ISSN 1055- 1425
August 2006
This work was performed as part of the California PATH Program of the
University of California, in cooperation with the State of California Business,
Transportation, and Housing Agency, Department of Transportation, and the
United States Department of Transportation, Federal Highway Administration.
The contents of this report reflect the views of the authors who are responsible
for the facts and the accuracy of the data presented herein. The contents do not
necessarily reflect the official views or policies of the State of California. This
report does not constitute a standard, specification, or regulation.
Final Report for RTA 65A0068
CALIFORNIA PATH PROGRAM
INSTITUTE OF TRANSPORTATION STUDIES
UNIVERSITY OF CALIFORNIA, BERKELEY
Development of the Advanced Rotary Plow
( ARP) for Snow Removal Operations
UCB- ITS- PRR- 2006- 17
California PATH Research Report
Han- Shue Tan, Fanping Bu, Bénédicte Bougler,
Shiang- Lung Koo, David Nelson, Joanne Chang,
Thang Lian
CALIFORNIA PARTNERS FOR ADVANCED TRANSIT AND HIGHWAYS
Development of the Advanced Rotary Plow ( ARP)
for Snow Removal Operations
Final Report
Han- Shue Tan, Fanping Bu, Bénédicte Bougler, Shiang- Lung Koo,
David Nelson, Joanne Chang & Thang Lian
i
Abstract
This final report describes the development and the initial field test of an automated
snowblower, focusing on one of the more difficult snow removal operations: blowing
snow off the freeway along side a guardrail without touching the guardrail. The objective
is to minimize damage to the snowblower, guardrail, and other elements of the
infrastructure by deploying highly accurate and robust automated steering. The automatic
steering is accomplished by following magnets embedded under the roadway. The
development process includes transforming this real- world automated highway winter
maintenance operation into a control problem, modeling snowblower, designing control
algorithms, devising human machine interface, instrumenting a 20- ton snowblower, and
conducting demonstration and field tests. The modified snowblower was equipped with
add- on sensors, actuator, computer and driver interfaces; the test site includes eight
guardrail sections between Kingvale and Soda Springs on the shoulders of Interstate- 80
in the Sierra Mountain region near Donner Summit in California, USA. The ride- along
and test data demonstrated that the prototype system achieved all initial performance
goals, and very positive feedback was received from various stakeholders as well as the
operators who tried it.
Keywords: Snowblower, automation, magnets, sensors
ii
Executive Summary
This project demonstrated an Advanced Rotary Plow ( ARP) with automatic steering
that allows the ARP to follow magnets embedded in the pavement. This project is one of
the first real- world applications derived from PATH/ Caltrans research in the area of
automated vehicle control. A modified snowblower with add- on sensors, actuator,
computer and driver interfaces was developed, and initial field tests were conducted
along the 8 guardrail sections between Kingvale and Soda Springs on the shoulders of
Interstate- 80 near Donner Summit. The results from the field tests and ride- along
demonstrated that the prototype system achieved all critical performance goals and
received very positive feedback from various stakeholders as well as the operators.
Although a full winter field operational tests were not completed due to the end of the
winter season, the successful development as well as the initial field tests suggests
feasibility toward deployment.
Objective
A snowblower is a key component of the snow removal strategy. To achieve effective
removal of the snow built up along the roadside, created by either a single snowplow or a
fleet of snowplows, an operator needs to drive the snowblower at the edge of the road in
order to eliminate the leftover snow “ bleeding” back into the highway. However, such
operations often cause severe damage to the guardrail. An operator generally uses the rear
steering joystick to position the snowblower to an appropriate “ crab” angle and “ tries” to
maintain constant contact between the blower head and the guardrail using his hands
( feeling the pressure), his ears ( hearing the contact sounds), and his eyes ( looking for the
snow poles and obstacles) as he plows forward. “ Riding on the guardrail,” as the
operators commonly put it, creates damage to the rail such as tilting, ripping and tearing
of the guardrail ( Figure 2). This damage leads to frequent repairs and replacements of the
guardrails often in treacherous mountain regions. While guardrails require rehabilitation
throughout all the areas maintained by Caltrans, the frequency of rehabilitation due to
snowblower damage represents a significant cost, thus becoming an opportunity for
excellent return through application of advanced technologies such as precision steering
control. Application of precision steering control has the following potential advantages:
1) Increased operational safety
• The driver knows where the guardrail is without having to " drive by feel".
• It reduces driver fatigue by allowing him to concentrate on the plow and not
where the guardrail is.
• It increases safety in areas that have steep ravines or canyons.
2) Reduced maintenance costs
• It reduces wear and tear on guardrails since the blower no longer needs to touch
the guardrail.
• It reduces wear and tear on the plow by reducing guardrail contact.
Development
In 2000, Caltrans, the Advanced Highway Maintenance and Construction Technology
Center ( AHMCT) at U. C. Davis, and the Partners for Advanced Transit and Highways
iii
( PATH) at U. C. Berkeley started a pooled fund study, “ Development of the Advanced
Rotary Plow ( ARP) for Snow Removal Operations,” with Nevada and Alaska’s DOT as
partners. Caltrans manages the overall project and coordinates resources for field tests
and evaluation. AHMCT conducts feasibility studies on the radar warning system, GPS
application and rotary protection device. PATH is responsible for the design and
development of the ARP automated control system. The ultimate goal of the ARP project
is to develop a prototype automated snowblower that will be used by the Caltrans’
operators and to perform real snow removal operations under harsh winter environments.
In 2002, the project responsibilities were divided more clearly between PATH and
AHMCT for efficiency. PATH is responsible for developing a turn- key lateral control
system that includes the design of HMI for lateral display functions. This report focuses
on the development of the ARP lateral control system.
Various lateral sensing and referencing technologies were investigated for this
application; what was found was that machine vision does not penetrate snow, and that
the GPS system does not provide sufficient reliability under possible multipath and
blockage scenarios. A magnetic- marker- based sensing system was chosen for the initial
implementation primarily because of its high reliability and accuracy ( better than 1 cm)
under all weather conditions. The mountainous highway I- 80 near Donner Summit ( close
to Lake Tahoe) was chosen to be the first field test site. In 2001, magnets were installed
along the eastbound and westbound guardrails of I- 80, at 4 feet apart and 4 feet away
from the guardrail. Binary coding of the magnetic markers was designed ( north pole up
vs. south pole up) to provide information about guardrail characteristics, such as the
shoulder side ( right or left of the blower) and the end of guardrail. Eight sections of the
guardrail were equipped with magnets for the initial feasibility operations with a total
length of 1.46 km ( 0.9 mile) between Soda Springs and Kingvale.
The basic performance requirements for the automated snowblower system requested
by the Caltrans’ Maintenance and formulated by the researchers are as follows:
• “ Tracks” accurately along guardrail ( i. e., lateral error: 2 to 4 inches)
• Supports various snow removal operations
• Survives harsh winter environments ( snow, ice, salt, water, dirt, wind)
• Employ simple operation procedure, tolerating operator mistakes, easy to train
• Create low distraction to operator
• Provide reliable and safe automated operation
The first prototype automated control was a truly “ add- on” system with the following
components added to a conventional Kodiak Northwest single- engine rotary snowplow
with full hydrostatics:
• A computer, together with a data acquisition unit, which processes information and
determines control and guidance actions
• Magnetometers underneath the blower body for measuring the field strength of
magnetic markers installed under the roadway
• A DC motor attached to the steering column with angular sensors as the steering
actuator
• A yaw gyro and speed sensor for measuring vehicle yaw rate and speed
iv
• Human Machine Interface ( HMI) or Driver Vehicle Interface ( DVI) consisting of
the local electronic circuit, a toggle switch, LED displays and an audible unit
The key software components that collectively constitute the necessary intelligence of
the automated system are:
• Reliable signal processing algorithm that provides consistent location estimates
despite large vehicle movement and environmental irregularities
• Smart steering servo that carries out the steering command under highly nonlinear
mechanical characteristics and unpredictable disturbances
• Robust high- gain “ lane- keeping” controller that accurately follows the “ magnets”
under all operational conditions without slope and curvature information
• Adaptive exception controls that cope with any imaginable “ abnormal” scenarios
such as sudden potholes, guardrail touching, actuator saturation, unknown
oscillations, operator mistakes or interventions
• A dependable “ transition” controller that executes “ on- demand” transitions between
automated and manual control under all operational conditions
• A simple and transparent DVI that facilitates clear operator state awareness and
prompts timely and correct responses under both normal and emergency scenarios
• A fault detection and management system that detects system irregularities and
provides a warning while at the same time conducting preventive actions
The effectiveness of the design is evident, for example, in the simple DVI system. It
consists of the following four elements ( see Section 11):
• A transition toggle switch, located under the radio, allowing the operator to switch
the system on and off
• The status LED’s, located underneath the air filter indicator, displaying the system’s
current status
• The guidance LED’s, located underneath the voltmeter, displaying the position of
the tip of the blower head with respect to the guardrail
• An audible unit that produces the following three different sounds:
acknowledgment ( transition to auto steering), end of magnets ( end of guardrail),
and emergency ( take over control now)
The automated operation is simple and straightforward. An operator simply
approaches the guardrail the same way as he always does. The operator can use the
guidance LED’s displays to observe the “ tip location” of the blower head. Once the
blower is within its appropriate crab angle range, the system is ready to transition to
automation, and the GREEN LED will be lit. Once the GREEN status LED is on, the
operator can switch to automated control any time he wishes by pushing down the AUTO
switch. With a soft acknowledgement sound, the BLUE status LED will then be lit,
indicating the blower is now under automated steering control. The operator can resume
manual control by pushing the MANUAL switch or by overriding the steering wheel at
any time. The flashing RED LED, with an emergency sound beeping simultaneously,
signals the driver to take over control immediately.
v
Result
On October 15, 2003, Caltrans conducted an ARP ride- along demonstration to more
than 30 stakeholders from 3 states at Kingvale. The demonstration used a simulated
guardrail and the ARP was tested under various operational scenarios for over 3 hours.
All comments received were positive about the system and performance, especially those
from people who had previous experience working with snow removal equipment.
During March 2005, three sets of initial field tests were successfully conducted along
the guardrails of the Interstate- 80 under real winter operational conditions. The last set of
the tests, on March 22, 2005, was conducted under a heavy winter storm, and the ARP
was blowing accumulated wet snow. Five operators test operated the automated
snowblower.
The initial operator trial and survey, operational test data, as well as the stakeholders’
feedbacks strongly indicated the following:
• The concept of applying automated steering control to snowblower operation is
feasible; the application will improve safety and efficiency of the snow removal
operations.
• The implementation of the current automation technology to the snowblower is
likely to succeed.
• The operators liked the system performance and would accept and use the system.
Recommendation
It is therefore appropriate to start moving toward the deployment of such technology.
