In the 1880s, the carriage lost its horse. Now, thanks to a car named Boss, the automobile could be about to lose its driver. This weekend, Boss, a Chevrolet Tahoe fitted with sensors and computers by a team of engineers from Carnegie Mellon University, won the most famous of robotic races: the Urban Challenge. With no human assistance, the vehicles competing in the race had to safely and quickly navigate city streets while staying in their lanes and avoiding other moving and parked cars. With the win, the Carnegie Mellon team, called Tartan Racing, takes home a $2 million prize from the U.S. Defense Advanced Research Projects Agency (DARPA), the organization that sponsored the race. The $1 million second-place prize went to Junior, Stanford University’s robot; Odin, Virginia Tech’s bot, came in third, winning $500,000.
The Urban Challenge is the third in a series of autonomous-vehicle competitions, designed to spur robotics innovation and inspire the next generation of engineers. In 2004, DARPA held the first race, the Grand Challenge, in the Mojave Desert. The race course was a 150-mile stretch of desert road, but the farthest any of the driverless cars got was about seven miles. In 2005, the second Grand Challenge was far more successful: five cars finished, and the prize went to Stanford’s car. Carnegie Mellon came in a close second.
This year’s race was far more complex than the previous two. The grounds of the former George Air Force Base in Victorville, CA, served as a mock city that the robots had to navigate. The course consisted of 60 miles of roads and parking lots and took about six hours to complete. The whole time, the robotic cars needed to obey traffic laws and avoid both cars driven by professional stunt drivers and the other robots on the course.
In the early morning on Saturday, with the sun rising behind them, 11 cars were set in motion in front of a crowd of thousands. The cars’ routes had been loaded into their onboard computers as series of Global Positioning System coordinates. Virginia Tech’s Odin was the first to drive away, and people cheered as the steering wheel rotated on its own, guiding the car through its first two turns, onto the course. Odin was followed, at intervals of about five minutes, by Junior, Little Ben, from Ben Franklin Racing, Talos of MIT, Terramax, the 12-ton truck from Team Oshkosh, Skynet from Cornell, AnnieWay from Team AnnieWay, the Ford Truck from Intelligent Vehicle Systems, Boss from Tartan Racing, Caroline from CarOlo, and Knight Rider from the University of Central Florida.
Within the first three hours, five of the teams had been eliminated. The Oshkosh truck, for instance, nearly hit a building, and AnnieWay halted at the entrance to a traffic circle for too long. In the end, six teams–Stanford, Cornell, Carnegie Mellon, MIT, Ben Franklin, and Virginia Tech–finished the race. (See pre-race interviews with members of these teams here.)
To the casual observer, the cars weren’t doing anything special, and over time, it was easy to forget that they had no drivers. Boss, for instance, came to a stop at an intersection and started to move forward, only to spot a car coming and back up to let it pass. During the race, there were a few traffic jams caused by cars’ stopping for too long at intersections, presumably to “think” about the best course of action. There was a flurry of excitement, however, during the fifth hour, when Cornell’s Skynet and MIT’s Talos collided. Skynet had stopped in the middle of the road. Talos approached, stopped behind it, and, determining that it was a stationary object, decided to pass it. As Talos curved back into the lane in front of Skynet, Cornell’s bot started up again, hitting Talos but not damaging either car.
Just under six hours after the race started, Junior rolled across the finish line, followed shortly thereafter by Boss and Odin. Little Ben, Talos, and Skynet came in within the hour. The final winner, however, was determined by a combination of race finish time and style on the course. If, for instance, a bot consistently rolled through stop signs but finished ahead of the pack, it most likely wouldn’t have won. On Sunday, after the DARPA judges sifted through the cars’ data and score sheets and watched video footage of each car’s activity, they announced that Boss had won the competition.
“A win’s always better than second or third,” said William “Red” Whittaker, leader of Tartan Racing. “It’s something we expected of Boss. It’s fast, it’s clean, and it’s perfected for driving in this race.”
At a press conference after the awards were presented, Whittaker said that he’d like to see more competitions like the Urban Challenge, and he’d like them to push the technology even further by subjecting the vehicles to more extreme conditions for a 24-hour period. “I’d like to see a race through the Rockies, through snow, rain, and fog,” he said. “See where we get in a day.”
Stanford’s team leader, Sebastian Thrun, said that he’d be interested in seeing robotic cars get better at more-difficult tasks, such as cornering and driving at high speeds. The vehicles in the Urban Challenge were able to handle mundane driving conditions, but to make consumers’ cars safer, digital technologies must be able to react to the sudden and the unexpected. Thrun proposed a man-versus-machine race that required more-difficult maneuvers at higher speeds. Charles Reinholtz, the leader of Virginia Tech’s team, was interested in seeing a competition in which all the vehicles communicated with each other, constantly sending and receiving data about their locations, a feature that he expects to see in autonomous vehicles in the future. The consensus was that within the next few years, technology deployed in the Urban Challenge will make its way into vehicles used for farming, mining, and exploring space. Completely autonomous consumer cars are more likely at least a decade away.
It’s unclear whether or not there will be another DARPA-sponsored robotic-car race, however. Tony Tether, the director of the agency, said that the races had already served an important purpose: dispelling the notion that it was impossible to build a car that could autonomously drive with traffic on city streets while obeying the rules of the road. “Once you show that something can be done,” he said, “then other people come out of the woodwork and say, ‘Hey, I can do better than that.’”
This new data poisoning tool lets artists fight back against generative AI
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
Rogue superintelligence and merging with machines: Inside the mind of OpenAI’s chief scientist
An exclusive conversation with Ilya Sutskever on his fears for the future of AI and why they’ve made him change the focus of his life’s work.
Data analytics reveal real business value
Sophisticated analytics tools mine insights from data, optimizing operational processes across the enterprise.
Driving companywide efficiencies with AI
Advanced AI and ML capabilities revolutionize how administrative and operations tasks are done.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.