DARPA Grand Challenge
The DARPA Grand Challenge was a series of prize competitions for driverless ground vehicles run by the United States Defense Advanced Research Projects Agency (DARPA). The first two events, held in the Mojave Desert on March 13, 2004 and October 8, 2005, asked fully autonomous vehicles to cover a long off-road course with no human intervention; a third event, the DARPA Urban Challenge of November 3, 2007, moved the contest to a mock city with moving traffic.[1][2][3][6] No vehicle finished in 2004. In 2005 five vehicles completed a 132-mile course and Stanford University's robot Stanley, a modified Volkswagen Touareg, won the $2 million prize. In 2007 six of 11 finalists finished the urban course, and Carnegie Mellon University's Tartan Racing team won with a Chevrolet Tahoe named Boss.[1][2][5] DARPA credits the challenges with helping to create the research community behind later autonomous vehicles, and when Google disclosed its self-driving car project in 2010, it said it had gathered "some of the very best engineers from the DARPA Challenges" to build it.[1][12]
Background
DARPA describes the Grand Challenge as its first major attempt to use a prize competition to draw in new performers and ideas from outside the usual defense research community.[2] The military motivation was explicit. A defense authorization bill passed in 2000 called for one third of U.S. ground combat vehicles to be unmanned by 2015, and DARPA conceived the Grand Challenge to speed up development of the necessary technology.[8] The Carnegie Mellon team behind Boss later wrote that the competition was a response to this congressional mandate and that existing ground-vehicle research programs were not expected to meet it.[5] DARPA's own account gives the longer-term aim as autonomous vehicles that could replace soldiers in dangerous jobs such as supply convoys.[2]
The rules for the desert events were simple. A vehicle had to drive itself over a desert route within 10 hours; no manual intervention of any kind was allowed once DARPA staff started the robot.[4] The route was kept secret until two to three hours before the start, when each team received it on a CD-ROM as a list of GPS waypoints; in 2005 it came in a DARPA-defined route definition data format (RDDF), which also set a corridor width and a speed limit for each segment.[4][7] Vehicles started at intervals, each followed by a DARPA chase vehicle.[7] In 2005, when a faster robot caught a slower one, officials paused the slower one so that it could be passed as a stationary obstacle.[4]
The 2004 Grand Challenge
DARPA announced the first event with a $1 million prize. More than a hundred teams applied; DARPA cut the field to 25 after technical reviews and site visits, and after nearly a week of qualifying trials at the California Speedway in Fontana, 15 vehicles were cleared to run.[7][1] Grand Challenge rules required each team to be led by a U.S. citizen.[7] The 142-mile route ran from Barstow, California, to Primm, Nevada.[2][9] The field was a mix of SUVs, dune buggies, large trucks and an autonomous motorcycle.[16][7]
The race began at dawn on March 13, 2004.[2] No vehicle came close to the finish. Sandstorm, a modified Humvee entered by Carnegie Mellon's Red Team under robotics professor William "Red" Whittaker, went the farthest, about 7.4 miles (DARPA's later account rounds this to 7.5 miles), before it high-centered on the berm of a mountain road while coming off a switchback.[9][2] IEEE Spectrum reported that the vehicle kept revving until much of the rubber on its front tires had burned away and both half-shafts had broken.[7]
| Vehicle or team | Approximate distance | How the run ended |
|---|---|---|
| Sandstorm (Carnegie Mellon Red Team) | 7.4 mi (about 12 km) | Stuck on a berm after a switchback |
| Elbit Systems and SciAutonics dune buggy | about 11 km | Stuck on an embankment |
| Team DAD Toyota pickup (David Hall) | about 10 km | Hung up on a large rock |
| Golem (Ford pickup) | about 8 km | Stopped on a steep incline, navigating by GPS alone |
| TerraMax (Oshkosh) | about 2 km | Backed into brush and was shut down |
Sources for the table: Carnegie Mellon, IEEE Spectrum and Velodyne.[9][7][13]
The $1 million prize went unclaimed.[1] IEEE Spectrum estimated that DARPA spent about $13 million organizing the event, and quoted DARPA director Anthony Tether as calling it "a most important first step in a long journey."[7] The day after the race, DARPA announced a second Grand Challenge for the autumn of 2005.[2]
The 2005 Grand Challenge