Additional R& D effort should address various deployment issues such as reliability, cost,
maintenance, and commercialization. With respect to the continuation of the ARP
technology development, improving safety and flexibility of automation technologies,
investigating operator interface with guidance and control system in real world, as well as
incorporating DGPS to extend the automated operation beyond guardrail sections are all
important possibilities.
vi
Table of Contents
Abstract....................................................................................................................... ........ i
Executive Summary ............................................................................................................ ii
Objective ......................................................................................................................... ii
Development ................................................................................................................... ii
Result .............................................................................................................................. v
Recommendation ............................................................................................................ v
Table of Contents............................................................................................................... vi
List of Figures .................................................................................................................. viii
Acknowledgements........................................................................................................... xii
1. Introduction................................................................................................................... 1
1.1 Background............................................................................................................... 1
1.2 Tasks and Responsibilities........................................................................................ 6
1.3 Accomplishments and Milestones .......................................................................... 12
2. Requirements and Solutions ........................................................................................ 14
2.1 Requirement Formulation ....................................................................................... 14
2.2 Solution Description ............................................................................................... 16
3. Software Architecture and Description........................................................................ 20
3.1 Software Architecture ............................................................................................. 20
3.2 Software Description .............................................................................................. 22
4. Magnetic Lateral Sensing ............................................................................................ 32
4.1 Magnetic Noise Effects........................................................................................... 33
4.2 Tire- induced Magnetic Noise ................................................................................. 34
4.2 Magnetic Sensing Algorithm .................................................................................. 36
4.3 Signal Processing .................................................................................................... 38
5. Magnet Installation ...................................................................................................... 40
5.1 Test Site .................................................................................................................. 40
5.2 Magnet Code Description ....................................................................................... 41
6. Hardware Modifications .............................................................................................. 44
6.1 Hardware Components............................................................................................ 44
6.2 Wiring and Circuit Diagram ................................................................................... 49
6.3 Rear Magnetometer Bar Installation....................................................................... 52
7. Steering Actuator ......................................................................................................... 56
7.1 Actuator System Configuration .............................................................................. 56
7.2 Position Servo Design............................................................................................. 59
8. Actuator Fault Detection.............................................................................................. 65
8.1 Fault Detection Method .......................................................................................... 65
8.2 Experimental Validation ......................................................................................... 70
9. Snowblower Tire Model .............................................................................................. 73
9.1 Impact of Snow Chains to Vehicle Lateral Dynamics............................................ 73
9.2 Impact of Low- Speed Tire Characteristics to Vehicle Steering Dynamics ............ 85
9.3 Improve Bicycle Model Validation ........................................................................ 99
10. Control Design .......................................................................................................... 104
10.1 Snowblower lateral dynamics modeling for control........................................... 104
vii
10.2 Lateral control design ......................................................................................... 107
10.3 Integrated Control ............................................................................................... 111
11. Human Machine Interface/ Driver Vehicle Interface ................................................ 113
11.1 Design Concept................................................................................................... 113
11.2 HMI Components and Location.......................................................................... 114
12. Procedure, Training, and Operator Survey ............................................................... 121
12.1 Operation Procedure ........................................................................................... 121
12.2 Test Procedure and Training............................................................................... 122
12.3 Operator Interview and Human Factor Study Preparation ................................. 124
12.4 Operator Feedback Questionnaire and Preliminary Results ............................... 127
13. Tests and Results....................................................................................................... 134
13.1 Initial Algorithm Test at RFS ............................................................................. 134
13.2 Initial Prototype System Test at RFS.................................................................. 136
13.3 Initial Kingvale Maintenance Yard Tests ........................................................... 138
13.4 Problem- Solving Test at Kingvale Maintenance Yard Tests.............................. 140
13.5 Simulated Guardrail and Operator Feedback at Kingvale .................................. 140
13.6 Stakeholder Demonstration................................................................................. 141
13.7 Final RFS System Calibration ............................................................................ 143
13.8 Initial I- 80 Guardrail Tests: ................................................................................ 149
13.9 Winter Operational Field Tests........................................................................... 151
14. Conclusion and Recommendation ............................................................................ 158
References..................................................................................................................... . 161
viii
List of Figures
Figure 1. 1 Two sections of guardrail damaged by snowblower ........................................ 2
Figure 1. 2 Example of a blower head with scratch mark .................................................. 2
Figure 1. 3 Example of guardrail rehabilitation.................................................................. 3
Figure 1. 4 Rotary Snowblower in Operation near Donner Summit. ................................. 3
Figure 2. 1 Illustration of snowblower crab angle and sensor range ................................ 15
Figure 2. 2 ARP Tasks ...................................................................................................... 16
Figure 2. 3 Automated Snowblower: prototype system components ............................... 18
Figure 2. 4 HMI display: Status lights and operations...................................................... 19
Figure 3. 1 Software architecture relationship.................................................................. 20
Figure 3. 2 Software architecture with respect to database .............................................. 21
Figure 3. 3 Lateral control software.................................................................................. 23
Figure 3. 4 Status DVI/ HMI ............................................................................................. 23
Figure 3. 5 Guidance DVI/ HMI........................................................................................ 24
Figure 3. 6 Lateral source code......................................................................................... 25
Figure 3. 7 Front magnetometer bar configuration........................................................... 26
Figure 3. 8 Rear magnetometer bar configuration ............................................................ 26
Figure 3. 9 Calibration table for snowblower rear center @ 2 cm division calibration.... 27
Figure 3. 10 Calibration table for snowblower ( vertical strength vs lateral position) ...... 27
Figure 3. 11 Calibration table for snowblower ( horizontal strength vs lateral position).. 28
Figure 3. 12 Transition state machine............................................................................... 29
Figure 3. 13 Control state machine ................................................................................... 29
Figure 4. 1 Examples of Snowblower magnetic field noise interference from tire .......... 35
Figure 4. 2 Snowblower tire magnetic noise vs. magnetometer sensor bar locations ...... 35
Figure 4. 3 Snowblower Front Magnetometer Calibration Tables ................................... 37
Figure 4. 4 Snowblower Rear Magnetometer Calibration Table...................................... 37
Figure 4. 5 Rear sensor new calibration & signal processing comparison. ...................... 38
Figure 4. 6 “ Peak- Mapping” Magnetometer Signal Processing Block Diagram.............. 38
Figure 4. 7 Peak detection block diagram......................................................................... 39
Figure 5. 1 Map of the test area ........................................................................................ 40
Figure 5. 2 Magnet Installation......................................................................................... 40
Figure 5. 3 Magnets along guardrail ................................................................................. 41
Figure 5. 4 Illustration of guardrail installed with magnets in I- 80 .................................. 41
Figure 5. 5 Illustration of beginning and ending of a magnet section .............................. 42
Figure 6. 1 Enclosure and components ............................................................................. 45
Figure 6. 2 Steering actuator ( not assembled) .................................................................. 46
Figure 6. 3 Yaw rate sensor and enclosure ....................................................................... 47
Figure 6. 4 Existing dented front magnetometer bar and a spare ..................................... 47
Figure 6. 5 Speed sensor ( left: old; right: improved)........................................................ 47
Figure 6. 6 Interface between snowblower sensors and computer ( 1).............................. 48
Figure 6. 7 Interface between snowblower sensors, commands and computer ( 2) .......... 48
Figure 6. 8 HMI and heart beat timing ............................................................................. 49
Figure 6. 9 Snowblower wiring and circuit diagram ( overall).......................................... 50
ix
Figure 6. 10 Snowblower wiring and circuit diagram ( steering actuator & transition
switches) ........................................................................................................................... 50
Figure 6. 11 Snowblower wiring and circuit diagram ( AT- MIO- 64E- 3 & magnetometers)
............................................................................................................................... ........... 51
Figure 6. 12 Snowblower wiring and circuit diagram ( I/ O boards).................................. 51
Figure 6. 13 Snowblower wiring and circuit diagram ( HMI circuit)................................ 52
Figure 6. 14 Snowblower wiring and circuit diagram ( heart beat detection) ................... 52
Figure 6. 15 Possible rear magnetometer bar location...................................................... 53
Figure 6. 16 Rear sensor bar housing and the magnetometers before final assembly ...... 55
Figure 6. 17 Rear sensor bar after assembly ..................................................................... 55
Figure 7. 1 Block diagram of steering actuator................................................................. 57
Figure 7. 2 Steering actuator installation .......................................................................... 57
Figure 7. 3 Schematic of steering actuator motor assembly ............................................. 58
Figure 7. 4 Current drive loop in ECU ............................................................................ 58
Figure 7. 5 Snowblower steering actuator open loop frequency response ....................... 60
Figure 7. 6 Friction effect ................................................................................................. 60
Figure 7. 7 Closed loop diagram of steering actuator position servo ............................... 61
Figure 7. 8 Step input of a PD controller .......................................................................... 61
Figure 7. 9 Friction effect for small amplitude command ................................................ 62
Figure 7. 10 Step input with friction compensation.......................................................... 62
Figure 7. 11 Overshoot when command input is large ..................................................... 63
Figure 7. 12 Step input with anti- windup ......................................................................... 63
Figure 7. 13 Closed loop response.................................................................................... 64
Figure 8. 1 Block diagram of motor dynamics ................................................................. 65
Figure 8. 2 Drive current in PWM motors: DC plus small ripples ................................... 66
Figure 8. 3 Schematic of fault detection ........................................................................... 66
Figure 8. 4 Large signals: the desired ( solid line) and actual ( dash line) voltages ........... 67
Figure 8. 5 Small signals: the desired ( solid line) and measured ( dash line) voltages ..... 67
Figure 8. 6 Probability distribution of system parameters under fault or no fault............ 69
Figure 8. 7 Components in the steering workbench ......................................................... 70
Figure 8. 8 Motor/ ECU in normal condition .................................................................... 71
Figure 8. 9 Motor/ ECU under fault................................................................................... 71
Figure 9. 1Typical lateral force versus slip angle [ 2] ....................................................... 75
Figure 9. 2 Lateral force versus slip angle ( RAW data) ................................................... 76
Figure 9. 3 Typical linear system identification ............................................................... 77
Figure 9. 4 Proposed identification procedure.................................................................. 77
Figure 9. 5 Non- parametric approach using look- up tables.............................................. 78
Figure 9. 6 Estimated force at each slip angle of the nonlinear relation........................... 80
Figure 9. 7 Flow diagram of the identification procedure ................................................ 80
Figure 9. 8 Experimental results on sand- covered road ( continued) ................................ 83
Figure 9. 9 # of points at each slip angle in configuration 1............................................. 83
Figure 9. 10 Experimental results on dry pavement ......................................................... 84
Figure 9. 11 General system diagram for a typical vehicle .............................................. 87
Figure 9. 12 Top view of ( a) lateral deflection and associated force ( b) yaw deflection and
associated moment ............................................................................................................ 89
Figure 9. 13 Freq. response: steering angle to yaw rate at ( a) V = 0.5m/ s ( b) V = 20m/ s.. 97
x
Figure 9. 14 Freq. response: steering angle to lateral acceleration at ( a) 0.5 m/ s; ( b) 20 m/ s
............................................................................................................................... ........... 98
Figure 9. 15 Freq. response: steering angle to yawrate at ( a) 0 m/ s ( b) 0.45 and 1.6 m/ s 102
Figure 10. 1 Frequency response from front steering angle to yaw rate......................... 106
Figure 10. 2 Block diagram of control loop................................................................... 107
Figure 10. 3 Synthesized and matched 6th order controller frequency responses from
lateral deviation at blower head to steering angle for vr = 1m/ s................................... 110
Figure 10. 4 Synthesized and matched 5th order controller frequency responses from
vehicle yaw angle to steering angle for 1 / r v = m s......................................................... 111
Figure 10. 5 Control algorithm structure ........................................................................ 112
Figure 11. 1 HMI system and components ..................................................................... 114
Figure 11. 2 Location of the transition switch ................................................................ 115
Figure 11. 3 Transition switch actions............................................................................ 115
Figure 11. 4 Emergency button and Auto system switch on the center console ............ 116
Figure 11. 5 Locations of the status display and guidance display................................. 117
Figure 11. 6 General meanings of the status display ...................................................... 117
Figure 11. 7 General meaning of the guidance display .................................................. 119
Figure 11. 8 Audible unit ................................................................................................ 120
Figure 12. 1 Automated Rotary Snow Plow Use Instruction ( 1).................................... 123
Figure 12. 2 Automated Rotary Snow Plow Use Instruction ( 2).................................... 124
Figure 12. 3 Camera views of snowblower operator testing .......................................... 128
Figure 13. 1 Initial test result at RFS test track ( 1st north bound run) ............................ 135
Figure 13. 2 Initial test result at RFS test track ( 2nd south bound run) ........................... 135
Figure 13. 3 Servo performances for the steering actuator............................................. 136
Figure 13. 4 Lateral and transition control...................................................................... 137
Figure 13. 5 HMI control and display results ................................................................. 138
Figure 13. 6 Snowblower arrived at Kingvale Maintenance yard .................................. 139
Figure 13. 7 Data collected during stakeholder demonstration ...................................... 143
Figure 13. 8 Simulated guardrail testing at Richmond Field Station.............................. 145
Figure 13. 9 Curve section on the simulated guardrail testing at RFS ........................... 146
Figure 13. 10 ARP Simulated Guardrail Tests at RFS ( Left Guardrail, 11- 01- 04):
Constant Rear Steering ................................................................................................... 147
Figure 13. 11 ARP Simulated Guardrail Tests at RFS ( Left Guardrail, 11- 01- 04):
Changing Rear Steering .................................................................................................. 148
Figure 13. 12 ARP Simulated Guardrail Tests at RFS ( Right Guardrail, 11- 01- 04): Rear
Steering Change, Wrong Crab Angle, Auto- Ejection, Sharp Speed Changes, Switch
on/ off............................................................................................................................ .. 148
Figure 13. 13 Automatic steering along guardrail on I- 80 on Dec. 2004 ....................... 149
Figure 13. 14 Test run on a fair weather condition ( 2/ 3/ 2005)....................................... 150
Figure 13. 15 Snowblower Tests along Guardrail on I- 80 ( Right Guardrail # 1, 02- 03- 05):
No Snow on the Ground ................................................................................................. 151
Figure 13. 16 Test run on a light snowy day ( 3/ 4/ 2005)................................................. 152
Figure 13. 17 Field test ( blowing wet snow during a winter storm 3/ 22/ 2005).............. 153
Figure 13. 18 Operator using automated steering control along I- 80 guardrail.............. 153
Figure 13. 19 Snowblower Tests along Guardrail on I- 80 ( Right Guardrail # 2, 03- 04- 05):
Light Snow on the Ground.............................................................................................. 154
xi
Figure 13. 20 Snowblower Tests along Guardrail on I- 80 ( Right Guardrail # 1, 03- 22- 05):
Heavy Wet Snow on the Ground .................................................................................... 155
Figure 13. 21 ARP Tests along Guardrail on I- 80 ( Right Guardrail # 2, 03- 22- 05): Heavy
Wet Snow on the Ground................................................................................................ 155
Figure 13. 22 Post- processed lowest snowblower operational speeds ........................... 157
xii
Acknowledgements
This work was sponsored by the California AHMCT Program, in cooperation with
the State of California Business, Transportation and Housing Agency, Department of
Transportation. The contents of this report reflect the views of the authors, who are
responsible for the facts and accuracy of the data presented herein. The contents do not
necessarily reflect the official views or policies of the State of California. This report
does not constitute a standard, specification, or regulation.
