Stanley: The Robot that Won
the DARPA Grand Challenge
Sebastian Thrun, Mike Montemerlo,
Hendrik Dahlkamp, David Stavens,
Andrei Aron, James Diebel, Philip Fong,
John Gale, Morgan Halpenny,
Gabriel Hoffmann, Kenny Lau, Celia Oakley,
Mark Palatucci, Vaughan Pratt,
and Pascal Stang
Stanford Artificial Intelligence Laboratory
Stanford University
Stanford, California 94305
Sven Strohband, Cedric Dupont,
Lars-Erik Jendrossek, Christian Koelen,
Charles Markey, Carlo Rummel,
Joe van Niekerk, Eric Jensen,
and Philippe Alessandrini
Volkswagen of America, Inc.
Electronics Research Laboratory
4009 Miranda Avenue, Suite 100
Palo Alto, California 94304
Gary Bradski, Bob Davies, Scott Ettinger,
Adrian Kaehler, and Ara Nefian
Intel Research
2200 Mission College Boulevard
Santa Clara, California 95052
Pamela Mahoney
Mohr Davidow Ventures
3000 Sand Hill Road, Bldg. 3, Suite 290
Menlo Park, California 94025
Received 13 April 2006; accepted 27 June 2006
••••••••••••••••• ••••••••••••••
Journal of Field Robotics 23(9), 661–692 (2006) © 2006 Wiley Periodicals, Inc.
Published online in Wiley InterScience (www.interscience.wiley.com). • DOI: 10.1002/rob.20147
This article describes the robot Stanley, which won the 2005 DARPA Grand Challenge.
Stanley was developed for high-speed desert driving without manual intervention. The
robot’s software system relied predominately on state-of-the-art artificial intelligence
technologies, such as machine learning and probabilistic reasoning. This paper describes
the major components of this architecture, and discusses the results of the Grand Chal-
lenge race.
© 2006 Wiley Periodicals, Inc.
1. INTRODUCTION
The Grand Challenge was launched by the Defense
Advanced Research Projects Agency 共DARPA兲 in
2003 to spur innovation in unmanned ground vehicle
navigation. The goal of the Challenge was to develop
an autonomous robot capable of traversing unre-
hearsed off-road terrain. The first competition, which
carried a prize of $1M, took place on March 13, 2004.
It required robots to navigate a 142-mile long course
through the Mojave desert in no more than 10 h. 107
teams registered and 15 raced, yet none of the par-
ticipating robots navigated more than 5% of the entire
course. The challenge was repeated on October 8,
2005, with an increased prize of $2M. This time, 195
teams registered and 23 raced. Of those, five teams
finished. Stanford’s robot “Stanley” finished the
course ahead of all other vehicles in 6 h, 53 min, and
58 s, and was declared the winner of the DARPA
Grand Challenge; see Figure 1.
This paper describes the robot Stanley, and its
software system in particular. Stanley was developed
by a team of researchers to advance the state-of-the-
art in autonomous driving. Stanley’s success is the re-
sult of an intense development effort led by Stanford
University, and involving experts from Volkswagen
of America, Mohr Davidow Ventures, Intel Research,
and a number of other entities. Stanley is based on a
2004 Volkswagen Touareg R5 TDI, outfitted with a six
processor computing platform provided by Intel, and
a suite of sensors and actuators for autonomous driv-
ing. Figure 2 shows images of Stanley during the race.
The main technological challenge in the develop-
ment of Stanley was to build a highly reliable system,
capable of driving at relatively high speeds through
diverse and unstructured off-road environments, and
to do all this with high precision. These requirements
led to a number of advances in the field of autono-
mous navigation, as surveyed in this paper. Methods
were developed, and existing methods extended, in
the areas of long-range terrain perception, real-time
collision avoidance, and stable vehicle control on slip-
pery and rugged terrain. Many of these develop-
ments were driven by the speed requirement, which
rendered many classical techniques in the off-road
driving field unsuitable. In pursuing these develop-
ments, the research team brought to bear algorithms
Figure 1. 共a兲 At approximately 1:40 pm on Oct 8, 2005, Stanley was the first robot to complete the DARPA Grand
Challenge. 共b兲 The robot is being honored by DARPA Director Dr. Tony Tether.
662 • Journal of Field Robotics—2006
Journal of Field Robotics DOI 10.1002/rob
from diverse areas including distributed systems,
machine learning, and probabilistic robotics.
1.1. Race Rules
The rules 共DARPA, 2004兲 of the DARPA Grand Chal-
lenge were simple. Contestants were required to
build autonomous ground vehicles capable of tra-
versing a desert course up to 175-miles long in less
than 10 h. The first robot to complete the course in
under 10 h would win the challenge and the $2M
prize. Absolutely no manual intervention was al-
lowed. The robots were started by DARPA personnel
and from that point on had to drive themselves.
