On March 13, 2004, a gaggle of engineers and a few thousand spectators congregated outside a California dive bar to watch 15 self-driving cars speed across the Mojave Desert in the first-ever Darpa Grand Challenge. (That’s the Defense Advanced Research Projects Agency, the Pentagon’s skunkworks arm.) Before the start of the race, which marked the first big push toward a fully autonomous vehicle, the grounds surrounding the bar teemed with sweaty, stressed, sleep-deprived geeks, desperately tinkering with their motley assortment of driverless Frankencars: SUVs, dune buggies, monster trucks, even a motorcycle. After the race, they left behind a vehicular graveyard littered with smashed fence posts, messes of barbed wire, and at least one empty fire extinguisher.
What happened in between—the rush out of the starter gate, the switchbacks across the rocky terrain, the many, many crashes—didn’t just hint at the possibilities and potential limitations of autonomous vehicles that auto and tech companies are facing and that consumers will experience in the coming years as driverless vehicles swarm the roads. It created the self-driving community as we know it today, the men and women in too-big polo shirts who would go on to dominate an automotive revolution.
I. The Challenge
In 2001, eager to keep soldiers away from harm in combat zones, the US Congress demanded that a third of the military’s ground combat vehicles be uncrewed by 2015. But defense industry stalwarts weren’t innovating quickly enough on the sensor and computing technologies that would enable autonomous driving. So in February 2003, Tony Tether, the director of Darpa, announced a 142-mile race for self-driving cars , open to anyone, with a $1 million prize for whoever finished its course the fastest. Tether held a kickoff event at the Petersen Automotive Museum in Los Angeles for prospective racers.
TONY TETHER (Darpa): I thought, “Hell, maybe we’ll get five or 10 people to show up.”
JOSE NEGRON (Darpa): The defense contractors that Darpa had been working with, they got stuck in a mind meld. They were thinking step by step, evolutionary, and they weren’t progressing. So we needed a revolutionary approach, a leap forward, and I kept telling Tony there would be hundreds of people interested in joining us, people who do not do DOD business. People who work late at night, in their garages and bedrooms, because they love what they do. For them, failure is OK as long as they learn something. But he didn’t believe me until he arrived at the Petersen for the kickoff.
TETHER: We were going to open the doors at 9 in the morning, and by the time I got there at 8:30 there was a line of people four abreast going all the way around the block. These were just ordinary people who dreamed of having a car that was driverless.
NEGRON: I told them this was gonna be a 142-mile challenge, in the desert. You’ll be going through switchbacks, up and down, paths as narrow as 10 feet, and the vehicle’s gonna have to sense its way through the course. They had a year to figure it out. One guy asked, “Why are you making the race so hard?” I said, “That’s what makes it the Grand Challenge.”
TETHER: Before all this, Darpa’s self-driving programs were working to minimize human intervention. That was the wrong criteria. We were trying to get it so a human never has to be involved.
JOSEPH BEBEL (Team Palos Verdes High School): My mom is an engineer and heard about the challenge. I was looking for new and interesting things to work on, so we brought the challenge to my high school as an extracurricular project.
DAVID HALL (Team DAD—Digital Auto Drive): We were appearing on BattleBots and Robot Wars around that time, which was mostly an excuse for advertising our Velodyne loudspeaker. Then, when this Darpa thing came about, we jumped on it as an opportunity to show off some computer skills and do something fun.
MELANIE DUMAS (Team Axion Racing): When we went to the Petersen event, we found an angel investor who gave us a bit of starter cash. We named our vehicle Spirit of Kosrae, for the island in Micronesia from which our investor was importing bottled water to sell.
JIM MCBRIDE (Ford Engineer): At the time I was working on automotive safety at Ford, and when I heard about the race I volunteered to work at it to see how remote sensing had progressed and to see if there was anything that could be carried over into Ford’s safety program. But most people thought that the notion of a car driving itself was for crackpots.
SAL FISH (Course Designer): I get a phone call from a guy saying, “I’m with Darpa, and we’re putting together an event with autonomous vehicles. Would you help us make a race course?” I made the course. It had rocks on it, left turns, right turns, dips, gullies, cactus all around the place. A drop-off on one side and then 5 miles later a drop-off on the other side. Barbed wire fences, animals that could come out of nowhere, train tracks. Anything and everything that a vehicle would encounter going through the desert.
NEGRON: I hired off-road champions to drive the route. To a man they said, “This is tough, even for us.”
TETHER: We got 106 team applications. We made them write technical papers, explaining what they were going to do. To cut it down, we’d go visit them. One guy’s wife came to the door and asked, “Are you the guys who have got my husband trying to mortgage this house to build his car?”