The authors thank the California State Department of Transportation for their support.
In addition, the researchers would like to thank the Caltrans Equipment Service Center
and Caltrans Maintenance for their invaluable contributions, particular in the
instrumentation and maintenance of the Advanced Rotary Plow. Finally, the authors
would like to thank the Sierra Snowfighters at the Kingvale Maintenance Center, without
their participation and feedback the project would not have been accomplished.
1
1. Introduction
1.1 Background
A snowblower, a. k. a. a rotary snowplow, is a massive snow removal apparatus that
blows snow high into the air and off the roadway. It is a key component of the snow
removal strategy employed by snow fighters, especially on highways that travel across
mountains. To effectively remove the snow built up along the roadside created by either a
single snowplow or a fleet of snowplows, an operator needs to drive the snowblower on
the edge of the road and often with a very tight tolerance range in order to prevent the
left- over snow from “ bleeding” back into the highway. This method of driving becomes
even more difficult when the snowblower is operated along a guardrail.
In current operation, an operator generally uses the rear steering joystick to position
the large snowblower in the appropriate “ crab” angle ( Figure 14) before he reaches a
section of guardrail. Typically, the rear edge of the vehicle is about 0.1- 0.6 of a meter
further away from the edge of the road or guardrail than that of the front end of the
blower. The operator then drives the huge vehicle body toward the guardrail until the
front side of the blower’s head touches it. He then “ tries” to maintain a somewhat
continuous contact between the blower’s head and the guardrail using his hands ( to feel
the pressure), his ears ( to hear the contact sounds), and his eyes ( to look for snow poles
and obstacles) as he plows forward. Since the blower’s head can weigh up to 6 tons, it
creates a natural oscillation when it hangs in front of the snowplow body. Consequently,
the snowblower continuously “ bounces” into and off the guardrail. “ Riding on the
guardrail,” as the operators commonly term it, creates damage such as tilting, ripping and
tearing of the guardrail that is serious enough to be easily identified by travelers passing
through ( see Figure 1.1 for an example of a section of damaged guardrail).
Such damage leads to frequent repairs and replacements of guardrails in treacherous
mountain regions. At an average cost of approximately $ 100/ meter of guardrail,
including material, equipment and labor, rehabilitation of guardrails is very costly. While
guardrails require rehabilitation throughout all the areas maintained by the Department of
Transportation, the frequency of rehabilitation due to snowblower damage, typically once
every couple of years, represents a significant cost, and thus becomes an opportunity for a
cost effective application of advanced lane- guidance technologies such as precision
steering control. In addition, the practice of guiding by guardrails often causes serious
damage to the approximately $ 300,000 snowblower, increasing the frequency of repair
and replacement. Please refer to Figure 1.2 for an example of a snowblower exhibiting
scratches on the head resulting from the above- mentioned operation; and to Figure 1.3 for
an example of guardrail rehabilitation. A successful application of precision steering
control can reduce; even eliminate contact between the snowblower and guardrail, while
improving the consistency and accuracy of the work performed. Furthermore, this
application will increase operational safety by allowing the operator to concentrate on
“ plowing”, remove the exhausting necessity of “ drive by feeling”, as well as reduce the
operator’s visual fatigue, a major complaint during long- hour winter operations. In
addition, limiting the damage to the guardrail also improve the safety of the traveling
2
vehicles in the event of an emergency situation. The current work, targets at application
in mountainous areas with guardrails, rather than areas without guardrails. However, the
researchers recognize that there could also be significant safety enhancements in areas
with steep ravines or canyons.
Figure 1. 1 Two sections of guardrail damaged by snowblower
Figure 1. 2 Example of a blower head with scratch mark
3
Figure 1. 3 Example of guardrail rehabilitation
Figure 1. 4 Rotary Snowblower in Operation near Donner Summit.
In addition, due to a great number of stalled or abandoned vehicles in mountainous
areas, combined with the buildup of snow to be removed by the blower, there is an
4
increased risk that the blower might collide with these vehicles. Natural objects, such as
large rocks and debris, also present collision hazards. Due to the large mass of the blower
vehicle and the action of the rotary mechanism, such collisions have a high potential for
damage to vehicles, even at the low operating speed ( approximately 1 - 5 MPH, 0.45 –
2.2 m/ s). Impact with foreign objects can also damage or destroy the expensive rotary
mechanism. Thus, inclusion of Collision Warning Systems technology will provide added
safety, as well as reduced liability and repair costs. Figure 1.4 shows a front- discharge
Kodiak rotary snowblower in operation near Donner Summit in California.
The Advanced Highway Maintenance and Construction Technology ( AHMCT)
Research Center at the University of California - Davis ( UCD), in partnership with the
California Partners for Advanced Transit and Highways ( PATH) of the University of
California at Berkeley ( UCB), proposed, in 2000, automation of the driving functions for
a rotary snowblower, including fully automated steering, and possibly automated throttle
and brake. Along with automation of the driving function, the research included
investigation of obstacle detection and collision warning in the context of the snowblower
operation. The proposed combination of automatic vehicle control and obstacle detection
is referred to as the Advanced Rotary Plow, or ARP.
Researchers at the AHMCT Research Center, as well as our research partners at the
California State Department of Transportation ( Caltrans) and PATH, have long
considered the benefits of providing guidance information and vehicle control to enhance
winter maintenance activities. AHMCT and PATH, along with the Western
Transportation Institute ( WTI) of Montana State University, have completed Phase I and
II of their Advanced Snowplow Project ( ASP- I & ASP- II), which provides lateral
guidance and collision warning information to significantly enhance the safety and
efficiency of the snow plowing operation. Based on the success of these projects, there is
an increased interest in applying similar technologies to related winter maintenance
activities, particularly on the rotary snowblower. Since the blower operation requires the
vehicle to operate very close to the guardrail without actually contacting it, tight
tolerances must be achieved. Attempting to drive within these tolerances, even with an
advanced display of all available roadway information similar to that in the Advanced
Snowplow, is not an easy task; therefore driver assistance in this form was not considered
in this project. Full automation can eliminate the high level of operator stress as they
attempt to operate very near the guardrail without impacting it. This endeavor provides a
unique opportunity to clearly demonstrate the near- term benefits of AVCSS and IVI
technologies, including vehicle automation and obstacle detection in a semi- controlled
and geographically limited operating environment. Caltrans has installed infrastructure
elements to support the development and testing of an automated snowblower at Donner
Summit on Interstate 80 during this project period.
This project started with developing a prototype automated snowblower to be used by
the California Department of Transportation operators and to perform real snow removal
operations under harsh winter environments [ 1]. Various lateral sensing and referencing
technologies were available to provide lateral position for the precision steering control.
For example, in [ 2][ 3], video cameras are used to determine the vehicle position for
5
guidance or for control. However, the vision based systems are generally more sensitive
to the environmental factors such as lighting, weather or pavement conditions; and the
machine vision does not penetrate snow and ice that cover the lane markings. GPS is
another way to determine vehicle position for the purposes of guidance or control with a
lower infrastructure cost [ 4][ 5][ 6]. However, current GPS system does not provide
sufficient reliability under possible multipath and blockage scenarios in the mountainous
areas. In order to quickly demonstrate the feasibility of the automatic lane guidance
concept, a magnetic marker- based sensing system [ 7][ 8] was chosen for the initial
implementation primarily because of its high reliability and accuracy ( better than 1 cm)
under all weather conditions [ 9]. The mountainous highway I- 80 near Donner Summit, 30
km from Lake Tahoe, was chosen to be the first field test site.
The overall goal of this project is to relieve the operator of the stressful task of
driving the vehicle in close proximity to the guardrail without physical contact. The
proposed system would also provide obstacle detection and warning to prevent injury or
property damage, thus allowing the operator to perform his duties safely and efficiently.
The application of AVCSS technologies can assist the blower operator in performing
snow removal, while preserving the integrity of the highway guardrail infrastructure, and
avoiding any objects or vehicles located in the path of the snowblower. A subsidiary goal
is the demonstration of the beneficial near- term application of AVCSS and IVI
technologies in the maintenance environment. The project culminates in demonstrations
in the Advanced Winter Maintenance test corridor on California’s Interstate 80.
As proposed, the blower automation functions include automated steering, possibly
automated throttle and brake, short- range forward collision warning, and the required
Human- Machine Interface ( HMI) technology. The proposed project included the
following developments:
a. Vehicle lateral control: The lateral control system includes a sensing system and
control algorithms. The main approach for lane position sensing was developed
by PATH using the embedded magnetic reference marker system for lateral
position indication within the lane. Feasibility of this technology was shown at the
1997 NAHSC and other automated vehicle demonstrations, and its current
application for ASP- I and ASP- II also suggests the technology is robust and well-suited
for this application. Alternative approaches, including magnetic tape, side-fire
radar, etc, were investigated early in the project. The research team has also
implemented multiple technologies on the rotary plow to achieve higher reliability
and robustness, as well as to comparatively evaluate them on a single field-deployed
platform.
As it turned out, the actuation mechanism and vehicle dynamics are significantly
different from any of the vehicle systems concerned in the previous work in the
area of vehicle automation. The blower operating conditions are safety and
operational critical ( i. e., large resistance forces and low tire/ road friction and
cornering forces), and the system is complicated with additional vehicle dynamics
such as tire and snow chain effects. PATH analyzed the control problems and
6
investigated approaches for providing robust and safe control for blower
operation.
b. Collision warning systems would be developed by AHMCT. Due to the short
operating range, low speed, and need to image obstacles through dense layers of
ice and snow, the requirements for the current application differ significantly from
those for ASP- I and ASP- II, For example, imaging through snowbanks with
varying height, density, debris, salt content, and conductivity, is expected to
present significant sensing and algorithmic challenges. AHMCT would
investigate various technologies to determine the best match for the current
application.
c. AHMCT and PATH would jointly develop the required Human- Machine
Interface ( HMI), which includes a display system, as well as necessary interfaces
to allow transition to and from automated control.
In 2002, the project responsibilities were divided more clearly between PATH and
AHMCT for efficiency. PATH is responsible for developing a turn- key lateral control
system that includes design of HMI for lateral display functions, while AHMCT takes
charge in developing an obstacle detection system, the HMI for obstacle display and
other functions, and optionally an alternative lateral controller. This report therefore
focuses on the development of the ARP lateral control system at PATH.
1.2 Tasks and Responsibilities
The tasks in the overall ARP project consist of the following components:
infrastructure and equipment, hardware and software, design and analysis, as well as
report and testing. All components are needed to support the automated functions.
Caltrains leads the efforts in overall project management and coordination,
infrastructure installation and snowblower acquisition, as well as field test support and
performance evaluation. PATH is responsible for automated steering system design and
development. It includes system architecture design, hardware installation, sensor signal
processing, control algorithms, HMI development, operator training and feedback
evaluation, performance review and improvements, as well as support field tests.
