
vision-based process performance monitoring systems. By
choosing the scan points away from the position of operation, it
is possible to monitor the procedure without the need to
interrupt the operation for cleaning the work piece. The
reliability of this technique is examined through observation of
similar results after repeating the experimentation in similar
conditions.
INTRODUCTION
Automation is essential for manufacturing products with
reasonable cost. During automated manufacturing, the quality
of the manufacturing process should be monitored. Many
methods have been developed to evaluate the manufacturing
operation; however, the complexity and the cost of these
systems limit their implementation in industry. The purpose of
this study was to develop a new structural health monitoring
method for condition monitoring of the quality of the
machining operations.
The structural health monitoring (SHM) community has
developed methods for the evaluation of the integrity of
structures, which resulted in improvement of reliability and
decrease in periodical maintenances. More recently, active
structural health monitoring methods have received a lot of
attention. These methods detect structural defects via exciting
the surface of the structure and monitoring its response at
another point. The major trends in the active health monitoring
techniques could be divided into two main categories: the lamb
wave based methods [1,2] or impedance based (EMI) methods
[3,4].
In the surface guided waves (lamb wave) based methods,
structural problems are detected from monitoring the reflections
of high frequency emitted signals. The changes of the
characteristics of the received signals are processed through
time-frequency techniques [ 5]. However, in the electromechanical impedance-based approaches [6], the variation in
the impedance of a piezoelectric element attached to the surface
of the structure is a criterion to detect structural damage.
Experimental studies have shown that the impedance
characteristics of a structure are subjected to considerable
changes when the structure is exposed to defects with different
sizes and types. It has been shown that the impedance-based
methods are more sensitive to loading status of structures
compared to lamb wave based methods [7].
The capabilities of the commercially available automated
manufacturing systems are limited to vision-based systems.
Due to bulkiness of equipment and expensive costs involved in
their processes and some limitations, the commercial
55
7F
8F
9F
10F
1F
12F
[] = [] [] 2
(2)
13F
=
[]
14F
15F
(1)
[]
(3)
56
within a wood frame. This wood frame and beam were fixed to
the experiment table to avoid any movements during the
operations. This way the scan points remained the same at all
steps of the operation. Figure 3 shows the beam, piezo element
and frame.
57
58
Function
Generator
Amplifier
$65
DAQ system
$6.54
Laser
vibrometer
Total
$15
$6.62
$93.16
CONCLUSION
In this study, we developed an automated process
monitoring of manufacturing processes through modification of
an already existing structural health monitoring technique. The
authors believe that once an efficient SHM process is
developed, it could be implemented with minimal expense and
equipment, and it would be possible to inspect several
manufacturing processes. The advantages of such a method are
non - interference with the manufacturing process and
evaluation of tool or product condition with complicated
configuration. Development of a non-contact SHM method for
the automated manufacturing process monitoring and feasibility
of embedded implementation via the SURE based algorithm
will considerably reduce maintenance and operation costs of
tools and quality control of products. Due to simplicity of the
proposed algorithm, the method avoided heavy computational
costs and expensive equipment required in methods like
machine vision based SHM. Due to the non-contact nature of
the laser beam, it was possible to collect data from any point on
the work-piece and locate the best possible points to locate the
piezoelectric elements.
59
60
ABSTRACT
This manuscript develops a visual lane detection algorithm
using the Hough transform to detect strong lines in a road
image as candidates for lane boundaries. The search space in
the Hough transform is reduced by searching for lane
boundaries where they were detected in the previous video
frame. The perspective transform is applied to determine the
position and orientation of candidate lines, which are trusted as
true boundaries if the detected lane width falls within a
specified tolerance of the actual width. Results from a nearly 8minute long video of highway driving in rain indicate that lane
boundaries are correctly identified in 95% of the images.
Detection errors occur primarily during lane changes and poor
lighting when entering underpasses. Including data from
inertial measurements, location on digital maps, and steering
direction would help to reduce or eliminate the instances of
incorrectly detected lane location.