BEBEL: The basic nature of the challenge was, follow the course they designed using GPS and sensors to avoid obstacles. We wanted to build something that, at least in theory, could complete the race. Our car was a pretty standard Acura SUV with a laser range-finding sensor on the front bumper. The laser created a one-dimensional picture of the road in front of you—if there was an obstacle, it could tell you where it was.
DUMAS: We had a Jeep Grand Cherokee, and we tried to keep it looking as normal as possible so that people wouldn’t think autonomy was so futuristic or out of reach. We also put a couple of surfboards on the roof, just to stand out. We ran parallel sets of algorithms to guide the vehicle. One took GPS data, one took camera input, one took laser range-finding data, and they all fed into a voting system, which decided where the car was and where it needed to go.
ALBERTO BROGGi (Team TerraMax): Our team specialized in machine vision, so we put three cameras on top of our vehicle—a bright-green, 3-axle military truck—along with two lidar sensors to spot obstacles ahead.
CHRIS URMSON (Red Team, Carnegie Mellon): We realized that, at the speed we needed to drive to finish the race in time, our Humvee wouldn’t be able to see that far ahead. So we gathered satellite imagery of the area. Whatever route Darpa gave us, we’d use those maps to figure out the best path, and our Humvee would use laser sensing to see the shape of the road.
TETHER: Anthony Levandowski was there, the fellow who’s alleged to have taken data from Google to Uber. He had a self-balancing motorcycle. It did a great job in the qualifying rounds.
NEGRON: Levandowski says motorcycles always win off-road races because of speed. And this was a speed event. That kid is a great engineer. He’s in a slight bit of trouble now. [Levandowski, the recently fired head of self-driving at Uber, declined to comment for this story.]
II. The Race
As race day approached, Darpa selected the 25 teams that had shown the most impressive progress and invited them to a qualifying round at California Speedway in Fontana, where they underwent safety inspections and technical tests. Fifteen teams performed well enough to advance. On March 13, 2004, the teams assembled in Barstow, two hours northeast of Los Angeles, to begin the 142-mile drive through California’s Mojave Desert, to Primm, just over the Nevada border.
FISH: We started at a saloon called the Slash X. A bloody shit-kicking, cowboy-type place.
RED WHITTAKER (Red Team, Carnegie Mellon): It was a cold desert morning, and everything from a motorcycle to an Oshkosh giant military truck showed up.
FISH: My God, these vehicles were something out of Mad Max.
HALL: It was like being at Woodstock for nerds.
NEGRON: Tony’s idea was not to tell anybody where we were going until two hours before the event, then give them a CD with all the waypoints.
TETHER: If we’d given out the coordinates weeks before, someone could have gone out and run the course. This way, they had to prepare their vehicle for everything. It added a little mystique, too.
WHITTAKER: We had a trailer, like a command center, and once we got the disc we went to work. Very quickly the vehicle team was running through everything from the fuel to the engine to the electronics, checking that everything was working.
HALL: You drive your car up to the line and you hop out of it and press the button, and it goes off by itself. It’s the damnedest thing you’ve ever seen.
WHITTAKER: Our vehicle, Sandstorm, was first off the line, so all that mattered was staying ahead.
TETHER: I’m sitting with this four-star general in the grandstand, and when Sandstorm goes off he just went, “Jesus.”
NEGRON: Sandstorm was doing 40 miles an hour.
TETHER: It was really tough terrain. The start was pretty flat, then they had to go up a high hill, with a lot of switchbacks.
NEGRON: The first few vehicles went out without a hitch. The back half of the field ... had some issues.
DUMAS: Our Jeep Cherokee came out of the chute, and it made the first turn. And then on the second turn, it made a U-turn and headed back to the starting line. I think it saw the gate as too narrow, and the sensors decided to send the car back to where it came from. We made it 20 feet. That was a little devastating.
TETHER: Levandowski’s two-wheeler, he got so excited, he forgot to throw the switch for the stabilization and it fell right over. It was too bad. I thought it had a shot at being faster than the four-wheelers. But it didn’t go a foot. Another little car went up a berm and flipped over. Another had a GPS problem—it tried to go through a barbed wire fence and got tangled up.
BEBEL: We’d been having a problem with the steering on our Acura, which we called Doom Buggy, for weeks, and we couldn’t test our final fix before the race. It failed, and Doom Buggy couldn’t turn. The race officials let it run 50 yards or so until it hit the concrete sidewall.
BROGGI: We put together all the Oshkosh’s software at the last minute, and it couldn’t discriminate between different types of obstacles. Our laser scanners picked out some brush and decided it was an unmovable obstacle.
TETHER: The Oshkosh truck backed up, and it saw another tumbleweed in the back. So it just went back and forth. Fourteen tons, taken down by two tumbleweeds.
BROGGI: We had no way to move around it. Almost a year of working on the project and we were finished in minutes.