AHMCT is responsible for various feasibility studies that include radar based collision
warning system, GPS system, and rotary protection.
In order to provide the automated steering control functions, the first prototype
automated steering system was developed and tested on a conventional Kodiak
Northwest single engine rotary snowplow with full hydrostatics. The system consists of
the following system elements:
- Magnetic markers installed along the highway shoulder, 4 feet from the guardrail at a 4-
foot spacing. Tolerances and binary coding were specified by PATH.
- Arrays of magnetometers installed on appropriate locations of the snowblower.
- Motion sensors, including accelerometers, a yaw gyro, and speed sensors.
7
- Steering actuators.
- Control computer
Detailed discussions of the tasks to develop the system capabilities are provided below.
Task 1 – Review Organization and Operator Needs
The project began with a thorough study of the needs of the DOT and its snowblower
operators. Interviews were conducted with operators, engineers, site managers,
equipment shop personnel, and others within Caltrans. This step ensures that the system
development targets the true needs of the customer, and provides the right capabilities to
enhance the safety, efficiency, and cost- effectiveness of the operation.
Task 2 – Develop System Specifications
Based on the results of Task 1, detailed system requirement specifications were
developed by the research partners, in conjunction with the appropriate parties within
Caltrans and subject to Caltrans review. The development turned out to be an iterative
process since the specifications were often modified based on the trial and field test
feedbacks. The resulting specifications were used to direct the development for the
project.
Task 3 – System Design
Subsystem and overall system design were conducted based on the specifications
obtained from Task 2. The design involved detailed design of each subsystem ( sensing,
actuation, power, HMI hardware, computer, electronics and software), as well as the
architecture of the overall integrated system. A design review process was employed to
ensure the incorporation of lessons learned from prior projects as well as feedbacks from
the field tests.
Task 4 – Vehicle Automation
The development of vehicle automation consists of three basic sub- tasks: sensor
development, actuator integration and controller design.
• Sensors
A number of sensing devices are installed on the snowblower in order to facilitate
automated control operation. These sensors include:
- Magnetic sensors: In this project, magnetic sensors are used as the primary location
sensors for snowblower steering control based on its proven accuracy and reliability
under the winter operation environment. Two arrays of magnetometers are installed under
the snowblower, which detects the magnetic field from the magnetic markers embedded
in the roadway. Through a signal processing algorithm, the lateral position of the vehicle
and information encoded in the magnetic markers are obtained from the magnetic field.
The number and the locations of magnetometers to install is determined based on the
requirement specifications as well as on the limitations that imposed by the configuration
of the equipment ( snowblower) used.
8
- Motion sensors: Motion sensors are installed on the snowblower for measuring vehicle
accelerations and rate. Two accelerometers ( for both lateral and longitudinal
accelerations) and an angular yaw rate sensor are installed. However, the yaw rate sensor
turns out to be the only motion sensor that is used for the lateral control.
- Steering angle sensor: In the prototype system, the steering angle measurements are
obtained through measuring the position of the steering actuator. A position sensor,
consisting of an encoder and a potentiometer, is installed as part of the steering actuator
design in order to provide accurate measurements of the steering angle.
- Vehicle speed: Accurate vehicle speed measurements are crucial for the control of low-speed
operation. Electronic circuitry that can be installed to interface the existing speed
measurement mechanism was attempted; however, the results did not achieved required
resolution and accuracy for the speed measurements. Investigation was conducted and
new speed sensor was installed in the drive shaft to provide speed sensing on the plow.
Field tests have also indicated that a speed sensor that can operate at speed at least as low
as 0.3 m/ s will be required.
- Brake pressure: A brake pressure sensor could be installed if it is required. However, It
has been determined not to pursuit longitudinal control in this phase the project.
- GPS: GPS can be used in conjunction with the motion sensors to provide position
measurements to supplement the magnetic– marker- based sensing system. AHMCT
explored the feasibility of using such sensor. An automated snowblower control based on
sensor fusion of position measurements of magnetic sensing and integrated GPS/ INS is
left for future study.
• Steering actuator
The experience and knowledge of developing automated steering vehicles has shown that
the steering actuator design is an integral part of the development of any automated
steering system. Moreover, the system analysis suggests that the practical limitations of
the steering actuator have an adverse effect on the lane- keeping performance, especially
when a look- down lateral sensing system, such as the magnetic- marker- based reference
system, is employed. The bandwidth and phase characteristics of the actuator have a
significant impact on the steering control design.
The configuration of the front steering actuator consists of the following components:
( 1) An add- on DC motor with gear interface on the steering column that drives the
existing hydraulic system;
( 2) Encoders and a potentiometer installed on the motor shaft and coupled with the
steering shaft, which measure the steering positions for the steering servo loop;
( 3) A computer that determines the steering command to the steering motor.
A torque sensor could be installed on the steering shaft for additional flexibility in the
HMI design. Due to the complexity, the sensor was not included during the design phase.
Should it be needed, the function of the torque sensor can also be approximated by the
9
current command of the steering actuator, since the motor is a current- mode commend
DC- motor.
Furthermore, although an array of LVDT could be installed on the rear steering
mechanism to measure or detect the rear wheel steering position or state for the rear
steering servo loop. The associated closed- loop rear steering control could be achieved
using additional servo valves. However, they are not installed in the development phase
of this project due to the reasons described below.
The possible steering servo loop designs are: ( 1) front wheel closed- loop control without
the direct measured knowledge of the rear wheel location, ( 2) front wheel closed- loop
control with knowledge of the locked rear wheel location, ( 3) front wheel closed- loop
control with limited rear wheel movement control, i. e. use of set- point positions for left,
right, and center steer of rear wheels, and ( 4) front and rear wheel complete closed- loop
controls. Option ( 1) involves the most challenging controller design, and it is also the
most preferred method, if it is achievable. Options ( 2), ( 3), and ( 4) require certain forms
of rear steering position measurements; Options ( 3) and ( 4) each requires a different level
of rear steering control capability.
Several design constraints favor not to install unnecessary new sensors or rear- steering
actuators in the snowblower. Since the current manual rear steering is an “ open- loop”
steer- by- wire hydraulic system, the survivability of the exposed rear steering sensors or
actuator is low. In addition, the ability of an operator to frequently adjust the rear- steering
angle for different speeds and load conditions turns out to be a very crucial factor for
operation. Therefore, the design focuses on Option ( 1).
To satisfy the performance requirements, iterations of the hardware and algorithm design
are performed in the development of the steering actuator. The design procedure
includes: model development and validation, control configuration design, data analysis,
linear compensator design, small signal and friction analysis, hydraulic evaluation,
nonlinear compensator design, benchmark and vehicle performance validation, user
interface development, software interface development, and fault management
development.
• Controller Design
The controller design involves the following processes: system requirement definition,
control configuration determination, snowblower model development and validation,
control algorithm design, control software development, fault management development
and vehicle testing. The control configuration are determined by the system requirement,
steering actuator configuration, snowblower dynamics, and HMI method.
The controller needs to satisfy all the system requirements under various uncertainties.
The system requirements include tracking accuracy, ride comfort, and easy driver
interaction; while the uncertainties include road adhesion variations, preview errors,
marker installation misalignments, actuator limitations, blower load changes, speed
10
variations, vehicle dynamic changes, suspension modes, and all reasonable sensor and
vehicle noise.
The control algorithm consists of the following elements:
( 1) A steering servo adapter algorithm that coordinates controls between front and rear
wheels;
( 2) A high- gain robust lane- keeping algorithm that guarantees small tracking error along
the magnet line;
( 3) A transitional algorithm that switches between manual and automated steering.
( 4) An adaptive lane- catching algorithm that provides smooth trajectories from manual
steering to automated steering;
( 5) A state machine that coordinates the above schemes based on the sensor signals,
available road information, and maneuver demands.
Task 5 – Collision Warning
The task involving collision warning will not be discussed in this report; however a brief
description is included here for completeness. The ARP proposal includes a forward
Collision Warning System ( CWS), which detects vehicles and other obstacles buried
under the large snow build- up to be removed by the blower. While AHMCT and PATH
have significant combined experiences in the application of CWS in a variety of
situations, e. g., the snow environment for the ASP, the unique operating conditions of the
rotary blower require innovative developments. Specific issues include low rotary blower
forward speed, reduced sensor range, increased sensor accuracy and resolution, large
snow build- up, and close proximity to fixed infrastructure, i. e. the guardrail. Furthermore,
Imaging through snow banks with varying height, density, debris, salt content, and
conductivity, also presents significant sensing and algorithmic challenges. These issues
place restrictions on the CWS hardware and algorithms. AHMCT is responsible for the
investigation of various sensing technologies, such as FMCW Doppler radar and surface
penetrating radar, to determine the best match for the current application. Additional
effort needs to be dedicated to developing the signal processing algorithms appropriate
for the identified conditions and the selected sensing technology.
In addition to the sensing aspect of the CWS, the nature of the warning to the operator
also needs consideration. Possible approaches are visual, audible, and tactile indication.
Audible indication is expected to be difficult, given the noisy operating environment in
the snowblower cab. In principle visual and/ or tactile warning is preferred.
Task 6 – Human Machine Interface
Although the snowblower is automated, it is necessary to provide information to the
operators so that they can supervise, make transition into or take over the automated
system. Lateral position, speed, curve information, and system status are the candidate
information available to the operator for the purpose of monitoring system performance
and integrity. Without a proper HMI, the efficiency and safety of the system are at risk.
This is especially true during whiteout and deep snow conditions when the operator has
difficulty observing a system fault.
11
The ultimate design criterion is “ simple and clear”. Of particular concern are workload
and driver/ machine control. Operator overload will result in high stress, while underload
may lead to driver inattention. The process for an operator to override or adjust automatic
control must be safe and intuitive. In addition, the frequency of transitions between
automated and manual control also influences the design requirements. The HMI design
started with conducting interviews with operators and other DOT personnel, as well as
analyzing the test site configuration and operations.
The HMI development began with an integrated approach that considers an “ operator” as
a part of the system, especially during transitions. Upon each modification, the
snowblower operators were invited to evaluate the modified HMI. Their comments, as
well as the observations of the research staff, were used to produce a HMI specific to
snowblower operation.
Following implementation, data are collected to examine the operator- display- activation
interaction. This data assists additional improvements and overall system validation.
Task 7 – Vehicle Integration
A snowblower provided by Caltrans was instrumented with a power system, sensors, the
steering actuator, and computers. Work of integration typically occurred during the
summer, so that the blower would be ready for the winter testing when the infrastructure
is available. However, due to various unplanned blower hardware maintenance issues
during the project period, the snowblower lost several opportunities for testing on I- 80
under snow removal operations.
The control computer, including sensor and actuator I/ O, software modules, and system
communications, was developed using standard industrial PC hardware and the QNX
Real- Time Operating System ( RTOS). This system architecture and software structure
has been used as the basis of the AVCSS research and development at PATH for many
years. Hardware and software improvements specific to this project, such as deduced
sensor spacing, low- speed sensing and control, and HMI control circuit, are the results of
the design and implementation iterations.
Task 8 – Infrastructure Installation
Caltrans installed magnetic markers in the highway shoulder at 4 feet away from the
guardrail ( half of the vehicle width). The distance between markers is also 4 feet ( 1.2m).
PATH designed, provided and double- checked tolerance specifications for both the
lateral and longitudinal placement of the magnets. The position of markers was carefully
established though survey to ensure smoothness, and binary coding was encoded in the
magnetic markers to provide information needed for control. Magnets were extended
beyond the length of the guardrail ( 15m) for an appropriate distance in each direction, in
order to provide transitions between automated and manual operation. Eight sections of
the guardrail were equipped with magnets for the initial feasibility operations with a total
length of 1.46 km ( 0.9 mile) between Soda Springs and Kingvale.
Task 9 – Testing and Demonstration
12
Various open- loop experiments have been conducted to verify the dynamic behavior of
the snowblower. These tests were designed to determine the steering responses, tire- road
interaction ( particularly the corning stiffness, the effects of tire chains), speed control,
and braking responses. Test data collected using on- board sensors was used to verify the
dynamic model, which in turn is used for the design of the controller.
The ARP system was finally tested in the Advanced Winter Maintenance Testbed around
the Donner area. The testbed was originally developed by Caltrans for the ASP project.
