clock menu more-arrow no yes mobile
Efi Chalikopoulou for Vox

Filed under:

Why we’re still years away from having self-driving cars

Self-driving cars were expected to roll out by 2021. Here’s what we need to solve and build first.

For the last five years, all anybody in the car world has talked about — well, apart from electrification — is autonomous driving. Carmakers began dropping the terms “self-driving” and “mobility” at car shows, Uber and its competitors poached engineers from university robotics labs en masse, and Tesla fans began squabbling on Twitter about whether the company’s Autopilot system can be called “autonomous.” (It can’t.) Meanwhile, Cadillac, Mercedes, Volvo, and others rolled out similarly equipped vehicles that aren’t quite autonomous but are more or less capable of driving themselves down highways, as long as drivers maintain a persistent vigil and nothing too weird happens along the way.

Meanwhile, visionary urban planners began rethinking city designs to envision what was sure to be a future uncluttered by automotive detritus — no more traffic signs or stoplights, no more cars parked by the side of the road. Vehicles would simply drop you off at your destination and vanish … somewhere. We were told cars would chat with each other and the roads themselves to modulate traffic flow, and that car accidents would no longer be a thing. In fact, the world was so optimistic about this future that then-US Secretary of Transportation Anthony Foxx declared in 2016 that we’d have fully autonomous cars everywhere by 2021.

Flash forward to today, and precious little has changed about our daily driving. You probably hear a lot less about self-driving cars than you did a few years ago, and the prospect of safely dozing off behind the wheel on long drives remains a distant fantasy, even if old-school carmakers are working with startups like Waymo, Cruise, Argo, and Zoox on the technology.

Why the radio silence? There are a lot of knotty problems to solve that are conspiring to delay the arrival of the technology — in fact, answers to these problems may redefine how self-driving cars will work.

Everything from programming vehicles to follow the rules of the road to getting them to communicate with human drivers and pedestrians — forever ending, for instance, that infuriating indecisiveness we all encounter when trying to determine who should go first at a four-way stop — is giving engineers fits. Even further in the weeds: developing sensors that can work flawlessly in all kinds of weather and visibility conditions, and teaching cars how to respond to all the so-called “edge cases” they’ll encounter on the road, such as comprehending the difference between a flock of birds dashing across the road or wind-blown leaves that are fine to run down. Also, cars don’t drive in a vacuum — the roads and infrastructure, as well as federal, state, and local regulations, have to accommodate fleets of robocars, and the public has to be on board, too. Many puzzle pieces must fall perfectly into place.

To put it more simply: Five years ago, as companies developing this tech talked a big game to lure talent and investment dollars, we were all more optimistic than realistic about the timeline for rolling out autonomous cars that are predictable, reliable, and as safe as possible.

“Those early estimates with really aggressive timelines for rolling out the services have turned into having a few research vehicles on the road by 2020,” notes Jeremy Carlson, an autonomy analyst with auto-industry research firm IHS Markit. “Even that might have been optimistic in some cases.”

The reality is that while roads themselves are generally orderly and well-known environments, what actually happens on them is anything but. Humans are proficient behind the wheel, but they’re also imprecise and occasionally wayward. So until 100 percent of the vehicles on the road are fully autonomous — something many analysts think is actually highly unlikely — every autonomous vehicle will have to be able to respond to the edge cases plus countless quirks and tics exhibited by human drivers on a daily basis. It’s the stuff we’re able to swat away without missing a beat while driving ourselves, but getting computers to try to manage it is a really big deal.

Pittsburgh-based Argo and the Bay Area’s Waymo, both frontrunners in the race to perfect self-driving tech, are solving for this challenge by training their autonomous-drive systems to rely as much on precisely scanned basemaps of the road as on sensors used to “paint” the environment around them. That means they will also be limited to areas that are fully mapped, as though operating in a real-world video game. It’s a process most developers will likely need to rely on, even though it requires persistent, continuously updated maps and will likely limit the ability to take vehicles “off-grid” as an owner or user might need from time to time.