INTRODUCTION
Within the past decade the development of autonomous
automobiles has grown significantly. Semi-autonomous
vehicles are becoming more common in the commercial car
industry. Audi, for example is reinventing their vehicles with
many semi-autonomous features such as adaptive cruise
control, side assist, and lane assist [1]. However, fully
autonomous vehicles are still not available to the public. One
factor of autonomous vehicles that requires refinement and
improvement is the road detection system. The most common
detection method is through a vision-based system due to the
ability to collect data in a nonintrusive manner [2]. Typically a
vision-based system uses a camera to capture footage and a
sensor or laser to scan the preceding road area. Several methods
are used to approach a vision-based road detection system. One
concept analyzes road textures, and segments the road image
based on road and non-road regions [3]. A system using visual
memory stores guiding images to determine a drivable path for
current road images [4]. A detection method for urban areas
was proposed to identify curbs by using a Markov chain to
detect and link curb points [5]. Issues involving these methods
include the lack of precision and reliability when functioning
on a variety of roads and the inability to identify lane
boundaries on multi-lane roads.
X
f=
and y
Z
Y
Z
(1)
(2)
61
=
M cot
=
and B L csc
(3)
HM
Hf
y=
x+
=+
mx b
B
B
(4)
B
L=
b
f +m
2
and = tan 1
b
mf
(5)
Nc
b
w
(7)
P(X,Y,Z)
L=
2b tan ( 1 ) N 1 + 1 R 2 + m 2
c
2
pix
= tan 1
x
X
y
Y
Figure 1: Geometry for projecting a point P in XYZ space
to its location p on the xy camera image plane.
The diagonal angle of view is the angle between lines
from opposite corners of the image to the XYZ origin. This
angle, along with focal length f and image aspect ratio R, define
the width of the image as
2 f tan ( 12 )
1 + 1 R2
(8)
mN c 1 + 1 R 2
w=
2bpix tan ( 12 )
and
(6)
62
m = tan
The and line with the largest Hough value within the final
bounded area is selected as a potential lane boundary line.
Notice how the two sides of the lane stripe are identified by the
detection algorithm, either of which are acceptable, and other
strong lines such as the road edge and guardrail are avoided.
other strong lines in image
successive lines identified by
the detection algorithm
(9)
12 N c
bpix =
sin
tan
(10)
(11)
cos
(pi
xel
s)
1
b
successive lines
identified by the
detection algorithm
boundary of
growing region to
locate maximum
a)
b)
(rad)
b)
Figure 5: Road image a) and corresponding Hough
transform b) used in the algorithm for detecting lane
boundaries.
EVALUATING THE ACCURACY OF DETECTED LINES
The line detection algorithm described above may fail for
any number of reasons such as poor lighting, occluded camera
view, nondescript lane boundary lines, etc. The perspective
transformation is used to identify false positives returned by the
detection algorithm. The and values for the left and right
potential lane boundaries are used with Equations (9)-(11) and
Equation (8) to determine the distance from and angle each line
makes with the camera position. The position of both boundary
lines are trusted if the distance between them is within 10% of
the known lane width for the road. An example of such a
situation appears in Figure 6a where both trusted lane
boundaries are marked with solid lines. Also indicated are the
63
distances from the camera of both detected lines (left and right),
and the difference between them (center) which is a reflection
of the detected lane width. If the distance between detected
lines is not within tolerance, one of the detected lines may still
be trusted according to:
Trust the detected left (or right) line if it was trusted in the
previous frame and its distance L has not changed more
than 1/6 of a lane width and the angle has not changed
more than 5 from the previous frame.
Trust the closer of the two detected lines if the distance
between them is within 10% of the known double lane
width (this covers the case where the detection algorithm
correctly identifies a two lane span instead of one)
a)
b)
64
65
SUMMARY
In total, detection failed continuously in only 8 instances of
more than 1 s during the almost 8 minute period explored. An
actual autonomous vehicle would obviously require reducing
such instances to essentially none. Rules for when to trust the
detected lines can be expanded, and selection of design
parameters fine-tuned, although care should be taken to avoid
micromanaging to deal with specific situations in any given set
of images. For reference, the number of outcome 4) failures is
80, 55, 42, and 35 for lane width tolerances of 5, 10, 15, and
20% of the known width, respectively. Note that reducing
outcome 4) failures from 55 to 35 by increasing the tolerance
results in approximately a doubling of both outcome 3) and 4)
failures. Since the optimal set of design parameters will be
specific to each set of road images, it is recommended to resist
adjusting parameters, while using a simple set of rules for
detecting boundary lines in order to succeed over a wide variety
of different road conditions. However, developing a more
sophisticated local maxima search for analyzing the Hough
transform image is likely to result in improved performance for
all road conditions. Perhaps the most promising improvement
involves fusing image data with other sources of information
such as speed/direction/location from an inertial measurement
unit, steering direction, or road curvature and lane
configuration by locating the vehicle on digital maps using
GPS.