DARPA doubled the prize to $2 million for the second event.[4] According to the Stanford team, 195 teams registered, 43 semifinalists took part in a National Qualification Event at the California Speedway from September 27 to October 5, 2005, and 23 vehicles raced.[4] The course both started and finished in Primm, Nevada, and DARPA gives its length as 132 miles over desert terrain near the California-Nevada state line.[4][1] Stanford received the route data, 2,935 GPS waypoints with speed limits of up to 50 mph, at about 4:10 a.m. on October 8, and its robot started at 6:35 a.m.[4]
Carnegie Mellon was the favorite. It brought two entries, a modified 1986 Humvee and a modified 1999 Hummer: an improved Sandstorm and a new vehicle, H1ghlander. Stanford had not taken part in 2004.[8] H1ghlander started first and led early on. Stanley, starting second, was paused twice by DARPA to keep its distance from the Carnegie Mellon robot, and was finally allowed to pass H1ghlander at mile 101.5, about 5 hours and 25 minutes into the race.[4] IEEE Spectrum later reported that Carnegie Mellon traced H1ghlander's loss of power, 12 years after the race, to a small electronic filter between the engine control module and the fuel injector.[8]
| Place | Vehicle | Team | Time |
|---|---|---|---|
| 1 | Stanley | Stanford Racing Team | 6:53:58 |
| 2 | Sandstorm | Carnegie Mellon Red Team | 7:05 |
| 3 | H1ghlander | Carnegie Mellon Red Team | 7:14 |
| 4 | Kat-5 | Sponsored by the Gray Insurance Company | 7:30 |
| 5 | TerraMax | Oshkosh Truck | 12:51 (over the 10-hour limit) |
Stanley's time is from the Stanford team's paper; DARPA rounds it to 6 hours, 53 minutes and IEEE Spectrum to 6 hours, 54 minutes. The other times are from IEEE Spectrum.[4][1][8] Kat-5 was named after Hurricane Katrina, which had struck the Gulf Coast about a month and a half earlier.[8]
Stanley
Stanley was built on a 2004 Volkswagen Touareg R5 TDI by a team led by Sebastian Thrun of the Stanford Artificial Intelligence Laboratory, with engineers from the Volkswagen of America Electronics Research Laboratory, Intel Research and the venture firm Mohr Davidow Ventures.[4] Its roof rack carried five SICK laser range finders, a color camera, GPS antennas and two 24 GHz radar units; six Pentium M computers in the trunk ran the software.[4] The radar was tested but not used in the race, and a stereo camera system was dropped because it added little to the lidar.[4]
The team's paper says the software "relied predominately on state-of-the-art artificial intelligence technologies, such as machine learning and probabilistic reasoning."[4] Three pieces illustrate the approach:
- State estimation. An unscented Kalman filter fused GPS, a GPS compass, an inertial measurement unit and wheel encoders to estimate the vehicle's position and orientation, and kept estimating through GPS outages.[4]
- Laser terrain mapping. A probabilistic model classified grid cells as drivable or not. Its parameters were tuned with data labeled simply by having a human drive over obstacle-free ground, and this tuning cut the false positive rate on an independent test set from 12.6% to 0.002%.[4]
- Adaptive vision. The lasers saw only about 22 m ahead, enough for about 25 mph. To drive faster, Stanley projected the patch of road the lasers had already judged drivable into the camera image, used those pixels to train a Gaussian mixture model of what road looked like, and then classified the rest of the image out toward the horizon. The paper estimates that without this vision module Stanley would have been held to 25 mph and finished in about 7 hours and 5 minutes, possibly behind Sandstorm.[4]
Stanley averaged 19.1 mph over the course, with a top speed of 38.0 mph.[4] The authors noted the limits of what had been shown: the race course was static, and Stanley could not drive in traffic.[4] Mike Montemerlo led the software effort.[12] After the race Stanley was lent to the Smithsonian's National Museum of American History in 2006 and became part of its permanent collection in 2008.[8]
The 2007 Urban Challenge
DARPA announced the Urban Challenge in April 2006. It asked vehicles to complete missions in a city setting, interacting with other moving vehicles and obeying the California Driver Handbook, with merging, passing, intersection precedence, parking and U-turns at blocked roads among the required skills.[5][6][3] Teams received a digital road map, the Road Network Description File, and aerial imagery of the site, and were then given missions as sequences of checkpoints.[6]