The various subsystems developed for the ARP were tested individually and as a system.
Subsystem tests began in laboratory development, continued at the test track in
Richmond Field Station ( FRS), and ended with tests in the snow environment at the
Kingvale test track. The overall system was tested in the deployment environment along
I- 80. Quantitative and qualitative measures are used. Quantitative measures include
control accuracy, and transition speed. Qualitative measures are obtained by interviews
with operators that include impressions of ease- of- use, HMI design, operator comfort in
automated operation and during transitions to and from automated mode.
Task 10 – Data Analysis and Reporting
The members of the research team provided quarterly reports at the end of each fiscal
quarter. These reports describe tasks initiated and/ or completed, percentage progress to
date, funds and percentage expended for current fiscal year as well as the overall project,
detailed description of tasks for the quarter, and anticipated work for the following
quarter. In addition, any problem areas related to the previous or subsequent quarter are
included as part of the quarterly reports. This final report presents details of the system
development process, background technical discussion, details of the system design
hardware and software, and results of quantitative and qualitative tests.
1.3 Accomplishments and Milestones
The major tasks related to the Automated Snowblower that have been accomplished
are listed as follows:
1. Completed system design ( 2002)
2. Installed, tested and refined sensors and signal processing algorithms ( 2002- 2004)
3. Installed, developed and tested the steering actuator, including its hardware,
software and servo algorithm ( 2002- 2003)
4. Developed, tested and refined automated control algorithms ( 2002- 2004)
5. Developed, installed, tested, and refined operator interface components including
sounds, display and switches ( 2003- 2004)
6. Successfully conducted operator training and interviews ( 2003- 2004)
7. Successfully demonstrated the first prototype system to stakeholders ( California,
Nevada and Alaska) at Kingvale yard with simulated guardrails ( 10/ 17/ 2003)
8. Tested and refined the second prototype “ turn- key” system along guardrails on I-
80 under no- snow conditions ( 12/ 2004- 3/ 2005)
9. Successfully conducted the first operational trial along guardrails on I- 80 under
heavy snow condition ( 3/ 22/ 2005)
13
The major milestones that have been reached are:
6/ 25/ 02: Snowblower reached Richmond Field Station ready for system installation
10/ 30/ 02: First prototype hardware and software drivers installed
12/ 17/ 02: First prototype control and operator interface system installation ready;
snowblower left Richmond Field Station
3/ 6/ 03: Snowblower arrived at Kingvale; initial system calibration started at
Kingvale yard
4/ 3/ 03: First prototype system ready; performance requirements are achieved at
Kingvale yard
4/ 29/ 03: Successfully conducted first operator trials at Kingvale yard with
simulated guardrail; survey results showed very positive responses from
the operators
10/ 17/ 03: Successfully conducted automated snowblower demonstration at Kingvale
yard for various stakeholders ( California, Nevada and Alaska)
3/ 22/ 05: Successfully conducted initial field tests along I- 80 guardrails under
winter operational conditions.
Various unplanned blower hardware maintenance issues affected opportunities for
winter field tests on I- 80. The following is a time line of these issues:
• Head gasket repair: 2001- 2002
• Blower head removal and modification: from after winter 2001/ 2002 to 11/ 02
• Hydraulic circuit breakdown: 9/ 02 – 10/ 02
• Warranty repairs: 12/ 02 – 2/ 03, 11/ 03 – 12/ 03
14
2. Requirements and Solutions
This section provides formulation of the snowblower steering control problem, as
well as a brief description of the solution.
2.1 Requirement Formulation
The initial “ performance requirement” from the Maintenance department seemed to
suggest that it is a difficult but straightforward project: controlling a snowblower at a
distance between 2 and 4 inches from the guardrail. An examination of the project
objectives revealed that the success of the project would stem on the positive responses of
the following questions: ( 1) Does the system reduce or eliminate guardrail damage
caused by the blower? ( 2) Does the system effectively support snow removal operations?
And ( 3) does the operator like the system and would the operator use the system? As a
result, the initial requirements for the automated snowblower system were defined as
follows:
• “ Track” accurately along guardrail ( 2 to 4 inches)
• Support various snow removal operations
• Survive harsh winter environments ( snow, ice, salt, water, dirt, wind)
• Simple operation procedure, tolerate operator mistake, easy to train
• Low operator distraction
• Reliable and safe automated operation
During the first winter’s ride- along observation in a snowblower, the researchers soon
realized that accurately controlling a 6- ton oscillatory blower head on a 20- ton vehicle
along the highway shoulder dotted with potholes while pushing and blowing snow and
ice was not easy! Let alone that the driver, from time to time, has to adjust the rear
steering angle to compensate for various cutting load and road curvature, move the head
( so- call “ box”) position and tilt angle to account for different road slope, inclination and
resistant force, as well as change the speed from stop to go to react for various road and
snow conditions. The control system must allow the operator to engage automation at
ease and to switch off any time he wants. The system also needs to survive both the
operator’s intervention, either intentionally or unintentionally; and the environmental
disturbances such as hitting a guardrail and running into an ice patch. Furthermore,
during the early literature survey stage, we also found out that little research work exists
in the area of snow chain effects as well as the under- damped low- speed heavy vehicle
model. Nevertheless, the project goals dictated that all obstacles needed to be overcome.
The system requirements were then modified to include the following additional specific
items:
• Automatically compensate operator’s rear steering action
• Robust against various blower head positions and the resultant front tire loading
conditions
• Robust against rough and uneven road surface conditions including potholes
• Provide sufficient control at any operational speeds including stop and go
• Allow on- demand operator transitions and interventions
15
• Not touching the guardrail and survive the consequence should it occur
In addition, there are several specific requirements that are the results from the
specific implementation with respect to either the magnetic sensing system or from the
existing steering hydraulic assist and DC- motor actuator:
• Less than 2 ft of effective operation sensor range ( the effective range starts from
magnetic sensor first “ sees” the magnets till when the blower head “ touches” the
guardrail – Fig. 2.1 for illustration)
• Nonlinear and under- powered steering hydraulic assist ( main nonlinearities:
under- power assist at very low speeds, insufficient assist when stop, large
variation in hydraulic assist power when other hydraulic components requires
power ( full hydrostatics)
Figure 2. 1 Illustration of snowblower crab angle and sensor range
Since the snowblower used for this study still performs normal winter snow removal
operations, several design constraints were imposed based on the considerations in safety,
operation and maintenance. First of all, the installation and application of any
components to the snowblower, especially the steering actuator, should not affect normal
driver manual operations, nor should it imperil or degrade performance of any existing
vehicle components. Second, unless a rear steering sensor can survive the harsh winter
exposure, it is not recommended. The rear wheel is actuated by open- loop hydraulic
valves. Driver controls the rear steering using a “ joy- stick” type controller with 7 LED’s,
each connecting to a contact switch, indicating the location of the rear wheel angle. Since
16
getting precise position reading by measuring flow rates of transmission fluid and
installing sensors on the linkages next to rear wheel are both difficult. High- precision
position sensors, such as rotary position encoder and linear transducer, are not
encouraged to mount on the rear steering mechanism that can be potentially encapsulated
in an ice ball. Finally, since the operator cuts in and out of the guardrail operation, the
only reliable information that is available through the magnetic pattern is the indicators
for left/ right shoulder, and for the approaching of the “ end of magnets.” Typical preview
road information such as curvature, super- elevation will not be available to the controller.
Many critical tasks were performed during the development of the automated
snowblower under the above limitations and requirements ( see Fig. 2.2 for a list of ARP
development tasks). It started with the above problem and requirement formulation;
followed by modeling and basic controller design. A system configuration were then
designed and rehashed based on the analysis results and the operational observations.
Hardware and software were developed that included sensor installation and signal
processing coding, actuator installation and servo controller design, computer setup and
circuit implementation. Human machine interface ( HMI) was then developed and
instrumented based on operational analysis, operator feedback, and field tests. Safety-critical
issues were designed and reviewed that included robust control, fault detection,
failure mode analysis, warning system and redundancy. Finally, various tests were
conducted to evaluate and refine the system design.
Figure 2. 2 ARP Tasks
2.2 Solution Description
To date, no automated precision steering control system has been designed to operate
under such harsh winter conditions subject to extreme external disturbances. And not
17
only that, but designed also with extensive un- modeled dynamics, under severely “ non-ideal”
actuating limitations, and requiring transparent “ interfacing” with an average
operator performing multiple tasks. As the project proceeded, especially under the short
time period the snowblower was available to the design team, the researchers soon
realized that mathematic models often do not portray certain important real
characteristics accurately. The design of this automated system is a combination as well
as iterations of “ design methodology” and “ design synthesis”. It requires continuously
evolving “ solutions” to all of the following elements: problem definition, requirement
specification, system configuration, hardware installation, software architecture, control
algorithms, human machine interface, fault detection and management, and testing and
evaluation.
The first prototype automated control was a truly “ add- on” system with the following
components, as shown in Figure 2.3, added to a conventional Kodiak Northwest single
engine rotary snowplow with full hydrostatics. A computer with a data acquisition unit
that processes information and determines control and guidance actions is the “ brain” of
the system. The lateral positioning system consists of two sets of magnetometers, one
underneath the front axle, and the other one mounted in between the front and rear
wheels, measuring the field strength of magnetic markers installed under the roadway. A
DC motor attached to the steering column with angular sensors is the steering actuator. A
yaw gyro and an axle speed sensor measuring vehicle yaw rate and speed are used as the
supplementary sensors during extremely low speed operations. Finally, a Human
Machine Interface ( HMI) unit ( or Driver Vehicle Interface ( DVI) unit), consisting of the
local electronic circuit, a toggle switch, LED displays and an audible device, interfaces
with the operator with essential information and commands for automation.
The key software components that collectively constitute the necessary intelligence of
the automated system are:
• Reliable signal processing algorithm that provides consistent location estimates
despite large vehicle movements and enormous environmental irregularities
• Smart steering servo that firmly carries out the steering command under highly
nonlinear mechanical characteristics and unpredictable disturbances
• Robust high- gain “ lane- keeping” controller that accurately follows the “ magnets”
under all operational conditions even without slope and curvature information
• Adaptive exception controls that cope with any imaginable “ abnormal” scenarios
such as sudden potholes, guardrail touching, actuator saturation, unknown limit
cycle oscillations, operator mistakes or interventions
• A dependable “ transition” controller that executes “ on- demand” transitions between
automated and manual control under all operational conditions
• A simple and transparent HMI ( DVI) that facilitates clear operator state awareness
and prompts timely and correct responses under both normal and emergency
scenarios
• A fault detection and management system that detects system irregularities and
provides a warning while at the same time conducting preventive actions
18
Figure 2. 3 Automated Snowblower: prototype system components
The effectiveness of the design is evident, for example, in the HMI ( DVI) system. It
consists of the following four elements:
• A transition toggle switch, located under the radio, allowing the operator to switch
the system on and off
• The status LED’s, located underneath the air filter indicator, displaying the
system’s current status ( Figure 2.4)
• The guidance LED’s, located underneath the voltmeter, displaying the position of
the tip of the blower head with respect to the guardrail
• An audible unit that produces the following three different sounds:
acknowledgment ( transition to auto steering), end of magnets ( approaching end of
guardrail), and emergency ( take over control now)
• An emergency button, located on the console just right to the operator right hand,
allowing the driver to “ kill” the steering actuator at any time.
The core of this HMI ( DVI) is four status LED’s: GREEN when the system is ready
for transition; WHITE when it is under driver’s control; BLUE when it is automated; and
RED when there’s a problem. It identifies the four key pieces of information for
automation: system on or off, ready for transition or not, current state of automation, and
fault or not. The operator simply approaches the guardrail the same way as he always
does. A separate supportive guidance LED’s displays the current “ tip location” of the
blower head. Once the blower is within its appropriate crab angle range, the system is
ready to transition to automation, and the GREEN LED will be lit. Once the GREEN
status LED is on, the driver can switch to automated control any time he wishes by
pushing down the AUTO switch. With a soft acknowledgement sound, the BLUE status
19
LED will then be lit, indicating the blower is now under automated steering control. The
operator can resume manual control by pushing the MANUAL switch or by overriding
the steering wheel at any time. The flashing RED LED, with an emergency sound
beeping simultaneously, signals the driver to take over control immediately.
