Communications of the ACM
On A Mission To Build The ncrashable Car
Ryan Eustice's interest in self-driving cars began 12,500 feet below the surface of the Atlantic Ocean. As a Ph.D. student in the joint MIT-Woods Hole Oceanographic Institution Program, Eustice focused on creating technologies for underwater vehicles to map and understand their environments.
"That's how I got into this line of work originally," explains Eustice, who is currently senior vice president of automated driving at Toyota Research Institute and associate professor at the University of Michigan. "From an engineering perspective, the focus would be on helping the robot better navigate and understand its surroundings."
At MIT and Woods Hole, Eustice would deploy robots on field cruises to take pictures or make a map of the seafloor using cameras, sonar, or LIDAR—an acronym for light detection and ranging. That map would then be used by a geologist or marine biologist for research purposes. A breakthrough in his career came in 2004, when he had the opportunity to send one of his robots to the site of the Titanic wreck, 12,500 feet below the water's surface off the coast of Newfoundland. "I was able to produce a very accurate reconstruction and map of the wreckage using the downward-looking camera imagery the robot collected."
Professor John Leonard, who served as Eustice's co-advisor while he was a Ph.D. student, found Eustice's work ethic infectious. One day, Leonard was facing a deadline to write some paragraphs for the literature review of one of Eustice's important papers. "I said I would try to write a few paragraphs—and Ryan said 'Do or do not, there is no try,'" recalls Leonard, in reference to the famous quip by Star Wars character Yoda. "I stayed late that night and wrote the paragraphs before going home."
After receiving his Ph.D., Eustice made his way back to his home-state of Michigan. He was offered a faculty position in the University of Michigan's Department of Naval Architecture and Marine Engineering, where he continued his work on underwater robotics. "I've been using some of the same technology that went into mapping the Titanic," Eustice says. "I'm looking at how robots can be deployed near naval ships so they can do inspection tasks or map the below-water portion of the hull."
Shortly after arriving in Michigan, Eustice was asked to apply the technology he was building for underwater vehicles to cars. In 2007, the U.S. Defense Advanced Research Projects Agency (DARPA) announced their Urban Challenge to build an autonomous vehicle that can drive and navigate everyday traffic scenarios. The team from Ford Motor Co., a few towns over from Ann Arbor, were looking for someone with expertise in mapping, navigation, and LIDAR technologies. Eustice fit the bill.
The Ford Motor team finished as finalists in the 2007 DARPA Urban Challenge. Eustice continued to work with them for nearly a decade, before joining Toyota Research Institute in 2016. At Toyota, Eustice leads a team developing a sensor-rich car built around artificial intelligence. Like many companies around the world, part of the team's research is focused on what they call "chauffeur mode"—where the human is the passenger and the car is fully capable of driving itself.
But according to Eustice, this kind of automation has multiple applications. "We are working on a technology stack that gets us to a full automation scenario, but at the same time we see a tremendous opportunity to use that technology in a different way," says Eustice. "Fundamentally, we want to build an un-crashable car."
With fully autonomous vehicles, a human driver has to be somewhat alert since the car is unable to handle all individually rare but collectively common scenarios that happen in day-to-day driving—a mattress flipping off a car in front of you, or a crossing guard motioning for you to stop, for example. In those situations, human drivers need to remain alert in the event they have to take over steering control. Humans are expected to watch the AI.
But Eustice and his team are developing technologies that flip that equation. "With 'guardian mode' we say, 'Well let's imagine a system where we have AI guard the human,'" Eustice says. It's a subtle change but has profound ramifications that can augment the human driver.
Eustice and his team have outfitted test cars with 360-degree sensing around the vehicle, using similar technologies he worked with as a graduate student at MIT. But instead of mapping oceanic environments, he now has one particularly lofty ambition, "to develop a car that is incapable of causing a crash."
Originally published on MIT News.
Reprinted with permission of MIT News.
No entries found