Selection had several stages. According to the Boss team, 89 teams registered interest; 53 were invited to demonstrate basic urban driving for DARPA; 36 were invited to the National Qualification Event at the former George Air Force Base in Victorville, California; and 11 qualified for the final.[5] The final took place on November 3, 2007, on the same base.[5][6] Besides the 11 robots, DARPA put 50 human-driven vehicles on the course as traffic.[5] The stated course length was 60 miles; the Boss team described the final event as 85 km, and Junior logged 55.96 miles over its three missions.[5][10][6]
Six vehicles finished, three of them without human intervention, and the only collision between vehicles happened at very low speed and caused no damage.[5] DARPA describes the Urban Challenge as the first time autonomous vehicles had interacted with both manned and unmanned traffic in an urban environment.[3]
| Place | Vehicle | Team | Prize |
|---|---|---|---|
| 1 | Boss (2007 Chevrolet Tahoe) | Tartan Racing (Carnegie Mellon University, General Motors and others) | $2 million |
| 2 | Junior (2006 Volkswagen Passat wagon) | Stanford Racing Team | $1 million |
| 3 | Odin (Ford Escape hybrid) | VictorTango (Virginia Tech) | $500,000 |
| Finished | Little Ben | Ben Franklin Racing Team (University of Pennsylvania and Lehigh University) | none |
| Finished | Talos | MIT | none |
| Finished | Skynet | Cornell University | none |
Sources for the table: the Boss and Junior papers, MIT Technology Review and Virginia Tech.[5][6][10][11]
Ranking was not purely on elapsed time. MIT Technology Review reported that the judges weighed driving behavior, such as violations at stop signs, as well as time.[10] DARPA's account says Boss crossed the finish line first, at 1:43 p.m., with a run time of just over four hours, and Junior followed 19 minutes later.[3] MIT Technology Review's report from the event gives a different order on the road: Junior, which had started earlier, rolled across the line first, followed shortly by Boss and Odin, and the winner was announced the next day after the judges reviewed the cars' data.[10] The Boss team gives its time as 4 hours, 10 minutes and 20 seconds, at an average of 22.5 km/h while enabled.[5] Stanford reports that Junior's run time, excluding pauses, was 4 hours, 5 minutes and 6 seconds, that organizers had paused it for a total of 1 hour, 17 minutes and 56 seconds, none of it at the team's request, and that DARPA set its adjusted total time at 4 hours, 29 minutes and 28 seconds.[6] Odin finished about 26 minutes behind Boss.[5] Each finisher had problems: Boss twice decided wrongly that it had to make a U-turn and drove an extra 2 miles, Junior repeatedly looped through one section because of a software bug, and Odin suffered a GPS error that put it partly off the road.[5]
Boss and Junior
Boss was named after Charles "Boss" Kettering, the automotive inventor. Tartan Racing drew on Carnegie Mellon, General Motors, Caterpillar, Continental and Intel, and Chris Urmson was its technical team leader.[5][12] Its software used a three-layer planner: a mission layer that chose streets, a behavioral layer that handled lane changes, intersection precedence and error recovery, and a motion planning layer that picked trajectories around obstacles.[5] The vehicle drove more than 3,000 km autonomously over 16 months of development.[5]
Junior, from Thrun's Stanford group with Volkswagen, Intel and Bosch, carried five laser range finders from IBEO, Riegl, SICK and Velodyne, an Applanix GPS-aided inertial navigation system, five Bosch radars and two Intel quad-core computers.[6] Its primary obstacle sensor was a Velodyne HDL-64E, a spinning unit with 64 lasers.[6]
That sensor came out of the Grand Challenge itself. David Hall, founder of the audio company Velodyne, entered a Toyota pickup in 2004 that relied on stereo cameras,[7] and for 2005 his team fielded a spinning 64-laser prototype, the forerunner of Velodyne's lidar products.[13] According to a 2011 Velodyne press release, the commercial HDL-64E was the obstacle detection sensor on five of the six vehicles that finished the Urban Challenge, Boss among them.[13] Boss also used a Velodyne HDL-64, and its team wrote afterward that the sensor came close to what urban driving needed but lacked angular resolution at long range and was unwieldy for commercial cars.[5]
Legacy
People and companies
