Autonomy. Lawrence Burns

teams would be evaluated by DARPA, competing to become one of twenty-three finalists to qualify for the actual race on October 8, 2005.

      The final day of testing was September 19. Whittaker’s culminating goal had Sandstorm and H1ghlander navigating 10 laps of a 30-mile-long course, to accumulate 300 miles in total, about double what the robots would have to do on race day. Once they achieved the distance, the team would freeze the software, store the robots and disperse to their own chosen habitats for the pre-race rest.

      By the afternoon of the nineteenth, Sandstorm was ready for the race but for a last-minute tire and oil change. Meanwhile, H1ghlander was nearing the final laps of its last test session. Following behind in AM General’s second donated Humvee was Peterson in the passenger seat and software engineer Jason Ziglar behind the wheel. Ziglar was doing his darnedest to keep up with H1ghlander, whipping the steering wheel this way and that, his foot jammed on the accelerator. With H1ghlander about to start its final lap, having already gone 270 miles, Peterson called Red in Pittsburgh, where he was handling some last-minute details. “The vehicle is driving really well,” Peterson told him. “But we’re really beating up on it.” What if something happened? Peterson recommended to Red that they call off the final lap. “It felt like we’d learned everything we were going to learn,” Peterson recalls.

      Giving up before the team had completed a goal wasn’t in Whittaker’s DNA. He made the call—finish the route. So they kept going. Moments later, H1ghlander was kicking up its usual dust cloud. From the passenger seat in the chase vehicle, Peterson couldn’t see the robot, but thanks to his laptop’s Wi-Fi connection, he could see what H1ghlander could see on the monitor. Approaching a leftward curve, the robot slowed down, the way its algorithms specified, and then accelerated into the curve. Except it swayed just a little bit to the right, off the path—and Peterson’s whole display went red. When the dust cleared, Peterson saw a dirt formation on the right side of the road that looked like the sort of thing a stunt driver would use to shoot a car up into a two-wheeled drive. In this case, the stunt jump had sent H1ghlander over on its side, and ultimately, onto its roof. The robot had caught the right wheels on the ramp at 30 mph and launched itself into the air.

      Another rollover.

      Having been through this before, the team leapt into action. No one broke down in tears over this one—Spiker was prepared. Many of the extra parts required to repair H1ghlander lay in the mechanics shed at the Nevada Automotive Test Center base. The rest, Spiker arranged to have shipped from Pittsburgh to Nevada.

      And that week’s worth of vacation everybody was supposed to go on the next day? Gone. Instead it turned into the biggest work session the Red Team had ever faced.

      Once Stanford’s AI class conducted its 8.4-mile test run, Thrun winnowed his team down to four key people. Thrun himself and Carnegie Mellon alum Mike Montemerlo were the first two. Among those who had taken his robot class, Thrun discovered a fellow German, a computer-vision expert and programming whiz named Hendrik Dahlkamp. The fourth was a grad student named David Stavens.

      A quartet was appropriate to the task because that’s how many occupants the Touareg comfortably fit. For a week at a time, Thrun and the other three would head out into the Mojave Desert and drive the trails. At first, they’d set the vehicle on a trail, watch it navigate itself, and eventually the robot would encounter something it couldn’t handle. Then someone would code a fix. As the process repeated itself dozens, and eventually hundreds, of times, the robot became sophisticated enough that it began to teach itself. In this phase, Thrun would drive Stanley through the desert, manning the controls, slowing down when the road became rough or steep, accelerating on smooth straightaways. After several days of this, Thrun would go back to the university, and Stanley, working overnight, would retroactively look at the data to engage in its own learning. Confronted with this terrain, Stanley would think, Sebastian chose to drive here—and I will do the same. “The robot would basically spend the night sorting through the data and bring order from chaos,” Thrun said.

      Stavens’s contribution was an algorithm that taught the robot how to regulate its speed. The roads Stanley drove in the Mojave featured rain ruts, puddles and potholes. Blasting through this sort of terrain at speed would have shaken the car to pieces. So Stavens wrote a program that regulated Stanley’s progress based on vibrations felt by the robot’s sensors, as well as the grade and width of the road. With the program loaded into the robot, Mike Montemerlo drove Stanley to create data the program could then analyze to develop rules that would guide its behavior.

      The problem here was that Montemerlo was too conservative. He’s incredibly detail oriented. A nice way of putting it is risk-averse. “We used to put stickers on his windows,” Thrun recalls. “So Mike couldn’t see how fast we were going.” Montemerlo had once protested to the team members that he would never get in a self-driving car that went more than 5 mph. Driving Stanley, Montemerlo would creep around the desert, easing up hills, wandering over rubble and stones. Then, once the vehicle was at home, the machine learning algorithm would look at the way Montemerlo drove and create rules that would guide Stanley in the future. Accustomed to high-speed driving on Germany’s Autobahn, Thrun didn’t like how slowly Stanley progressed once it had crunched Montemerlo’s data. So one week, when Montemerlo went away on vacation, Thrun set Stanley to go 20 percent faster.

      Then came the day in 2005 when Thrun received an unexpected visitor at his Stanford office. He looked up and saw a figure in the doorway. The figure came forward and introduced himself: “Hi,” the man said. “I’m Larry Page.”

      Thrun knew who Page was, of course. What surprised him was how interested Page was in the project. “Larry’s always been a robotics enthusiast,” Thrun says, explaining that had Page not started Google, he might have pursued a PhD in robotics. Page was fascinated with Thrun’s project. He had about a million questions. He wanted to see how real the technology was—how close are driverless cars? A century? Decades? A couple of years? What did Thrun think? In fact, Page was so interested that he told Thrun he planned to attend the second Grand Challenge. Through their shared enthusiasm for driverless cars, Thrun and Page developed a friendship that deepened, because the two men both relished taking on tasks that everyone else dismissed as impossible. Thrun had no idea, at that point, that Page would change the course of his life.

      At 4:30 A.M. on October 8, 2005, the day of the race, DARPA officials provided a Red Team member a USB key featuring a computer file of 2,935 waypoints—the course of the second Grand Challenge. The whole of the route totaled 132 miles, starting and ending in Primm, Nevada.

      The next bit bore many similarities to the first race. The team member sprinted to Red Team’s command center. Another member loaded the route network definition file onto Red Team’s shared hard drive. A computer program analyzed the waypoints and added thousands more, so a route originally specified every eighty yards now featured a dot every yard or two. Next, the route was divided up among team members to go over. The pre-planning team went through each part of the route to ensure the new waypoints kept Sandstorm and H1ghlander on navigable road.

      In the anxious moments that passed while the pre-planning team worked, Whittaker, Urmson and Peterson discussed strategy. The experience of the first race eighteen months before was fresh in everyone’s minds. That time, they’d gone for speed. And perhaps they’d pushed Sandstorm beyond what was good for it.

      So the three decided Red Team should take a tortoise-and-hare approach with its two vehicles. One of the vehicles would take it easy, going so slowly that it would be certain to finish the race. This way, in the event that no one else finished, at least Red Team would have a vehicle that crossed the finish line.

      Sandstorm consistently came in 10 percent slower than H1ghlander—a symptom, the engineers thought, of the way the electronics box floated, which made it difficult for the robot to pinpoint exactly where it was. So H1ghlander would be Red Team’s hare, while Sandstorm was the tortoise.

      In terrain