“Self-driving cars equipped with electric motors and deployed in a transportation service model are poised to become the biggest thing to hit the automobile industry since the invention of the automobile itself,” wrote Lawrence D. Burns, former vice president of Research and Development for General Motors, in his 2018 book, “Autonomy: The Quest to Build the Driverless Car—And How It Will Reshape Our World.”
For years, the Defense Advanced Research Projects Agency (DARPA) — the U.S. government’s lab credited with the invention of the internet and drone technologies — was looking to develop self-driving vehicles to deploy in war zones.
In the aftermath of the terrorist attacks of Sept. 11, 2001, the military stepped up the urgency, as the U.S. was losing soldiers in Afghanistan and Iraq because of explosive devices planted under the roads.
Now, DARPA was charged with developing the robot vehicles that could travel those roads to deliver supplies and conduct reconnaissance missions without risking precious human lives.
Unfortunately, the agency found the problem too difficult for their engineers and the military contractors of the time. They didn’t have the expertise because no one had done it before.
So DARPA director Tony Tether came up with a new approach: DARPA would launch a challenge, a series of races, for robot cars.
The DARPA Grand Challenge was the first long-distance competition for driverless cars in the world. The ultimate goal was making one-third of ground military forces autonomous by 2015.
First Grand Challenge Across the Desert
To kick-start the competition, DARPA announced the first race in 2002. It was open to teams from all over the world as long as one of their members was an American citizen. A $1 million prize was waiting for the team whose vehicle arrived first after a 150-mile journey through the desert in less than 10 hours.
DARPA held the race on the morning of March 13, 2004, in the Mojave Desert. One-hundred-and-six teams applied; after a series of qualifying events, 15 teams made it to the starting line in tiny Barstow, California, hoping to make it to the finish line in Primm, Nevada.
To ensure that the race was not pre-programmed into the vehicles in advance, DARPA did not release the final route — a file with a series of 2,500 GPS waypoints — until a few hours before the start.
Not one vehicle could arrive at the finish line. Most of the cars were disabled or ran into obstacles within the first 20 miles of the race.
However, of those that competed that day, two stood out: the Stanford Racing Team and the Carnegie Mellon Red Team. Neither could finish, but their vehicles showed that they could make driverless cars — they just needed more time.
Tony Tether was very disappointed, mainly because the media widely covered the race and several TV stations broadcasted the failure, some portraying DARPA as an example of government bureaucracy. Then Tether took the stage and announced that DARPA would hold a second race, in about a year, with a prize of $2 million.
The Grand Challenge second race took place on Oct. 8, 2005. Five vehicles successfully completed the 132-mile course. The winner was “Stanley,” the car from the Stanford Racing Team. The Carnegie Mellon Red Team took second and third places, with “Sandstorm,” which competed in the first race, and “H1ghlander.”
The third and final Grand Challenge race, commonly known as the DARPA Urban Challenge, took place on Nov. 3, 2007, at the site of the now-closed George Air Force Base in Victorville, California. The winner of the $2 million prize was Tartan Racing, a collaborative effort by Carnegie Mellon University (CMU) and General Motors, with their vehicle, “Boss,” a heavily modified 2007 Chevrolet Tahoe.
This time, Stanford came in second with “Junior,” a modified 2006 Volkswagen Passat.
The Stanford Racing Team
Sebastian Thrun was entertaining the idea of self-driving cars for many years. Born and raised in Germany, he was fascinated with the power and performance of German cars. Things changed in 1986, when he was 18, when his best friend died in a car crash because the driver, another friend, was going too fast on his new Audi Quattro.
As a student at the University of Bonn, Thrun developed several autonomous robotic systems that earned him international recognition. At the time, Thrun was convinced that self-driving cars would soon make transportation safer, avoiding crashes like the one that took his friend’s life.
In 1998, he became an assistant professor and co-director of the Robot Learning Laboratory at Carnegie Mellon University. In July 2003, Thrun left Carnegie Mellon for Stanford University, soon after the first DARPA Grand Challenge was announced. Before accepting the new position, he asked Red Whittaker, the leader of the CMU robotics department, to join the team developing the vehicle for the DARPA race. Whittaker declined. After moving to California, Thrun joined the Stanford Racing Team.
On Oct. 8, 2005, the Stanford Racing Team won $2 million for being the first team to complete the 132-mile DARPA Grand Challenge course in California’s Mojave Desert. Their robot car, “Stanley,” finished in just under 6 hours and 54 minutes and averaged over 19 mph on the course.
Carnegie Mellon’s Red Team
The other team came from Pittsburgh’s Carnegie Mellon University, which has one of the top robotic departments in the world. William “Red” Whittaker led Carnegie Mellon’s Red Team.
Whittaker is considered one of the leading roboticists in the world. He and his colleagues at CMU have created robots to climb into the craters of active volcanoes, inspect and perform repairs in damaged nuclear reactors, and explore the terrain of Mars.
The technical lead of the team was Chris Urmson, another professor of robotics at CMU. The “Red” team developed three vehicles for the DARPA races: Sandstorm, H1ghlander, and Boss.
Google’s Page wanted to develop self-driving cars
Two years after the third Grand Challenge, Google co-founder Larry Page called Thrun, wanting to turn the experience of the DARPA races into a product for the masses.
When Page first approached Thrun about building a self-driving car that people could use on the real roads, Thrun told him it couldn’t be done.
But Page had a vision, and he would not abandon his quest for an autonomous vehicle.
Thrun recalled that a short time later, Page came back to him and said, “OK, you say it can’t be done. You’re the expert. I trust you. So I can explain to Sergey [Brin] why it can’t be done, can you give me a technical reason why it can’t be done?”
Finally, Thrun accepted Page’s offer and, in 2009, started Project Chauffeur, which began as the Google self-driving car project.
The Google 101,000-Mile Challenge
To develop the technology for Google’s self-driving car, Thrun called Urmson and offered him the position of chief technical officer of the project.
To encourage the team to build a vehicle, and its systems, to drive on any public road, Page created two challenges, with big cash rewards for the entire team: a 1,000-mile challenge to show that Project Chauffeur’s car could drive in several situations, including highways and the streets of San Francisco, and another 100,000-mile challenge to show that driverless cars could be a reality in a few years.
By the middle of 2011, Project Chauffeur engineers completed the two challenges.
In 2016, the Google self-driving car project became Waymo, a “spinoff under Alphabet as a self-driving technology company with a mission to make it safe and easy for people and things to move around.”
Urmson led Google’s self-driving car project for nearly eight years. Under his leadership, Google vehicles accumulated 1.8 million miles of test driving.
In 2018, Waymo One, the first fully self-driving vehicle taxi service, began in Phoenix, Arizona.
From Waymo to Aurora
In 2016, after finishing development of the production-ready version of Waymo’s self-driving technology, Urmson left Google to start Aurora Innovation, a startup backed by Amazon, aiming to provide the full-stack solution for self-driving vehicles.
Urmson believes that in 20 years, we’ll see much of the transportation infrastructure move over to automation.