The most direct line runs to Google. In an October 2010 post on Google's official blog, Thrun wrote that the company had developed cars that drive themselves, that they had logged more than 140,000 miles, and that the team included Chris Urmson, "technical team leader of the CMU team that won the 2007 Urban Challenge," Mike Montemerlo, software lead of the Stanford team that won in 2005, and Anthony Levandowski, who had built an autonomous motorcycle for the Grand Challenge.[12] Waymo dates the start of the Google self-driving car project to 2009, and the project became Waymo, a self-driving technology company under Alphabet, in 2016.[17] Thrun also worked on Google's Street View before the car project.[8] In 2017 Urmson co-founded Aurora Innovation with Sterling Anderson and Drew Bagnell, the latter also an author of the Boss paper.[14][5]
Other participants stayed closer to defense work. DARPA noted in 2014 that Oshkosh Defense had developed the TerraMax unmanned ground vehicle for the U.S. Marine Corps, and that TORC Robotics, one of the Urban Challenge finishers, was developing autonomous capabilities for Marine Corps platforms.[2] TORC had written the software for Virginia Tech's Odin with the student team.[11] IEEE Spectrum notes that Carnegie Mellon's Red Team went on to work with NASA on planetary rovers and on self-propelled agricultural harvesters.[8]
Technical influence
The events left a public engineering record. The Journal of Field Robotics published detailed papers on the winning vehicles, and Springer collected team papers in two edited volumes, The 2005 DARPA Grand Challenge (2007) and The DARPA Urban Challenge (2009), both edited by Martin Buehler, Karl Iagnemma and Sanjiv Singh.[4][5][6][15][18] Among the lessons the Boss team drew were that no off-the-shelf sensor was sufficient on its own for urban driving, and that estimating a vehicle's position relative to a prior road map could replace estimating road shape on the fly; the team named applying that approach at national scale as the next step.[5] It also listed problems the contest left unsolved, among them a richer representation of the scene and the lack of any formal way to verify that a vehicle interacting with the world behaves correctly.[5]
Stanley's authors presented the robot as an application of artificial intelligence to field robotics, and credited "the pervasive use of machine learning, both ahead and during the race" with making it robust and precise.[4]
Later DARPA challenges
DARPA says the Grand Challenge invigorated the prize challenge model of promoting innovation, and in 2014 three other DARPA challenges were building on it: the Spectrum Challenge, the DARPA Robotics Challenge and the Cyber Grand Challenge.[2] IEEE Spectrum also lists the Robotics Challenge for disaster-response robots and the Subterranean Challenge for robots that work underground among the contests that followed.[8]
Assessments
DARPA's timeline says the challenges helped create a research community that, a decade later, made fleets of autonomous vehicles look close to certain.[1] Writing in IEEE Spectrum in February 2025, in a retrospective headlined as looking back 20 years, Allison Marsh, a University of South Carolina professor, was more cautious. Her article's subtitle says the technology was "still niche" two decades after Stanley; she noted that civilian deployment had been confined to specific uses such as Waymo robotaxis in San Francisco, that robotic ground vehicles made up only 3 percent of the global armored-vehicle market as of 2023, and that the U.S. military had mostly adopted semiautonomous, operator-assisted systems instead.[8] Marsh also stressed that the 2005 result rested on a longer path of research, tracing Stanley's approach back to Minerva, a museum tour-guide robot Thrun had worked on at Carnegie Mellon in 1998 that had to learn where it could move safely among visitors.[8]
Summary of the events
| Event | Date | Location | Prize | Field | Finishers | Winner |
|---|---|---|---|---|---|---|
| Grand Challenge | 2004-03-13 | Barstow, California, to Primm, Nevada (142 mi) | $1 million (unclaimed) | 15 starters | 0 | None |
| Grand Challenge | 2005-10-08 | Loop from Primm, Nevada (132 mi) | $2 million | 23 starters | 5 | Stanley (Stanford) |
| Urban Challenge | 2007-11-03 | Former George Air Force Base, Victorville, California | $2 million, $1 million, $500,000 | 11 finalists | 6 | Boss (Tartan Racing) |
Sources: DARPA and the team papers.[1][2][3][4][5][10]
Continue reading
- DARPA (Defense Advanced Research Projects Agency)The Defense Advanced Research Projects Agency (DARPA) is the central research and development agency of the United States Department of Defense, responsible for funding high-riskMentioned 37 times above
- Autonomous vehicleAn autonomous vehicle, also known as a self-driving car, driverless car, or robotic vehicle, is a vehicle that uses artificial intelligence, sensors, and software to navigate and operate without human input.Mentioned 5 times above
