View from a Visionary

January 1, 2019

Lindsay Brooke

“Over the last fifteen years of development on autonomous vehicles, if there is one figure who has been there, on the ground, getting his palms filthy with engine grease, breathing carbon monoxide exhaust and burning himself with electrical solder to solve each little problem as it comes up, it is Chris Urmson.” — from the 2018 book, Autonomy, by former GM R&D chief Lawrence Burns.

When a blueprint for “the 21st century mobility engineer” finally emerges — defined by expertise in computer engineering, robotics, and the ability to sling a mean soldering gun — Chris Urmson may well be its model. Fifteen years ago while a Ph.D candidate at Carnegie Mellon University, he assisted a NASA team test a robot for Mars exploration. Today he’s CEO and co-founder of Aurora, a new company developing self-driving technology, and aiming to deliver its benefits safely, quickly, and broadly. As part of that Aurora is working with Volkswagen, Hyundai and Byton, the Chinese luxury self-driving-EV startup.

Along the way Urmson, 42, directed the CMU team that won the grueling 2007 DARPA Urban Challenge. That event effectively kicked off the race to develop commercial AVs. Later he became CTO of Google’s pioneering self-driving vehicle project that’s now the Waymo juggernaut. As Burns also noted, Urmson “was programmed with an engineer’s duty to seek out the coolest and most interesting projects that could change the lives of most people.”

Developing safe self-driving vehicles requires “understanding what the other actors on the road are doing — where they are in the moments, and what they’re going to do over the next five to 10 seconds,” Urmson told the audience of The Financial Times’ “Future of the Car” event last November in Detroit. And more than 99 percent of the time, the other cars are going to stay in their lane.

But it’s that fraction of a percent, he noted, where the other vehicle makes an unexpected U-turn in front of you, or a pedestrian who is transfixed on their phone steps into the road, that presents the most daunting technical challenge. To earn the full, abiding trust of the public and regulators, AVs must have the capability to reliably sense and react to an infinite range of driving scenarios.

“Much of what Machine Learning is about is picking up the ‘norms,’” Urmson explained. “So if you start to naively train a machine learning system, it’s going to predict basically that all the cars will continue going down their lane because so much of the time that’s what they do. The vehicle is starting to slide to its left and a lane change is about to occur, or the pedestrian isn’t slowing as they approach the intersection — so that means they’re about to step into the road. How do you ‘pull’ that out of the signal and translate it into safe action?”

The technical problem is “counter-intuitively hard,” he offered, “because we drive vehicles every day and are used to that experience, and yet dealing with the subtleties of human error on the road is difficult.”

The challenge inspired Urmson in 2017 to found Aurora, with Sterling Anderson, formerly head of Tesla’s Autopilot development, and Drew Bagnell, who led autonomy research at Uber’s Advanced Technology Group. The company “sits at the intersection of rigorous engineering and applied machine learning,” developing the Aurora Driver, its self-driving system which includes the hardware, software and data services for SAE Level 4 and 5 driving. It is working with vehicle OEMs to design the Aurora Driver so it can seamlessly integrate into vehicle platforms.

Aurora’s expertise in artificial intelligence and machine learning is a competitive advantage, the company believes.

Headquartered in both Palo Alto and Pittsburgh, it also has offices in San Francisco and tests its AV fleets in all environments.

Urmson acknowledges that Alphabet-owned Waymo is a competitor. So are GM-owned Cruise Automation, Argo AI (in which Ford has a $1 billion stake), and Aptiv. The AV tech space is indeed packed—more than 60 companies are now claiming to be LiDAR “manufacturers,” for example. Software start-ups in the U.S. alone number in the hundreds. Urmson expects a significant contraction in the number of serious players in the coming years.

“Only a handful of companies globally will get to having a sufficiently self-driving car,” he said, given the difficulty and importance of the technology, and the geopolitical dynamics around it.