But don’t fret. Highly automated driving remains a very real proposition, one that’s being enabled not merely by fast-talking CEOs but by technology that’s indeed racing forward — even if it’s not as fast as we’d hoped. Computer-processing capabilities continue to surge annually while sophisticated artificial intelligence systems are learning to, if not necessarily think like humans, at least run through enough options for every decision to pick the best solutions. The onboard sensor systems that are required to detect vehicles, monitor their behavior, and “read” the environment grow more compact and affordable every year.

Then there’s the ubiquitous communication systems that will tie everything together — namely, cloud computing and the forthcoming 5G cellular network, which will eventually make wireless speeds exponentially faster than the 4G you’re already familiar with and is deploying around the world. The cloud system allows engineers to offload a lot of the data processing away from the vehicles themselves and onto more capable and rigorously updated servers — meaning that the autonomous-drive systems remain persistently state of the art. But it’s the 5G network that could enable a lot of key features within these systems. Though the cars will by and large be able to operate without connectivity, having a more robust, faster, higher-bandwidth wireless data system will significantly boost the autonomous vehicle network’s capabilities. Cities will be able to optimize traffic patterns, cars will know ahead of time what the traffic signals will be at every intersection, and vehicles will communicate with each other to ease everything from lane changes to routing strategies based on congestion or weather.

According to Carlson, that will generate a kind of universal awareness on the part of vehicles, akin to the way GPS navigation will today reroute you based on congestion. “With a better and more robust network, you’ll have longer detection ranges for other vehicles and incidents and have lots of different types of information pumped into the system,” he says. “There’s real value there in terms of how it can make driving better and more efficient.” Researchers have already demonstrated the systems’ abilities to precisely coordinate autonomous vehicles threading their way past each other with millisecond timing, thanks to all the vehicles automatically gauging their relative positions and deciding who goes where. This can only be done when vehicles are communicating wirelessly with each other.

Finally, at the risk of further muddying the waters about when and what we can expect, there’s another variable that’s slowing down self-driving cars: Covid-19. Many carmakers are yet again recalibrating their expectations and timelines for the vehicles, noting that consumer behavior might change permanently as a result of the pandemic and that could mean both a reluctance to use shared-car services — which many had targeted as a significant launching pad for the technology — or, conversely, an increased desire to stay away from mass transit, thus making self-driving options more appealing. Ford announced in April that it would delay its anticipated 2021 rollout of its autonomous car service to 2022, using the time to gauge reassess the market.

The pandemic might also spur increased interest in contactless delivery, of the sort Mountain View, California-based autonomy startup Nuro’s engineers are developing via R2, its self-driving delivery vehicle that’s nearly the size of a small car. It’s a more compact form of the same kind of technology that passenger-carrying vehicles would possess, adhering to the same principles and rules of the road, both real and virtual. “As an industry, we’ve seen an unprecedented shift in consumer demand for on-demand home delivery since the beginning of the pandemic — with online grocery sales increasing nearly five times,” says David Estrada, Nuro’s chief legal and policy officer. “We partnered with several nonprofits to help make meal deliveries for local food banks, deliver quarantine kits for those sheltering in place, and use R2 to bring meals to frontline workers who were treating Covid-19 positive patients in pop-up medical facilities.”

Of course, plenty of other technologies for autonomous systems are already appearing in full-sized vehicles, as well, in the guise of advanced driver assistance systems (ADAS), such as adaptive cruise control, traffic signal alerts, and emergency braking and maneuvering. (Technically, all semi-autonomous or “self-driving” systems are ADAS, it’s just that some are more advanced than others.) These will ease consumers into accepting and using them over time while the development of technology for the fully evolved systems continues in the background. The path to autonomy truly does appear to be starting out small, working up to something much larger and more impactful as all the pieces fall into place.


Local police should not be your go-to source for iPhone safety news

The Future Perfect 50

Jerry Chow wants to make quantum computers you can actually count on

The Future Perfect 50

Alec Stapp and Caleb Watney have a plan for national progress

View all stories in Technology