Figure 2. 4 HMI display: Status lights and operations
20
3. Software Architecture and Description
3.1 Software Architecture
The software architecture consists of a set of processes running on the control
computer ( a six slot industrial computer) and communicating through the
Publish/ Subscribe database. All of the software is written in C and runs on the QNX
real- time operating system. The functions of the real- time software are to process the
signals obtained from the various sensors, give control commands to the steering actuator
and send display parameters to the Driver Vehicle Interface. To achieve those functions,
the real- time software is structured as ( as shown in Figure 3.1):
• device drivers
• database manager
• steering controller
Figure 3. 1 Software architecture relationship
The computer is decked with four cards:
• PC- TIO- 10: for timers inputs and digital I/ O
• AT- AO- 6: for analog outputs and digital I/ O
• AT- MIO- 64E- 3: for timer inputs, analog inputs and outputs, and digital I/ O
• EIC- 325: encoder interface card from Industrial Microcomputers Systems
21
Figure 3. 2 Software architecture with respect to database
Ten processes are running together on the control computer. The database runs at
priority 25, the highest among all the processes. Device drivers run at priority 19; since
hardware interrupt handlers are part of the device drivers, they inherit their priority. The
lateral control process runs at priority 18 because it needs to read the magnetometer
channels every 2 msec. The steering actuator inner loop is running at priority 17 simply
because the processing speed of computer processor used is too slow ( otherwise the
steering controller becomes unstable because of the time delay). The vehicle input/ output
process runs at priority 15. All other processes run at priority 10, which is the default.
Regarding scheduling, the static- priority scheduling policy of QNX is used. Each process
is assigned a priority, from 0 ( lowest) to 31 ( highest). At any time, a highest- priority
process is chosen to run among the ready ( i. e. non blocked) processes
Below is a table of the processes as well as their output variables written to the
database and the process priorities.
Table 3.1 Processes, output variables and priorities
Process name Process description Output variables Process
priority
db_ slv Database manager - 25
atme_ rse Interface to AT- MIO- 64E-
3 card - 19
22
Process name Process description Output variables Process
priority
EIC_ 325 Interface to EIC- 325
encoder interface card - 19
pctio10 Interface to PC- TIO- 10
card - 19
ataocard Interface to AT- AO- 6
card - 19
veh_ iobl Vehicle Input Output
long_ input
lat_ input_ front_ ma
g
lat_ input_ rear_ mag
lat_ input_ sensors
15
gyroread Interface to E- Core gyro gyro 10
steerctl Steering actuator lat_ output
lat_ steer_ input 10
t_ driver Steering actuator inner
loop
lat_ steer_ output
lat_ control_ input 17
sbl_ lat Lateral control
lat_ control_ output
lat_ dvi_ output
lat_ heartbeat_ outpu
t
18
The database variables exchanged by data I/ O and control processes are created and
stored in the database. There is a single producer for each variable, that is, each variable
is updated by only one process, though it can be read by many processes.
3.2 Software Description
Lateral Control Software
The lateral control software gets a trigger from the front magnetometers, i. e. every 2
msec, and reads the following six structures from the database:
• front magnetometers ( DB_ LAT_ INPUT_ FRONT_ MAG)
• rear magnetometers ( DB_ LAT_ INPUT_ REAR_ MAG)
• steering actuator inputs ( DB_ LAT_ CONTROL_ INPUT)
• other lateral sensors ( DB_ LAT_ INPUT_ SENSORS)
• vehicle speed ( DB_ LONG_ INPUT)
• yaw rate from the gyro ( DB_ GYRO)
For debugging purpose, it also reads:
• output from the steering actuator inner loop ( DB_ LAT_ STEER_ OUTPUT)
• output from the steering actuator driver ( DB_ LAT_ OUTPUT)
The lateral control software ( see Figure 3.3) writes the three following structures to
the database:
23
• lateral outputs for the steering actuator ( DB_ LAT_ CONTROL_ OUTPUT) every
2 msec
o steering actuator mode ( 0= manual, 1= auto high, 2= auto low)
o steering command in deg
o steering actuator control mode
• computer heartbeat ( DB_ LAT_ HEARTBEAT_ OUTPUT) every 50 msec
• DVI ( HMI) outputs to the LEDs and speaker ( DB_ LAT_ DVI_ OUTPUT) every
50 msec
Figure 3. 3 Lateral control software
The DVI/ HMI outputs are of 3 kinds:
• lights to the control DVI/ HMI
• lights to the guidance DVI/ HMI
• sound to the speakers
Figure 3. 4 Status DVI/ HMI
24
The control DVI/ HMI ( see Figure 3.4) has 4 lights:
• control manual ( white LED on)
• control auto ( blue LED on)
• control warning ( red LED on)
• control ready ( green LED on)
The guidance DVI/ HMI ( Figure 3.5) has 7 lights:
• guidance left ( if the snowblower is on the far left side of the magnets)
• guidance center left ( if the snowblower is on the left side of the magnets)
• guidance center ( if the snowblower is on top of the magnets)
• guidance center right ( if the snowblower is on the right side of the magnets)
• guidance right ( if the snowblower is on the far right side of the magnets)
• guidance up ( LED on if speed is too low)
• guidance down ( LED on if speed is too high)
Figure 3. 5 Guidance DVI/ HMI
Three different sounds are sent to the speakers:
• audible emergency ( when there is a fault)
• audible takeover ( when end of magnets, i. e. end of guardrail)
• audible acknowledge ( transition to auto steering)
7 kinds of fault are detected:
• yaw rate sensor fault
• steering actuator sensor ( potentiometer) failure
• HMI/ DVI fault ( not used)
• magnetometer ( rear magnetomer health signal) or speed sensor fault
• steering actuator fault ( motor failure, power off, command failure, driver failure,
encoder failure or startup failure)
• system fault ( if we have a continuous spike under automated control)
• multiple faults ( if we have 2 faults of more)
25
Only the 4 last faults require emergency control.
Lateral Source Code
The main file for lateral control is sbl_ main. c. The compilation command is “ make
exec/ sbl_ lat”, to be executed from the “ lat” directory. Below is the list of the main files
and they associated functions ( see also Figure 3.6):
• hst_ cont. c: steering controller
• sbl_ code. c: decoder calls for all sites
• sbl_ db. c: database communication
• sbl_ dvi. c: DVI/ HMI controller
• sbl_ func. c: basic functions
• sbl_ i80. c: decoder for I- 80 ( shoulder side and end of magnets)
• sbl_ mark. c: magnetometer signal processing
• sbl_ obs. c: observer and fault detection
• sbl_ stat. c: state machines
• sbl_ trajc.: trajectory planning
and the associated header files are:
• nat. h: definition of the structures ( front and rear magnetometers, and
DVI)
• constant. h: definition of constants used in different files
• sites. h: definition for the different sites ( RFS, Crows Landing and I- 80)
Figure 3. 6 Lateral source code
The magnetometer calibration tables are in the “ mag_ tab” directory. The calibration
was performed for a ceramic type magnet, for the 6 sensors at the front of the
snowblower and the 7 sensors at the rear of the snowblower. The low and high heights
for calibration chosen were 7 and 11 inches ( 0.18 and 0.28 m) for the front magnetometer
bar, and 7.5 and 11.5 inches ( 0.19 and 0.29 m) for the rear one. The magnetometers are
26
installed as follows, with a total sensor range of [- 0.84 cm, 0.84 cm] on the front and
[- 1.1 m, 1.1 m] on the rear.
See Figure 3.7 for the front magnetometer bar configuration; and Figure 3.8 for the rear
magnetometer bar.
Figure 3. 7 Front magnetometer bar configuration
Figure 3. 8 Rear magnetometer bar configuration
The magnetometer calibration files are generated automatically using the calibration
software. The magnetic calibration tables consist of vertical and horizontal magnetic
strength data that were stored during calibration process. Such table can be plotted as
magnetic strength data at the low and at the high calibration heights as discussed above.
See Figure 3.9, 3.10 and 3.11 for plots of one such table. The snowblower magnetometer
calibration files consist of the following “. h” files:
• t_ cer_ fll. h: table for the front most left magnetometer
• t_ cer_ fl. h: table for the front left magnetometer
• t_ cer_ fcl. h: table for the front center left magnetometer
• t_ cer_ fcr. h: table for the front center right magnetometer
• t_ cer_ fr. h: table for the front right magnetometer
• t_ cer_ frr. h: table for the front most right magnetometer
• t_ cer_ bll. h: table for the rear most left magnetometer
• t_ cer_ bl. h: table for the rear left magnetometer
• t_ cer_ bcl. h: table for the rear center left magnetometer
• t_ cer_ bc. h: table for the rear center magnetometer
• t_ cer_ bcr. h: table for the rear center right magnetometer
• t_ cer_ br. h: table for the rear right magnetometer
• t_ cer_ brr. h: table for the rear most right magnetometer
27
− 1000 − 500 0 500 1000
0
500
1000
1500
2000
2500
3000
Calibration table for snowblower rear center sensor
Figure 3. 9 Calibration table for snowblower rear center @ 2 cm division calibration
− 250 − 200 − 150 − 100 − 50 0 50 100 150 200 250
0
500
1000
1500
2000
2500
3000
Bv Vs Lat. Pos
Figure 3. 10 Calibration table for snowblower ( vertical strength vs lateral position)
28
− 250 − 200 − 150 − 100 − 50 0 50 100 150 200 250
− 1000
− 500
0
500
1000
Bh Vs Lat. Pos
Figure 3. 11 Calibration table for snowblower ( horizontal strength vs lateral position)
Lateral Control Algorithm
The detailed operational procedure of the algorithm of the lateral control module is
described as follows:
1. Get command option and set internal flag accordingly.
2. Call all initialization routines, which include:
• initialization of testing site and decoder
• initialization of observer, state machine, controller
• initialization of magnetometer signal processing
• initialization of database
3. Set priority of the program to 18
4. Set exit point
5. Wait from trigger from database
6. Read database to retrieve sensor measurements
7. Process front and rear magnetometer signals
8. Decode the markers for shoulder side and end of magnets code
9. Call observer for vehicle angle and road curvature
10. Determine which side the blower is tracking on
11. Do fault detection
12. Call finite state machine to transition between different operational states
13. Call finite state machine to switch between different controller states
14. Determine DVI ( HMI) states
15. Call steering controller and switch controller mode through finite state machine
16. Write steering command, DVI ( HMI) outputs and heartbeat to database
17. Save lateral data to file
18. Goto step 5
29
The transition state machine in Step 12 above has 8 states as shown in Figure 3.12.
Figure 3. 12 Transition state machine
Furthermore, the control state machine in Step 13 above has 4 states as shown in Figure
3.13.
Figure 3. 13 Control state machine
30
Software operation procedure
When the snowblower is turned on ( ignition on), all the processes are automatically
started using the script file “ phstart” in / home/ blower/ test. On the other hand, to start the
lateral control manually, the operating procedure is the following:
1. Turn on the ignition switch of the snowblower and the on switch for the PC in the rear
of the cabin.
2. Start the control menu by logging in on the laptop as " blower" with the password
“ avcs”. Go to QNX Photon and get three windows: use one to run the startup menu,
and two to do any debugging.
3. Change to the test directory by typing " cd test".
4. Run the menu by typing " avcs". The menu looks like this:
1. Show configurations
2. Run device drivers
3. Vehicle → MENU
4. Controller & Site → MENU
5. Run lateral control
6. Plot lateral data
7. Plot DVI data
8. Stop device drivers
5. In general, the control menu works by:
• using numerical options to run items.
• the letter ‘ q’ to quit.
• pressing the return key to return to a menu after the menu task is complete.
• text entry to change parameters.
6. Menu item # 4: check the controller and the site.
1. Set lateral controller
2. Show lateral controller configuration
3. Change lateral controller configuration
4. Show site location
5. Change site location
The site should be set to 2 for I- 80 at Donner Summit.
7. Menu item # 5: run the lateral control. The device drivers have already been
automatically started. The lateral control can be terminated by pressing the return key.