Teams only saw their robots at the starting line and,
with luck, at the finish line.
Both the 2004 and 2005 races were held in the
Mojave desert in the southwest United States. The
course terrain varied from high-quality graded dirt
roads to winding rocky mountain passes; see Figure
2. A small fraction of each course traveled along
paved roads. The 2004 course started in Barstow,
CA, approximately 100 miles northeast of Los Ange-
les, and finished in Primm, NV, approximately
30 miles southwest of Las Vegas. The 2005 course
both started and finished in Primm, NV.
The specific race course was kept secret from all
teams until 2 h before the race. At this time, each
team was given a description of the course on CD-
ROM in a DARPA-defined route definition data for-
mat 共RDDF兲. The RDDF is a list of longitudes, lati-
tudes, and corridor widths that define the course
boundary, and a list of associated speed limits; an
example segment is shown in Figure 3. Robots that
travel substantially beyond the course boundary risk
disqualification. In the 2005 race, the RDDF con-
tained 2,935 waypoints.
The width of the race corridor generally tracked
the width of the road, varying between 3 and 30 m
in the 2005 race. Speed limits were used to protect
important infrastructure and ecology along the
course, and to maintain the safety of DARPA chase
drivers who followed behind each robot. The speed
limits varied between 5 and 50 mph. The RDDF de-
fined the approximate route that robots would take,
Figure 2. Images from the race.
Figure 3. A section of the RDDF file from the 2005
DARPA Grand Challenge. The corridor varies in width
and maximum speed. Waypoints are more frequent in
turns.
Thrun et al.: Stanley: The Robot that Won
• 663
Journal of Field Robotics DOI 10.1002/rob
so no global path planning was required. As a result,
the race was primarily a test of high-speed road
finding, obstacle detection, and avoidance in desert
terrain.
The robots all competed on the same course;
starting one after another at 5 min intervals. When a
faster robot overtook a slower one, the slower robot
was paused by DARPA officials, allowing the second
robot to pass the first as if it were a static obstacle.
This eliminated the need for robots to handle the
case of dynamic passing.
1.2. Team Composition
The Stanford Racing Team team was organized into
four major groups. The Vehicle Group oversaw all
modifications and component developments related
to the core vehicle. This included the drive-by-wire
systems, the sensor and computer mounts, and the
computer systems. The group was led by researchers
from Volkswagen of America’s Electronics Research
Lab. The Software Group developed all software, in-
cluding the navigation software and the various
health monitor and safety systems. The software
group was led by researchers affiliated with Stan-
ford University. The Testing Group was responsible
for testing all system components and the system as
a whole, according to a specified testing schedule.
The members of this group were separate from any
of the other groups. The testing group was led by
researchers affiliated with Stanford University. The
Communications Group managed all media relations
and fund raising activities of the Stanford Racing
Team. The communications group was led by em-
ployees of Mohr Davidow Ventures, with participa-
tion from all other sponsors. The operations over-
sight was provided by a steering board that included
all major supporters.
2. VEHICLE
Stanley is based on a diesel-powered Volkswagen
Touareg R5. The Touareg has four-wheel drive
共4WD兲, variable-height air suspension, and automatic
electronic locking differentials. To protect the vehicle
from environmental impact, Stanley has been outfit-
ted with skid plates and a reinforced front bumper. A
custom interface enables direct electronic actuation of
both the throttle and brakes. A DC motor attached to
the steering column provides electronic steering con-
trol. A linear actuator attached to the gear shifter
shifts the vehicle between drive, reverse, and parking
gears 关Figure 4共c兲兴. Vehicle data, such as individual
wheel speeds and steering angle, are sensed auto-
matically and communicated to the computer system
through a CAN bus interface.
The vehicle’s custom-made roof rack is shown in
Figure 4共a兲. It holds nearly all of Stanley’s sensors.
The roof provides the highest vantage point of the ve-
hicle; from this point, the visibility of the terrain is
best, and the access to global positioning system
共GPS兲 signals is least obstructed. For environment
perception, the roof rack houses five SICK laser range
finders. The lasers are pointed forward along the
driving direction of the vehicle, but with slightly dif-
ferent tilt angles. The lasers measure cross sections of
the approaching terrain at different ranges out to
25 m in front of the vehicle. The roof rack also holds
a color camera for long-range road perception, which
is pointed forward and angled slightly downward.
