Our self-driving future has yet to materialize. What’s next for AV tech?

Beep. Beep. Beep.

Edwin Olson can still hear the faint sound of the autonomous vehicle he helped build as part of MIT’s team for the DARPA Urban Challenge in 2007 as it returned from the hourslong competition.

“It was the most amazing feeling,” he told Tech Brew.

Nearly 20 years later, Olson is the CEO of an AV company of his own, May Mobility, and the industry is in a much different position than it was in the early days of self-driving vehicle technology.

Plenty of would-be contenders have come and gone. Leaders have emerged. Lots of modern vehicles now come equipped with some automated features, and fully self-driving vehicles are operating on a limited basis—progress is being made. But hype around people being ferried to and fro in AVs on a mass scale has collided with technical, regulatory, and scalability challenges. Yet industry stakeholders are still working toward a future where many more of our transportation needs are met by driverless vehicles—and someday by steering wheel- and pedal-free ones.

“The public sentiment was significantly ahead of the technical reality…And then what basically happened is, the technical reality got ahead of public sentiment. And I think that’s where we are right now,” Olson said. “It can’t do everything. But it works pretty damn well, and it can solve real problems.”

Starting line

Research into self-driving vehicles dates back decades, but many point to the DARPA challenges as seminal events that helped kickstart the modern AV sector.

In the early 2000s, the US Defense Advanced Research Projects Agency (DARPA) started testing AVs on a 150-mile off-road course, according to the University of Michigan. The DARPA Grand Challenge series pitted major universities like MIT, Stanford, and Carnegie Mellon against each other, starting with a competition in the desert in 2004 in which none of the AVs completed the course.

Olson was part of MIT’s team for the DARPA Urban Challenge in 2007, and recalled it as a “humbling” experience that involved creating what was essentially a rudimentary joystick to control the vehicle.

“The hardware was terrible…The vehicle control was very imprecise,” Olson said. “Everything was dodgy. The ride quality was bad. The autonomy was really bad. But we got there.”

The challenges are widely credited with establishing the foundation of the AV sector as we know it today, including contributing to the start in 2009 of Google’s self-driving vehicle project—what is today Waymo, the robotaxi sector’s leader.

“Since then, we’ve gone up and down in hype and disappointment and adjusting expectations and whatnot,” Jeremy Carlson, associate director for autonomy at S&P Global Mobility, told Tech Brew, “because it turns out that it’s a pretty challenging task, what we’re trying to do in building autonomous vehicles to replicate and improve upon human driving.”

Expectations, meet reality

While the DARPA challenges generated lots of momentum around AV tech, it became clear soon enough that there were massive technical hurdles to overcome, according to experts and stakeholders. One of them, Carlson noted, is that it’s nearly impossible to teach AV systems the infinite number of complex driving scenarios they could encounter in the real world.

“We don’t have a system today that’s going to address every single possible corner case in the world,” Carlson said. “Whether or not we need to build that in the first place is an argument that most developers will say, ‘Hey, it’s impossible to actually cover every corner case.’ To do that kind of rules-based approach, which is where a lot of the software and programming side of this started, it’s just kind of unrealistic.”

Then there are the realities of the automotive business, which is capital-intensive and slower-moving than the tech industry’s hype cycles, Carlson said. And AV tech in and of itself is incredibly costly—not just in terms of hardware like cameras, sensors, and radar, but in terms of establishing operating bases and maintaining large vehicle fleets.

These challenges are one reason why we’ve seen much of the traditional auto industry put its resources into advanced driver-assistance systems, or ADAS, which partially automate the driving experience (for example, allowing drivers to take their hands off the wheel on the highway or automatically change lanes) but still require human involvement.

In one of the starkest examples of this shift, GM last year announced it was pulling the plug on its robotaxi business, Cruise, and focusing, at least for now, on ADAS. The company attributed the decision to “the considerable time and resources that would be needed to scale the business, along with an increasingly competitive robotaxi market.”

“I think as we transition from that rules-based, software approach to what we might describe as an end-to-end architecture, something that’s a little bit more flexible in identifying the situation, understanding what it’s supposed to do…that will cover a lot more of those corner cases,” Carlson added. “It will make those systems a lot more adaptable, very much like a human brain would be.”

Envisioning the future

Though the highest levels of vehicle automation operating on a wide scale may still be many years away, in the next several years, experts expect to see expansions in ADAS features and the more sophisticated automation we’re already seeing today for use cases like ride-hailing and trucking. Think of packages being delivered via autonomous vans, robotaxis launching in more and more cities, and long-haul routes no longer featuring human drivers behind the wheel.

“You’ve got Waymo out showing everybody this can happen,” Olson said. “And I think this is really important, too, because the way that we build our cities today…this personal car ownership thing, it’s a disaster.”

He and other industry stakeholders describe a future in which multi-car households go down to one car, and eventually, just maybe, to zero—perhaps eventually spurring complementary zoning and development changes to meet the needs of a society less dependent on personal vehicles.

Johann Jungwirth, EVP of autonomous vehicles for Mobileye, touted potential benefits like restoring lots of free time to people no longer beholden to their daily commutes, increased productivity at work, more choices around where people live if transportation is less of a concern, better transit accessibility for people with disabilities, and huge safety improvements. These boons, he believes, will be delivered both by autonomous personal vehicles and fleets.

“I think we are just at the beginning of this,” Jungwirth said. “There is so much more to come, once vehicles can be designed without steering wheels or pedals.”

In the next five years, Carlson expects to see AVs become more integrated into many people’s everyday lives—largely in the areas where the tech is available today, like California, Arizona, and Texas.

“Those vehicles are very much a part of people’s lives today. It’s obviously in limited areas,” Carlson said. “But that is beginning to expand quite quickly.”

He also expects to see gradual advancement into small- and medium-sized cities in the coming years, and that technology advancements will “also make it possible to put it into locations where we might have inclement weather or other obstacles that we don’t see them being deployed in today.”

May Mobility, which has partnerships with Lyft and Uber and provides autonomous microtransit services, is working toward a future of transportation in which autonomy, paired with public transit, outcompetes the experience of owning a car.

“I think we’re very, very close to that,” Olson said. “When people ride in an autonomous vehicle…it’s like getting somewhere in a way that doesn’t suck.”