- WaymoWaymo LLC is an American autonomous driving technology company and subsidiary of Alphabet Inc. that develops the Waymo Driver, a Level 4 autonomous driving system, and operates Waymo OneMentioned 3 times above
- CarsAI in cars covers the use of machine learning, computer vision, neural networks, and large language models in passenger vehicles, commercial trucks, and the systems that build and insure them.Mentioned 3 times above
- Chris UrmsonChris Urmson (born 1976) is a Canadian roboticist, engineer, and entrepreneur who is a co-founder and the chief executive officer of Aurora Innovation, a company that develops self-driving technology for…Mentioned twice above
- LiDARLiDAR is an active optical sensing method that estimates distance from transmitted laser light and its return from a target.Mentioned twice above
More from See also
More to read
References
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10Defense Advanced Research Projects Agency. "Grand Challenge". DARPA Innovation Timeline. darpa.mil/...grand-challenge
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12Defense Advanced Research Projects Agency. "The DARPA Grand Challenge: 10 Years Later". DARPA News, March 13, 2014. darpa.mil/...grand-challenge-ten-years-later
- ^1 ^2 ^3 ^4 ^5Defense Advanced Research Projects Agency. "DARPA Urban Challenge". DARPA Innovation Timeline. darpa.mil/...darpa-urban-challenge
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21Thrun, S., Montemerlo, M., Dahlkamp, H., Stavens, D., et al. "Stanley: The Robot that Won the DARPA Grand Challenge". Journal of Field Robotics 23(9): 661-692, 2006. doi:10.1002/rob.20147. robots.stanford.edu/...thrun.stanley05.pdf
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21Urmson, C., Anhalt, J., Bagnell, D., Baker, C., et al. "Autonomous Driving in Urban Environments: Boss and the Urban Challenge". Journal of Field Robotics 25(8): 425-466, 2008. doi:10.1002/rob.20255. ri.cmu.edu/...urmson_christopher_2008_1.pdf
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10Montemerlo, M., Becker, J., Bhat, S., Dahlkamp, H., et al. "Junior: The Stanford Entry in the Urban Challenge". Journal of Field Robotics 25(9): 569-597, 2008. doi:10.1002/rob.20258. robots.stanford.edu/...junior08.pdf
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9Kumagai, Jean. "Dusted: No winners in DARPA's $1 million robotic race across the Mojave Desert". IEEE Spectrum, March 2004. spectrum.ieee.org/...-race-across-the-mojavedesert
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11Marsh, Allison. "The Starting Line for Self-Driving Cars". IEEE Spectrum, February 1, 2025. spectrum.ieee.org/darpa-grand-challenge
- ^1 ^2 ^3Spice, Byron. "Carnegie Mellon's Red Team Sandstorm Robot Sets Distance Record in DARPA Grand Challenge". Carnegie Mellon University School of Computer Science News, March 15, 2004. cs.cmu.edu/...istance-record-darpa-grand-challenge
- ^1 ^2 ^3 ^4 ^5Greene, Kate. "Champion Robot Car Declared". MIT Technology Review, November 5, 2007. technologyreview.com/...champion-robot-car-declared
- ^1 ^2Virginia Tech News. "Virginia Tech team takes third place $500,000 prize in DARPA Urban Challenge". Virginia Tech, November 2007. news.vt.edu/...2007-660
- ^1 ^2 ^3 ^4Thrun, Sebastian. "What we're driving at". Official Google Blog, October 9, 2010. googleblog.blogspot.com/...what-were-driving-at
- ^1 ^2 ^3Velodyne LiDAR (press release). "Velodyne Donates LiDAR and Robotic Artifacts to Smithsonian". LIDAR Magazine, May 21, 2011. lidarmag.com/...d-robotic-artifacts-to-smithsonian
- ^Aurora Innovation. "Company". Aurora. aurora.tech/company
- ^Buehler, Martin, Iagnemma, Karl, and Singh, Sanjiv (eds.). "The 2005 DARPA Grand Challenge". Springer Tracts in Advanced Robotics, Springer, 2007. doi:10.1007/978-3-540-73429-1. doi.org/...978-3-540-73429-1
- ^Davies, Alex. "An Oral History of the 2004 Darpa Grand Challenge". Wired, August 3, 2017. wired.com/...darpa-grand-challenge-2004-oral-history
- ^Waymo. "About Waymo". Waymo. waymo.com/about
- ^Buehler, Martin, Iagnemma, Karl, and Singh, Sanjiv (eds.). "The DARPA Urban Challenge". Springer Tracts in Advanced Robotics, Springer, 2009. doi:10.1007/978-3-642-03991-1. doi.org/...978-3-642-03991-1
Improve this article
Add missing citations, update stale details, or write a clearer explanation. Anyone can edit: changes go live when you publish, and every edit is checked by an editor.
1 revision · v1 · 3,342 words · full history
Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify
Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here
Cite this page: AI Wiki. "DARPA Grand Challenge." aiwiki.ai, updated 18 Nov 2026. CC BY 4.0. https://aiwiki.ai/wiki/darpa_grand_challenge