SEBASTIAN THRUN (Former head of Google’s self-driving project): These vehicles didn’t fail because they weren’t rugged enough. They failed because they didn’t take in enough environmental information. None of them saw anything.
By the time the last of the 15 vehicles left the starting gate, the land around the Slash X saloon was a robotic graveyard. The race organizers and observers pinned their hopes on the four vehicles that were still driving an hour after departure, especially David Hall’s Velodyne pickup and Carnegie Mellon’s Sandstorm, last seen ripping through the desert flats and headed for the course’s marquee challenge: a winding road up and over a steep hill called Daggett Ridge.
HALL: There was this one hill you had to get over, at mile seven. And if you made it over that hill, it was absolutely straight, flat all the way to the finish line.
FISH: I was going, “My God, they’re gonna do it.”
WHITTAKER: The important thing that we had to avoid on that hill was going off-center. But a week before qualifying, we rolled our Humvee and damaged the sensors on our roof. It had the effect of the sensors looking a little right, and driving the vehicle a little left.
TETHER: Sandstorm got straddled.
WHITTAKER: We went a touch too far over the edge, and that’s all it took. The tire just spun and spun until it burned. Smoke was pouring off of it. The race officials hit the emergency kill, and the race was over for Sandstorm. I thought, “Oh hell, it’s a robot, I’m not going to get emotional about it.” I just made sure the team was OK, and I told them I appreciated the great thing they’d just done.
TETHER: We paused the race. David Hall’s vehicle was about half a mile behind, still going. But when the race was paused, it rolled up against a rock.
HALL: It was there for an hour or two while they were trying to get Sandstorm out of the way. We were back at our tent, waiting for information from them. We had no idea what the hell was happening.
TETHER: We cleaned up the track and started Hall’s car up again.
HALL: When they started it back up again, it wouldn’t jump over this rock underneath the front tire.
TETHER: And that was the end of it. I got in the helicopter and flew to Primm, where all the TV crews were waiting to see the cars cross the finish line. It was 11 o’clock in the morning, and they said, “How’s it going?” I said, “Well, it’s over. One car went 7.4 miles, it started smoking, blah blah blah. It’s over.” A reporter asked, “Well, what are you gonna do?” I said, “We’re gonna do it again, and this time it’s going to be a $2 million prize.” It was so successful and yet so not successful, I had to do it again.
III. the Aftermath
Despite the slap-stick failures of the first race, most of the participants—and many onlookers—saw the same possibilities that Tether saw when he announced the next race.
TETHER: I was disappointed, but it was a spectacular thing, and the people involved understood—this had never been done before.
THRUN: None of what is happening in self-driving today would have happened without the original challenge—it created a new community. They were all newcomers, and the innovation doesn’t come from the core of the field itself but from the outside. The experts are usually the lowest performers, because they’re totally bound in their way of thinking. Very few self-driving car people knew anything about machine learning at the time, for example. David Hall’s Velodyne lidar development was triggered by the Grand Challenge.
HALL: At the first challenge the sensors were all unreliable—they would cause the vehicle to stop or veer off course. So a year after the first Grand Challenge, I started working on developing a new lidar sensor, with a bunch of lasers giving me a 360-degree, 3-D view. Darpa convinced me to build these sensors and sell them to the other teams for the third challenge.
THRUN: Around 2008, Larry Page convinced me to build Google’s self-driving car team. I met the best people in the field from the Darpa challenges. I hired Chris Urmson and Anthony Levandowski.
URMSON: There was an incredible sense of camaraderie, and that community fostered the folks who are leading a lot of the technology today.
TETHER: And that all comes from the fact that in 2004, some crazy-ass people decided to have a challenge.
the oral historians, Then and Now
Then: Director of Darpa (2001–2009)
Now: Private technical consultant
Then: CEO of Velodyne, a loudspeaker company
Now: CEO of Velodyne, Now making lidar laser sensors
Then: CEO of Score International, which organized desert races like the Baja 1000
Now: Retired
Then: Robotics researcher at Carnegie Mellon
Now: Same
Then: Darpa’s day-to-day manager for the Grand Challenge
Now: Cyberwarfare consultant
Then: Voice- recognition engineer
Now: Program manager on Google’s security team
Then: Lead on Team TerraMax
Now: General manager of VisLab, a computer vision developer
Then: Machine-learning researcher at Stanford
Now: Head of online university Udacity
Then: Student at Palos Verdes High School
Now: PhD student in computer science at USC
Then: Engineer in Ford’s safety department
Now: Autonomous vehicle specialist at Ford
Then: Robotics student at Carnegie Mellon
Now: Head of Aurora Innovation, a self-driving startup
Alex Davies (@adavies47) is a senior associate editor at WIRED.
This article appears in the August issue. Subscribe now.