8. Menu item # 6: plot lateral data. It contains 4 windows with 4 graphs each:
• window 1:
o measured front lateral measurement before and after trajectory planning
( m)
o measured rear lateral measurement before and after trajectory planning
( m)
o measured and desired steering wheel angle ( deg)
o computed head position of the blower ( m)
• window 2:
o vehicle velocity ( m/ s)
o magnet spacing ( m)
o estimated vehicle angle ( rad)
o estimated road curvature ( 1/ m)
31
• window 3:
o transition state
o steering actuator status and mode
o controller state
o gyro rate ( deg/ sec)
• window 4:
o computed vehicle travel angle ( rad)
o desired and sent steering torque
o mode
o clutch state and fault mode
32
4. Magnetic Lateral Sensing
The development of a reliable and accurate lateral referencing system is crucial to the
success of the lateral guidance system for any steering guidance and control systems of
heavy vehicles, let alone when such vehicle is under severe weather, as well as large road
and load variations. Since the snowblower steering control system operating along
guardrail has very strong accuracy requirements as described in Section 2, the accuracy
when snowblower is operating at a very close distance to the guardrail was used to set the
benchmark accuracy for the magnetometer sensor design for the snowblower. For any
precision steering/ lateral control system, the accuracy requirement for the lateral sensing
system is directly proportional to the required lane tracking accuracy. The desired
tracking accuracy tolerance is set to be no more than 10 cm; such tracking tolerance
under large disturbances and uncertainties implies the need for a high gain control
system. The lateral sensing accuracy requirement was set to be at about 1 centimeter
based on ( a) it is 1/ 10 of the maximum tracking tolerance; and ( 2) it is half of the smallest
operational distance ( 2cm) to the guardrail. The assumption also include that the
installation and measurement accuracy are randomly and evenly distributed along the
correct position.
PATH has proposed and developed a lateral referencing and sensing system that is
based on the magnetic markers embedded under the road to provide the lateral position
and road geometric information. The automatic steering guidance system based on such
technology provides the control system with the following two fundamental pieces of
information: the vehicle position with respect to the roadway, and the current and future
road geometry. Two arrays of magnetometers, one located just under the front axle and
the other at about mid- point behind the front and rear axle.
Extensive development and experiments have been performed on magnetic marker-based
lateral sensing systems for many PATH vehicles equipped with automated steering
control. The vast knowledge available about this lateral sensing technique as well as its
high reliability under winter operational conditions was two of the primary reasons that
this technology was first chosen to support the snowblower automation. Other positive
characteristics of this lateral sensing technique include good accuracy ( better than one
centimeter), insensitivity to weather conditions, and support for binary coding. The
requirement of modifying the infrastructure ( installing magnets) and the inherent “ look-down”
nature ( the sensor measures the lateral displacement at locations within the vehicle
physical boundaries, versus look- ahead ability) of the sensing system are two known
limitations of this technology. The principle idea for this sensing system is
straightforward. Magnetic markers are installed under the roadway delineating the center
of each lane or any other appropriate lines for the specific applications. Magnetometers
mounted under the vehicle sense the strength of the magnetic field as the vehicle passes
over each magnet. Onboard signal processing software calculates the relative
displacement from the vehicle to the magnet based on the magnetic strength and the
knowledge of the magnetic characteristics of the marker. This computation is designed to
33
be insensitive to the vehicle bouncing ( e. g., heave and pitch) and the ever- present natural
and man- made magnetic noises.
Furthermore, the road geometric information, such as road curvatures and super-elevation
can be encoded as a sequence of bits, with each bit corresponding to a magnet.
The polarity of each magnet represents either 1 ( one) or 0 ( zero) in the code. In the
snowblower operations, only four kinds of information are coded: right- side or left- side
guardrail, end- of- magnets, begin- of- bridge. No curvature information is coded. The basic
reason is to ensure a short enough magnetic code length that the snowblower can start
tracking the guardrail as soon as it “ sees” the magnets. In our case, the snowblower
knows the direction of the guardrail within three magnets. In addition to the lateral
displacement measurement and road preview information, other vehicle measurements
such as yaw rate, lateral acceleration, and steering wheel angle may also be used to
improve the performance of such a lateral guidance system.
4.1 Magnetic Noise Effects
Four major noise sources are usually present in the magnetic signal measurements in
a typical vehicle operational environment: earth field, local magnetic field distortion,
vehicle internal electromagnetic field, and electrical noise.
The most frequent external disturbance is the ever- present earth’s permanent
magnetic field, which is usually on the order of 0.5 Gauss. The value of the earth field
measured by the magnetometers on the vehicle depends on the location of the vehicle on
earth as well as the altitude and orientation of the vehicle. Although the earth magnetic
field usually changes slowly, sharp turns and severe braking can quickly change the field
measurements along the vehicle axes.
The most serious noise problems are caused by local anomalies due to the presence of
roadway structural supports, reinforcing rebar, and the ferrous components in the vehicle
or under the roadway. Underground power lines are another source of such local field
distortion. Rebar or structural support usually creates a sharp change in the background
magnetic field and sometimes is difficult to identify. Most signal processing algorithms
will have some difficulty recovering from such sharp distortions. The ferrous components
in the vehicle, on the other hand, can be isolated as long as their locations are fixed with
respect to the magnetometers, or are located at a significant distance from the sensors.
A third source of noise comes from the alternating electric fields generated by various
motors or rotating permanent magnets or magnetized materials operating in the vehicle.
These rotating “ magnets” may include alternator, fan, electric pump, steal belts inside
tires, compressor and other actuators. However, their effects vary according to the
rotational speed and distance from the magnetometers. The higher the rotating speed, or
the farther it is placed away from the magnetometers, the less the resultant noise
becomes. Sometimes modest changes in sensor placement can alter the size of such
disturbances.
34
The last common noise source arises from the electronic noise in the measurement
signal itself. Such noise can be created by the voltage fluctuations in the electrical
grounding or from the power source. It can also be a result of poor wiring insulation
against electromagnetic disturbances. Usually, the longer the wire, the higher such noise.
Although low- pass filtering can reduce the magnitude of such disturbances, noticeable
degradation of the magnetic sensor signal process algorithm occurs when such noise level
exceeds 0.04 Gauss. Digital transmission of magnetic field measurements or local
embedded processor is two possible approaches that can significantly reduce such noise.
4.2 Tire- induced Magnetic Noise
One less- common magnetic noise observed in the snowblower is the tire- induced
magnetic noise. Such noise typically appears as alternating magnetic fields from the
magnetized steel- belted tires. Magnetic field strengths have been measured as a function
of frequency directly at the tire rotation. Measurements at the tire showed field strength
up to 1 Gauss ( 100 microTesla).
To ensure such magnetic noise does not affect the accuracy of the lateral
measurements. Before a new magnetometer sensor bar can be installed on the
snowblower, PATH has constructed a temporary wooden magnetometer bar that can be
“ strapped” to the snowblower to investigate the magnetic noise effect resulted from the
magnetized rear tires. For example, before mounting the rear magnetometer bar, four
different sensor bar locations were tested: at locations 13, 19, 25, and 31 inches to the
rear tires; and all at 10 inches above the ground. Figure 4.1 shows two examples of the
magnetic field interference characteristics from the magnetized rear tires. The left- most
and right- most sensors were chosen as examples because they are the closest to the rear
tires and hence have the most noise impact. As can be observed in Figure 4.1, the right-most
sensor exhibits the strongest interference from the tire magnetic field; and the
further away the sensor, the less the interference. Moreover, the magnetic field
measurement resulted from the “ noise” of the rear tire can be as high as 200 mV
( 1V= 0.67G) peak to peak at a distance of 13 inches to the tire rim; a 10 folds noise
increase to the nominal 20 mV static noise of the magnetometers. Figure 4.2 shows the
tire magnetic noise effect ( peak- to- peak value) to the 3 right magnetometers with respect
to various magnetometer sensor bar locations. It concludes that the sensor bar would
requires at least 19 inches distance from the tire rim in order to has a noticeable reduction
of the tire magnetic interference.
35
Figure 4. 1 Examples of Snowblower magnetic field noise interference from tire
Figure 4. 2 Snowblower tire magnetic noise vs. magnetometer sensor bar locations
36
4.2 Magnetic Sensing Algorithm
One of the important attributes of the lateral sensing system is its reliability.
Currently, there exist several algorithms designed to detect the relative position between
the marker and sensor ( magnetometer), as well as to read the code embedded within a
sequence of these markers. Three magnetic marker detection and mapping algorithms
have been experimented with by PATH. The first is called the “ peak- mapping” method
that utilizes a single magnetometer to estimate the marker’s relative lateral position when
the sensor is passing over the magnet. The second algorithm is the “ vector ratio” method
that requires a pair of magnetometers to sample the field at two locations. It returns a
sequence of lateral estimates in a neighborhood surrounding, but not including the peak.
The third is the “ differential peak- mapping” algorithm that compares the magnetic field
measurements at two observation points to eliminate the common- mode contributions
and reconstructs a functional relationship between the differential sensor readings and the
lateral position using the knowledge of the sensor geometry. The “ peak- mapping”
algorithm was selected for the snowblower project because it has been proven effective
over a wide range of speeds and has been widely applied in many experimental
applications conducted by PATH.
In the heavy vehicle operational environment, the magnetic field maps can deviate
quite significantly from the theoretical dipole equation prediction because of the massive
amount of ferrous material from the body structure located just above the magnetometers.
Numerical mapping created by empirical data gathering ( calibration) is used to create the
associated inverse maps. Figures 4.3 and 4.4 show the front and rear magnetic tables for
the snowblower, respectively. The figures consist of tables of the seven magnetometers
starting from the right side of the bus to the left, designated as follows: right- right, right,
center- right, center, center- left, left and left- left. Each table is obtained with two sets of
calibration data, one at a lower sensor height ( at around 7 inches from the magnetometer
to the magnet) and the other at a higher sensor height ( at 11 inches from the
magnetometer to the magnet). Each half- circle in the table consists of vertical and
horizontal fields of the marker that are collected at 2- cm interval of lateral displacement.
The magnetic tables clearly depict the nonsymmetrical natural for the magnetic field due
the adjacent ferrous material. The calibration process was repeated for every
magnetometer to ensure that the static local magnetic effects for each magnetometer were
accounted for.
When a magnetometer bar is not properly calibrated, the lateral position measured
can exhibit significant error. Figure 4.5 shows the both the problem areas before the
proper calibration ( using the rear center table for all rear magnetometers) and the smooth
rear measurements ( using appropriate calibration tables) when the snow blower is driven
across the magnet track from right side of the road toward the left.
37
Figure 4. 3 Snowblower Front Magnetometer Calibration Tables
Figure 4. 4 Snowblower Rear Magnetometer Calibration Table
38
Figure 4. 5 Rear sensor new calibration & signal processing comparison.
4.3 Signal Processing
Low Pass
Filter
Low Pass
Filter
Bv
Bh
z- 1 z- 1 z- 1 z- 1 z- 1
z- 1 z- 1 z- 1 z- 1 z- 1
Variance
Calculation
latest peak
Logic based
on Variance
Peak latch
Peak Finder
Earth latch
Bv
Bv
Peak latch
Earth latch
Bh
Bh
Synchro
+
-
+
-
Mapping
Bv
Bh
y
y
1
2
y
3
y
4
yr
Figure 4. 6 “ Peak- Mapping” Magnetometer Signal Processing Block Diagram
The magnetometers signal processing for the “ peak- mapping” method involves three
procedures: peak detection, earth field removal and lateral displacement table look- up
( see Figure 4.6 for block diagram of signal processing algorithm, and Figure 4.7 for one
of the peak detection algorithm). Although it is straightforward in principle, it becomes
complicated when the reliability of the process is the major concern. Many parameters in
39
the lateral sensing signal processing software need to be tuned in order to provide
consistent lateral displacement information regardless of vehicle speeds, orientations,
operating lateral offsets and vehicle body motions. Debugging can become very time
consuming when failure conditions cannot be recreated. To improve the reliability of the
lateral sensing system with the magnetic road markers, PATH has developed a
“ reconstructive” software system for the lateral sensing signal processing that supports
the tuning of the parameters using stored real- time data. In such a setup, any erroneous
situation can be recreated in a lab environment and debugged with ease.
Figure 4. 7 Peak detection block diagram
40
5. Magnet Installation
5.1 Test Site
Magnets were installed along the eastbound and westbound guard rails of Interstate
80 at Lake Tahoe around Donner Summit ( see a map in Figure 5.1) in 2001, at 1.2 meter
spacing and 4 feet away from the guardrail. See Figure 5.2 for a photo of the installation
process; and also Figure 5.3 for a photo of the magnets installed along the guardrail.