REFERENCES
[1] Audi Driver Assistance Systems. Audi USA News, [online]
January 10, 2012, http://www.audiusanews.com/pressrelease/
2757//driver-assistance-systems (Accessed: August 2014)
[2] Agunbiade, O. Y., Zuva, T., Johnson, A. O., and Zuva, K..,
2013, Enhancement Performance of Road Recognition system
of Autonomous Robots in Shadow Scenario, Signal & Image
Processing : An International Journal., 4(6), pp. 1-12.
[3] Graovac, S., and Goma, A., 2012, Detection of Road
Image Borders Based on Texture Classification, International
Journal of Advanced Robotic Systems., 9, pp. 242-254.
[4] Courbon, J., Mezouar, Y., and Martinet, P., 2009,
Autonomous Navigation of Vehicles from a Visual Memory
Using a Generic Camera Model, IEEE Transactions on
Intelligent Transportation Systems., 10(3), pp. 392-402.
[5] Tan, J., Li, J., An, X., and He, H., 2014, Robust Curb
Detection with Fusion of 3D-lidar and Camera Data,
Sensors., 14, pp. 9046-9073.
[6] Cheng, H., Jeng, B., Tseng, P., and Fan, K., 2006, Lane
Detection with Moving Vehicles in the Traffic Scenes, IEEE
Transactions on Intelligent Transportation Systems., 7(4), pp.
571-581.
66
67
ABSTRACT
This study is fully based on previously published papers
and subsequent research done as a continuation during the last
year [1-3].
Developing an accurate model for an Unmanned Ground
Vehicle (UGV) on stochastic terrain is not an easy task due to
complex phenomena such as stochastic terrain properties. A
stochastic terrain mathematical model was developed and
introduced as a disturbance load in the inverse dynamics-based
control algorithm to introduce the on-line (real time) influence
of the terrain properties on a UGV. A control algorithm was
based on a developed strategy that utilized the inverse
dynamics approach and individual wheel angular velocity
control. This indirectly provides the required torque by
providing the required angular velocity and rolling radius for
each wheel to overcome wheel load torque produced from the
stochastic behavior of the terrain properties.
The objective of this study is to develop an analytical
method for a UGV angular velocity control that is based on
inverse longitudinal dynamics to indirectly provide the required
torque for each wheel to overcome stochastic wheel load torque
produced from the stochastic changes in the terrain properties.
The inverse dynamics control algorithm was integrated
with the DC motor mathematical model. The simulation results
show that the integrated model is distinguished by its
robustness to stochastic external disturbances. In addition, it
shows promising adaptability to disturbances in loading, wheel
load torques and changes in stochastic terrain properties.
INTRODUCTION
A skid-steered wheeled UGV requires a comprehensive
tire-terrain interaction model. Several studies have attempted to
define the nature of circumferential forces acting on a rigid
wheel moving in deformable media and the pressure
distribution along the contact area. However, much less effort
Vladimir V. Vantsevich
Mechanical Engineering Department
University of Alabama at Birmingham
Birmingham, Alabama, USA
has been undertaken to investigate the interaction between a
UGVs deformable tires on deformable soft terrain (sand) [4-7].
Many researchers have investigated the wheel-terrain
interaction. A fundamental aspect of terramechanics was
described in [8], and the wheel traction performance,
considering stress distribution over the contact patch, has been
investigated in [6, 7]. Bekker, Janosi, and Hanamoto introduced
dependences to describe soil-shear diagram and equations for
stresses and forces acting on the tire-terrain interaction [8].
There are numerous sources of uncertainty. It can be due to
spatial variability of soil properties, the changing moisture of
the soil or to measurement errors. The simplest way to
represent uncertainty in vehicle dynamic simulations is to
model it as uncorrelated random variables using a random
sampling method such as the Monte-Carlo method, which is
easy to implement.
The ability to overcome terrain irregularities is mostly
dependent on the drive system of the UGV and its control. A
control algorithm can be used to improve terrainability, which
is related to minimizing wheel slip. Krebs et al., [9] proposed a
torque control method to improve the rover terrainability by
taking into account the whole mechanical structure of the rover.
