Aurora Innovation's self-driving cars roll through Pittsburgh

It’s pretty convenient to have forward collision warning tech in your car, warning you to slow down or, in some models, even applying the brakes automatically.

Those systems have been shown to reduce collisions by about 40 percent, explained Sterling Anderson, co-founder and chief product officer at self-driving startup Aurora Innovation.

But that’s still nowhere close to the world of self-driving vehicles, he said during a talk about autonomous vehicles safety on Tuesday evening at Aurora’s Lawrenceville office.

And that’s why the company is taking a “conservative” approach to developing its robot cars, Mr. Anderson told a group of over 50 mobility enthusiasts at a meetup hosted by SAE International, a Warrendale-based organization that develops standards for engineers around the world.

Advertise with us

In a past life, Mr. Anderson was in charge of developing Tesla’s Autopilot system, which allows those affluent enough to purchase one of the luxury electric vehicles to get as near to a personal autonomous vehicle as possible.

When you develop those driver assistance technologies, you know that a human is in control of the vehicle at all times, he explained.

“You expect, therefore, that the worst thing you can do is cause harm,” he said. “You do not want to brake falsely on the interstate for something that you expect is a thing, when there is no thing, and you cause an accident by virtue of you braking.”

So automatic lane-changing, self-parking and forward collision systems are more OK with false negatives — situations where the car may not “see” something with its suite of sensors. That’s better than a false positive that would force the car to slam the brakes for a shadow it thought was a person, for example.

“You don’t have that luxury in self-driving systems,” Mr. Anderson said. You have to get it right every time.

Advertise with us

That fits into the Aurora mantra: “One should not attempt to build a ladder to the moon.”

Aurora, which is headquartered in Pittsburgh but also has offices in Palo Alto, Calif. and San Francisco, Calif., employs 272 people, according to data from LinkedIn. It’s just one of four companies building and testing autonomous vehicles on Pittsburgh streets, including rivals Argo AI, Aptiv and Uber.

According to an October 2018 “Notice of Testing” document that the company sent to the Pennsylvania Department of Transportation, the startup is operating three autonomous 2017 Lincoln MKZ cars in Pennsylvania and employs eight full-time safety drivers here and 14 in California.

Despite having less money than Argo AI — Ford invested $1 billion into that Strip District startup two years ago — and having less clout and manpower than Uber, Aurora has managed to catch the eye of Amazon and Sequoia Capital, which contributed to a recent $530 million funding round.

The team is still taking its time.

Advertise with us

Engineers usually want to incrementally build autonomous vehicles, Mr. Anderson said, because it’s gratifying to see progress each day. That’s in contrast to building up a solid infrastructure, creating custom hardware and software, and rigorously testing it before deployment.

Mr. Anderson said Aurora takes the latter approach, which is partly why most people haven’t heard much about the company, outside of the occasional funding announcement.

This is deliberate, he explained.

“Our focus here with the engineering team is ‘put your heads down, work on the problem we know we have to solve, focus on the foundational infrastructure for that problem and when we get there, we will look like we are slower early on. This is fine.’”

A ladder will not get the Aurora team to the moon, so to speak. But some fuel will, Mr. Anderson said, and that comes in the form of slow, intentional development.

Advertise with us

The team has spent much time creating custom hardware around the central computer, he said. That makes it possible to synchronize information from sensors across a vehicle at micro-second precision.

Aurora has also built an automotive connector that will make it easier to plug the autonomous system into its partners’ vehicles, creating a smooth integration.

Maps took a great deal of time to develop, despite the fact that Aurora originally had no interest in developing them, Mr. Anderson said.

But probably most importantly, Aurora takes its time in simulation.

Mr. Anderson showed the audience a video of one such simulation from a scene in Pittsburgh. A pedestrian starts to cross the street and the Aurora car comes to a halt.

Advertise with us

He shows that the driver used caution, taking over and braking, even though the car may well have stopped itself.

That’s because it’s better to be safe than sorry in the real world, he explained. The team can learn just as much from a simulation as it would from driving toward the pedestrian.

So they take the footage back to the office and run the simulation to see what the car would have done in autonomous mode without any intervention. In other words, they’re looking at how the self-driving system makes decisions after-the-fact rather than exposing the public to a potentially dangerous situation.

And that’s exactly why Mr. Sterling and his team won’t tell us when to expect a fully autonomous Aurora vehicle. They themselves don’t quite know yet and are in no rush to expedite the process.

“Self-driving is a very, very difficult problem,” Mr. Anderson said. “Akin to getting 250,000 miles to the moon.”

Advertise with us

Courtney Linder: clinder@post-gazette.com or 412-263-1707. Twitter: @LinderPG.