There are 5 sections on the westbound shoulder and 3 sections on the eastbound shoulder,
between Soda Springs and Kingvale. The total number of magnets installed is 1222, i. e. a
total length of about 1 mile.
Figure 5. 1 Map of the test area
Figure 5. 2 Magnet Installation
41
Figure 5. 3 Magnets along guardrail
There are 8 sections of guardrail with magnets installed, including:
• 5 sections on the WB lane ( 4 on the right shoulder, 1 on the left shoulder)
• 3 sections on the EB lane ( 2 on the right shoulder, 1 on the left shoulder)
The longest section is 477 magnets long, the shortest one 55 magnets. Figure 5.4
illustrates these guardrails with magnets. The sharpest curve is 457 m radius to the left.
Magnets were also installed on both shoulders of the Kingvale overpass, in each
direction. All the magnets are ceramic type, except for the Kingvale overpass where they
are rare earth.
Figure 5. 4 Illustration of guardrail installed with magnets in I- 80
5.2 Magnet Code Description
For each section of guardrail, 25 magnets were installed before the guardrail ( to turn
the auto- steering on) and 10 magnets after ( to turn the auto- steering off). The magnet
polarity is 0 ( south pole up) on the right- side shoulder and 1 ( north pole up) on the left-side
shoulder. The code for the “ end of magnets” is to the interchange the polarity for the
42
last 12 magnets, for example [ 101010101010]. Note that there is no curvature
information coded due to the required fast “ control initialization” as well as the long code
length to encode different curvature and curvature changes.
Figure 5. 5 Illustration of beginning and ending of a magnet section
The westbound shoulder magnets are as follows:
• Section 1:
station 3395.6 -> 2824.4
477 magnets on the right shoulder
25 magnets before the guard rail, 10 magnets after
code on last 12 magnets
curvatures in the section: 762 m ( to the left) , straight , - 701.04 m ( to the right)
• Section 2:
station 2280.0 -> 1986.0
246 magnets on the right shoulder
25 magnets before the guard rail, 11 magnets after
code on last 12 magnets
curvatures in the section: straight, 457.2 m ( to the left)
• Section 3:
station 13536.3 -> 13451.1
72 magnets on the right shoulder
25 magnets before the guard rail, 10 magnets after
code on last 12 magnets
curvature in the section: straight
• Section 4:
station 13000.6 -> 12891.4
92 magnets on the right shoulder of the bridge
25 magnets before the guard rail, 10 magnets after
magnets 49 & 50: [ 11] for beginning of bridge
magnets 78 & 79: [ 11] for end of bridge
code on last 13 magnets
curvature in the section: straight
• Section 5:
station 12997.0 -> 12891.4
89 magnets on the left shoulder of the bridge
25 magnets before the guard rail, 10 magnets after
magnets 49 & 50: [ 00] for beginning of bridge
magnets 78 & 79: [ 00] for end of bridge
code on last 13 magnets
curvature in the section: straight
43
The eastbound shoulder magnets are as follows:
• Section 1:
station 12521.5 -> 12586.3
55 magnets on the right shoulder
25 magnets before the guard rail, 10 magnets after
code on last 12 magnets
curvatures in the section: 792.48 m ( to the left)
• Section 2:
station 12832.1 -> 12956.9
105 magnets on the right shoulder of the bridge
25 magnets before the guard rail, 10 magnets after
magnets 59 & 60: [ 11] for beginning of bridge
magnets 94 & 95: [ 11] for end of bridge
code on last 10 magnets
curvature in the section: straight
• Section 3:
station 12854.9 -> 12956.9
86 magnets on the left shoulder
25 magnets before the guard rail, 10 magnets after
magnets 59 & 60: [ 00] for beginning of bridge
magnets 94 & 95: [ 00] for end of bridge
code on last 10 magnets
curvature in the section: straight
44
6. Hardware Modifications
6.1 Hardware Components
As illustrated in Figure 2.3, the automated steering control is implemented to a
conventional Kodiak Northwest single engine rotary snowplow with full hydrostatics.
The concept of implementation is to maintain all “ manual” operational functionalities the
same as those of the original snowblower. All automated steering components are “ add-on”
devices or systems. The main add- on components are described below. A computer
with a data acquisition unit processes information and determines control and guidance
actions. The lateral positioning system, consisting of two arrays of magnetometers,
measures the field strength of magnetic markers installed under the roadway. The
steering actuator, using a custom- made DC motor attached to the steering column with
angular sensors, steers the front wheels. A yaw gyro and an axle speed sensor provide
supplementary motion data under extremely low speeds. A Human Machine Interface
( HMI) or Driver Vehicle Interface ( DVI) unit, consisting of electronic circuit, a toggle
switch, LED displays and an audible device, interfaces with the operator with information
and commands for automation. Table 6.1 details these add- on components.
Table 6.1 Automated snowblower add- on components
Component Description Functions
Computer 10 slot industrial computer Control/ actuation/ signal
processing/ HMI/ fault
detection
Power supply & inverter EGS ( GLQ- 04- 200) power
supply & Statpower PROSine
1000 inverter
Provide power to computer
& circuits
Front magnetometer bar 6 Applied Physics ASPS535
magnetometers with custom-made
enclosure
Measure magnetic strengths
from buried magnets
Rear magnetometer bar 7 Applied Physics ASPS535
magnetometers with custom-made
enclosure
Measure magnetic strengths
from buried magnets
I/ O board
( inside computer)
National Instrument AT- MIO-
64E- 3
Use for timer inputs, analog
inputs and outputs, and
digital I/ O including data
from magnetometers
I/ O board
( inside computer)
National Instrument AT- AO- 6 Use for analog outputs and
digital I/ O
I/ O board
( inside computer)
PC- TIO- 10 Use for timer and digital
I/ O
I/ O board
( inside computer)
Microcomputer Systems EIC-
325
Use for interfacing encoder
( steering sensor)
Steering actuator NSK custom- made DC- motor actuator with
current- mode control and
steering angle sensors ( see
45
Section 7)
Speed sensor and circuit Magnetic pick- up and custom-made
circuit
Measure drive shaft speed
Yaw rate sensor KVH E- core Fiber optic rate
gyro
Measure yaw rate
HMI: control circuit Custom- made Provide independent HMI
control
HMI: switch & button Custom- made Allow driver to input to the
automated system
HMI: LED display Custom- made ( LED’s) Provide driver with
information about the
automated system
HMI: audible unit Custom- made, speaker &
sound board
Provide driver with audible
information
Enclosure Provide weather- proof
enclosure for computer, I/ O
boards, power supplies, and
HMI circuit
Figure 6. 1 Enclosure and components
46
Power supplies
The 12V power supply, coming from the snowblower batteries, is used for powering
most sensors and electronics on the system. The 12V batteries are also connected to a
120V AC inverter for powering the computer ( see Figure 6.1). All the circuit boards and
electronics are powered by 12V bus bar supply through 10A circuit breaker. The steering
actuator is powered by 12V connected to the battery through 60A circuit breaker. All the
sensors power supply is directly from 12V bus bar except for the magnetometers. The
magnetometers are powered by +/- 15V which is powered from the 12V bus bar. Bypass
capacitors are put at the output of the power adapters to reduce the power noise.
Steering actuator
The motor assembly is manufactured by NSK as shown in Figure 6.2. The steering
actuator motor assembly consists of a steering column, DC motor actuating steering
column, an electromagnetic clutch, angle sensors ( incremental encoder and
potentiometer), and ECU. See Section 7 for detailed description.
Figure 6. 2 Steering actuator ( not assembled)
Sensors
Yaw rate sensor ( see Figure 6.3) is behind the cab of the snowblower. Magnetometers
are mounted on weather- proof enclosures under the front axle and in between the front
and rear axle. The front sensor bar has a dent resulted from a past winters operation. The
performance of the front sensor bar has been verified despite the dent. Figure 6.4 shows
the dented front sensor bar and a spare in case of a severe damage in the future. On the
other hand, several installation issues with respect to the rear magnetometers were
encountered during the development and test periods; see Section 6.2 for detailed
description on those issues. For steering position, steering pot and encoder ( see Figure
6.2) are used and installed as part of the steering actuator ( Refer to Section 7 for more
detailed descriptions). The speed sensor has been modified several times in the
development and test cycles. The original speed sensor ( Figure 6.5, left) operated to
speed as low as 2 mph (~ 1m/ s) based on the initial speed requirement of minimum 3mph.
The current speed sensor, as shown in Figure 6.5- right, was reinstalled and the associated
47
speed- signal processing algorithm was tested to extend to the low speed range from 2
mph to 0.25 mph (~ 0.5m/ s).
Figure 6. 3 Yaw rate sensor and enclosure
Figure 6. 4 Existing dented front magnetometer bar and a spare
Figure 6. 5 Speed sensor ( left: old; right: improved)
48
Sensor interface
Sensors are connected with computer through I/ O boards as illustrated in Figure 6.6
and 6.7. Thirty- nine input channels are used on AT- MIO- 64E- 3 card. Twenty- six
channels are used for thirteen magnetometers ( two channels for each magnetometer), and
two channels are used for steering potentiometer and steering motor condition. Three
digital inputs and two analog outputs are used on AT- A0- 6. Thirteen digital input
channels are used for magnetometer health signals on PC- TIO- 10. In addition, two digital
inputs are used for auto/ manual transition switches. Sixteen channels outputs are used for
outputs to HMI and steering actuator. Three channels for triggering three different sounds
recorded in the alert audible system. Eleven channels are for various LED’s: seven for
guidance display and four for status display. There are two additional channels, one is for
steering clutch, and the other for steering torque command.
Figure 6. 6 Interface between snowblower sensors and computer ( 1)
Figure 6. 7 Interface between snowblower sensors, commands and computer ( 2)
49
HMI & HMI circuit
The HMI ( human machine interface) consists of a transition toggle switch, a set of
status LED’s, a set of guidance LED’s, an audible unit, a system switch, and an
emergency button. They are controlled by a HMI circuit which decides whether the
control is based on the HMI circuit or through the computer commands. This HMI circuit
will warn the driver when the automated control failed to function. This warning system
listens to the beat sent from computer. It will do nothing as long as the heart beat is on.
This system will trigger the emergency sound when the heart beat dies or the beat misses
beating more than twice as shown in the diagram below ( Figure 6.8). The timer chip is
set to be triggered every 250ms ( millisecond). The out put of the chip will be held high
as long as the chip is triggered every 250ms, therefore, the collector of transistor is low
and the emergency sound will be silenced. The circuit will also provide LED and sound
command during computer boot- up period as well as when there is no heart- beat.
However, it will also relinquish LED/ sound control to the computer command when there
is a heart- beat.
Figure 6. 8 HMI and heart beat timing
6.2 Wiring and Circuit Diagram
This section exhibits various wiring diagrams for the snowblower hardware
installation.
Figure 6.9 shows the overall snowblower automated system wiring and circuit
diagram. The highlighted areas in Figure 6.9 indicate those diagrams that with a more
detailed figure followed. Figure 6.10, 6.11, 6.12, 6.13, 6.14 illustrate these more detailed
wiring/ circuit diagrams; they are steering actuator & transition switches, AT- MIO- 64E- 3
& magnetometers, I/ O boards, HMI circuit, and heat- beat detection, respectively.
50
Figure 6. 9 Snowblower wiring and circuit diagram ( overall)
Figure 6. 10 Snowblower wiring and circuit diagram ( steering actuator & transition
switches)
51
Figure 6. 11 Snowblower wiring and circuit diagram ( AT- MIO- 64E- 3 & magnetometers)
Figure 6. 12 Snowblower wiring and circuit diagram ( I/ O boards)
52
Figure 6. 13 Snowblower wiring and circuit diagram ( HMI circuit)
Figure 6. 14 Snowblower wiring and circuit diagram ( heart beat detection)
6.3 Rear Magnetometer Bar Installation
One problem encountered during hardware installation worthwhile noted relates to
the rear magnetometer bar installation. The rear magnetometer bar was first installed
under the rear bumper of the snowblower to minimize the noise amplification during
vehicle angle computation by maximizing the distance between the front and rear bars.
53
Through the discussions with the Kingvale maintenance yard lead mechanics, this
original rear magnetometer bar was not installed at an appropriate location. The tail end
of the snowblower may not be compatible with one of the snow removal operations that
performed by