Another control strategy has been proposed to improve vehicle
handling and stability based on the yaw control strategy [10].
The idea was that the driving torques of the inner and outer
wheels were changed in the same range at the same time during
turns. A method for wheel-ground contact angle measurement
and a traction control strategy minimizing slip in rough terrain
was presented in [11].
Reference [12] investigated motor torque-based active yaw
moment control law, based on which a vehicle stability control
algorithm for a four-wheel-drive electric vehicle that uses a
fuzzy logic control method is proposed.
Hallowell and Ray developed control algorithms for scale
model testing of an all-wheel drive system [13]. The control
algorithms seek to keep all four wheel speeds within a certain
percentage of one another to limit slip, while a torque
68
/+
0.999
/+2
1528.4
1.1
/2
1.04
Degrees
28
10
0.999
0.029
0.951
1.047
1528.4
44.79
1457.6
1601.2
1.1
0.032
1.045
1.154
1.04
0.03
0.960
1.115
28
0.57
26.58
29.27
10
0.001
9.512
10.497
69
2133.59
2132.72
2068.4
2133.15
-5222.21
-5224.63
-5265.29
-5223.69
2070.82
2071.71
1982
2071.54
2107.48
2106.09
2033.08
2106.7
3193.42
3192.12
3123.17
3192.63
5490.3
5480.52
5464.29
5485.39
70
71
(1)
[(+)](+1)/
1
2
180
sin(+)
= 2 sin cos 2 + 2 +
+ 2 tan(45 + )
2
where
540
(4)
= tan + 1 cos 2 ,
= ( tan ) cos2 ,
= cos1 (1
3 (90)
(2)
+ (+1)
3.581 2 (0.0349sin 2 )
(2 )
(3)
()
= ()
+ . () + () 0
()
+ ()
(5)
72
()
were calculated from Eqn. (1), which was used in
()
= 2=1
()
= 2=1 +
+ (6)
2 (2 2 ) = 1 1 + + 0 +
(7)
73
2 = 0 +
2
(8)
1 =
2 0
2
(9)
(10)
(11)
RESULTS
Once applying a stochastic load, the output wheel angular
velocity is no longer able to reach the reference input
(specified) angular velocity, (see Fig. 11). It should be
emphasized here that this computational test is conducted under
the condition where all the wheels are independent from each
other, i.e., there is no simulation of mechanical connection
74
between the wheels and the vehicle frame. In the case when no
control is applied, the front wheels have lower angular
velocities due to higher rolling resistance. The multi-pass effect
has a huge influence on the rolling resistance for the rear tires
as shown before in Fig. 5. Therefore, the rotational speed of the
rear wheels is higher than that of the front wheels.
()
75
CONCLUSION
In this study, an analytical method for an all-wheel drive
off-road UGVs individual angular velocity control is
developed based on inverse longitudinal dynamics. The method
includes a stochastic terrain model and an inverse dynamicsbased control algorithm of a UGV 4x4 drive pneumatic wheels
to overcome stochastic terrain behavior.
A stochastic terrain mathematical model was developed
and used as a disturbance load in the control algorithm to
introduce the on-line (real time) influence of the terrain
conditions on loading of a single wheel of a UGV.
Experimental values for the relationship between the supplied
voltage and the DC motor angular velocity were introduced to
confirm the validity of the DC motor parameters used in this
model.
The outlined method of UGV dynamics control is based on
the inverse dynamics formulation. The provided algorithm for
wheel angular velocity control by varying the wheel torque is
based on the inverse wheel dynamics and the control-byacceleration principle. The model includes the developed
control algorithm to control the angular velocity of four
independently driven wheels by equating the slip of each
wheel.
The inverse dynamics control algorithm was integrated
with the DC motor mathematical model. The integrated model
is distinguished by its robustness to stochastic external
disturbances. It showed promising adaptability to disturbances
in loading, wheel load torques and changes in stochastic terrain
properties.
Future work will be conducted considering variable rolling
radii in driven mode for each wheel individually and
investigating energy losses due to slippage for each wheel. This
will bring about further advances in vehicle performance
control by distributing optimal power to the drive wheels.
REFERENCES
[1] Salama, M., and Vantsevich, V. V., 2013, "Stochastic TireTerrain Interaction for Unmanned Ground Vehicle Modeling,"
Proc. ASME Early Career Technical Conference, ASMEECTC2013.
76
77
78