For long-range detection of large obstacles, Stanley’s
roof rack also holds two 24 GHz RADAR sensors,
supplied by Smart Microwave Sensors. Both RADAR
sensors cover the frontal area up to 200 m, with a cov-
erage angle in azimuth of about 20°. Two antennae of
this system are mounted on both sides of the laser
sensor array. The lasers, camera, and radar system
Figure 4. 共a兲 View of the vehicle’s roof rack with sensors. 共b兲 The computing system in the trunk of the vehicle. 共c兲 The
gear shifter, control screen, and manual override buttons.
664 • Journal of Field Robotics—2006
Journal of Field Robotics DOI 10.1002/rob
comprise the environment sensor group of the system.
That is, they inform Stanley of the terrain ahead, so
that Stanley can decide where to drive, and at what
speed.
Further back, the roof rack holds a number of ad-
ditional antennae: One for Stanley’s GPS positioning
system and two for the GPS compass. The GPS po-
sitioning unit is a L1/L2/Omnistar HP receiver. To-
gether with a trunk-mounted inertial measurement
unit 共IMU兲, the GPS systems are the positioning sensor
group, whose primary function is to estimate the lo-
cation and velocity of the vehicle relative to an exter-
nal coordinate system.
Finally, a radio antenna and three additional GPS
antennae from the DARPA E-Stop system are also lo-
cated on the roof. The E-Stop system is a wireless link
that allows a chase vehicle following Stanley to safely
stop the vehicle in case of emergency. The roof rack
also holds a signaling horn, a warning light, and two
manual E-stop buttons.
Stanley’s computing system is located in the ve-
hicle’s trunk, as shown in Figure 4共b兲. Special air
ducts direct air flow from the vehicle’s air condition-
ing system into the trunk for cooling. The trunk fea-
tures a shock-mounted rack that carries an array of
six Pentium M computers, a Gigabit Ethernet switch,
and various devices that interface to the physical sen-
sors and the Touareg’s actuators. It also features a
custom-made power system with backup batteries,
and a switch box that enables Stanley to power-cycle
individual system components through software.
The DARPA-provided E-Stop is located on this rack
on additional shock compensation. The trunk assem-
bly also holds the custom interface to the Volkswagen
Touareg’s actuators: The brake, throttle, gear shifter,
and steering controller. A six degree-of-freedom IMU
is rigidly attached to the vehicle frame underneath
the computing rack in the trunk.
The total power requirement of the added instru-
mentation is approximately 500 W, which is pro-
vided through the Touareg’s stock alternator. Stan-
ley’s backup battery system supplies an additional
buffer to accommodate long idling periods in desert
heat.
The operating system run on all computers is
Linux. Linux was chosen due to its excellent network-
ing and time sharing capabilities. During the race,
Stanley executed the race software on three of the six
computers; a fourth was used to log the race data
共and two computers were idle兲. One of the three race
computers was entirely dedicated to video process-
ing, whereas the other two executed all other soft-
ware. The computers were able to poll the sensors at
up to 100 Hz, and to control the steering, throttle and
brake at frequencies up to 20 Hz.
An important aspect in Stanley’s design was to
retain street legality, so that a human driver could
safely operate the robot as a conventional passenger
car. Stanley’s custom user interface enables a driver to
engage and disengage the computer system at will,
even while the vehicle is in motion. As a result, the
driver can disable computer control at any time of the
development, and regain manual control of the ve-
hicle. To this end, Stanley is equipped with several
manual override buttons located near the driver seat.
Each of these switches controls one of the three major
actuators 共brakes, throttle, and steering兲. An addi-
tional central emergency switch disengages all com-
puter control and transforms the robot into a conven-
tional vehicle. While this feature was of no relevance
to the actual race 共in which no person sat in the car兲,
it proved greatly beneficial during software develop-
ment. The interface made it possible to operate Stan-
ley autonomously with people inside, as a dedicated
safety driver could always catch computer glitches
and assume full manual control at any time.
During the actual race, there was of course no
driver in the vehicle, and all driving decisions were
made by Stanley’s computers. Stanley possessed an
operational control interface realized through a
touch-sensitive screen on the driver’s console. This
interface allowed Government personnel to shut
down and restart the vehicle, if it became necessary.
3. SOFTWARE ARCHITECTURE
3.1. Design Principles
Before both the 2004 and 2005 Grand Challenges,
DARPA revealed to the competitors that a stock
4WD pickup truck would be physically capable of
traversing the entire course. These announcements
suggested that the innovations necessary to success-
fully complete the challenge would be in designing
intelligent driving software, not in designing exotic
vehicles. This announcement and the performance of
the top finishers in the 2004 race guided the design
philosophy of the Stanford Racing Team: Treat au-
tonomous navigation as a software problem.
Thrun et al.: Stanley: The Robot that Won • 665
Journal of Field Robotics DOI 10.1002/rob