“This technology is too important to the U.S. to have it be, say, powered by a Chinese company,” Urmson asserted. “And I think similarly, it will be too important in China to have self-driving technology there powered by a U.S. company. So, I imagine there will at least be players in those two countries,” as well as perhaps a dominant player in Europe.

“It will be hard to demonstrate [one player per global region] early on,” he added. “If the technologies come to market comparably at a similar time, they will grow in their perceived safety in parallel. If a couple companies make it there dramatically sooner, it will be hard for a company to enter later,” he opined.

‘Responsible’ development

Many close observers of the AV development race put Waymo in the lead — an observation with which Urmson currently agrees. “I’m really proud of that team [Waymo]; they’re doing good stuff,” he said, adding that few companies have the depth of experience that the Waymo and Aurora teams have in how to solve the self-driving puzzle.

However, he doesn’t see that it matters if one company gets there first, because the “greenfield opportunity” in this space — estimated by UBS to offer about $2.3 trillion in economic activity by 2030 — is so enormous.

What matters most for the mobility industry, he said, is how quickly, safely, and robustly the leaders can transition to volume-scale products.

Urmson, Anderson and Bagnell have no ambition to be AV manufacturers, and are thus focused on serving their OEM partners with a powerhouse development team. Aurora's VP of software engineering, Jinnah Hosein, hails from Google and SpaceX, where he “launched and landed rockets,” Urmson said. Another top engineer is well versed in the complexities of autonomous ore haulers used in open-pit mining. Aurora’s leaders also are committed to maintaining optimum integrity across the organization and through their hiring process [see sidebar].

The team is adamant that AV development be conducted responsibly, given the risk of human lives in testing, and the threat of regulations that could stall progress. “This is something we need to be thoughtful about as an industry — that the opportunity to save lives and give people better access to and reduce the cost of transportation is really profound,” Urmson said. “It’s an important mission because transportation is fundamental to everything we do. Bad actors who see just a pile of money and don’t see the social consequences could screw it up.”

He said he has faith in America’s “mentality of innovation” and “freedom to explore” and its “robust legal system that punishes you if you do something reasonably dangerous.”

Toward mass deployment

The subject of whether self-driving vehicles need to be 100 percent safe — and what is termed “acceptable” related to the costs involved — came up during the FT event. “It’s not for me to answer,” Urmson commented, explaining that he believes it’s “something that we have to answer as a nation and figure out what the benefits are.

“When you look at this as an engineer, the answer is ‘as soon as we are infinitesimally better than a person, on average, then that’s when we should launch this. Because at that point, that’s one life saved,” he said. “And we know that because these vehicles are going to share the same software and same driver, effectively, then as they make mistakes over time they’ll get eventually better — to the point where in a very short time period they’ll surpass human driving dramatically.”

Urmson mentioned studies that show if a machine does a job five to seven times better than a person, it would be acceptable to use even if it had defects. Even for an AV pioneer who’s watched the technology’s evolution, the subject is difficult to discuss. But the most important thing is, “it’s easy to forget that the status quo is broken — that 40,000 people die on [U.S.] roads every year,” he said. “We shouldn’t accept that.”

AV deployment for public use will begin with the ride-hailing industry, Urmson believes, because of the favorable economics demonstrated by the rapid rise of Uber, Lyft, and Didi. As the various costs of the technology are reduced, it will spur even lower price points for moving people around, which will attract more users.

“Over time, I think this technology does make it into personal car ownership,” Urmson predicted. “But I think now the economics are so favorable for ride-hailing first.”

And how soon will mass autonomy be apparent in a typical streetscape? “I think over the next five years we’re going to see tens of thousands of these vehicles on the road,” he said. “Then the story we’ve been telling about the increased opportunity being demonstrated, we’ll see the capital investment to scale the production of these vehicles.”

In that near-term period, Aurora and other leading AV developers must prove the value of the technology — in its potential safety and mobility benefits, and what Urmson calls “the democratization of transportation.” In the meantime, the broad and safe deployment of self-driving technology awaits.

---