Why your first driverless car is decades, not years, away
The technology needed to realize the dream of fully autonomous personal vehicles is growing steadily. But so is the list of hard-to-solve problems.
There was a time not that long ago when it felt like consumer-level driverless vehicles might be just around the corner. The car companies were hyping the technology. The media was too. You can even find examples of people like Obama’s Secretary of Transportation Anthony Foxx predicting that by 2021, a person would be able to walk out their door and hail a driverless car to take them to work or school.
Well, now it’s actually 2021, and there still aren’t market-ready options for fully autonomous vehicles (AVs). In fact, the hype atmosphere around personal driverless cars seems to be giving way to a sense that we might actually be pretty far from the finish line. When we asked UM-Dearborn Associate Professor Sridhar Lakshmanan, who specializes in vehicle perception systems, how long it might be before we have autonomous cars that can give us a chauffeurlike experience, he initially said “years.” Later, when he offered more specificity, his answer was, “maybe 10 to 15 years.” “That’s probably the nearest realistic estimate I could see for when you and I could go down to the car dealership and actually buy something like that. And widespread use would probably be beyond that.”
One of the fundamental reasons that Lakshmanan is less bullish about the timeline is that the list of challenges seems to be growing at least as fast as the technology. “I think the biggest general challenge for autonomous vehicles are so-called ‘edge’ cases,” Lakshmanan says. “These are situations where vehicles have to respond effectively to subtle changes in the environment. Humans are highly adaptable to new environments, but vehicles have to be trained for them. I think what we’re seeing now is the number of cases like this that you have to test these vehicles for is ever-increasing. The edges just keep multiplying.”
Lakshmanan’s point is that we have likely underestimated just how messy driving is — and how relatively good humans are at dealing with the chaos. Under typical conditions, today’s incarnations of driverless vehicles can replicate the decisions of human drivers with almost spooky levels of performance. But when you throw something weird at them that they’ve not been coded to respond to, things can sometimes take a turn for the worse. Pedestrian interactions or situations involving aggressive drivers are two edge categories that are particularly tricky, not to mention high stakes, says Lakshmanan. “We humans have a sixth sense for whether someone is going to suddenly jaywalk out in front of us. Or if you see someone doing really crazy things on the highway, you might respond by pulling over into the slow lane. I’m not saying it can’t be done, but I think that kind of adaptiveness is something we’re learning is very hard to replicate in autonomous vehicles.”
The whole technological paradigm for solving issues like this is something that’s evolving too. Professor Paul Richardson, who works on vehicle-to-vehicle and vehicle-to-infrastructure communications, says car companies’ initial vision was to have all the systems contained in the vehicle itself. Nowadays, that’s not considered the only, or even the clearest, path to the finish line. For example, a car that, like humans, relies primarily on perception to navigate through space has serious limitations once the world changes. Winter weather obscuring lane lines is enough to render optical vision systems ineffective. “Even mud on a sensor basically can blind the car,” Richardson says.
Researchers are now looking at how HD mapping systems (which are far more precise than GPS), or communications technology that lets vehicles talk to highway infrastructure or each other, could help AVs maintain their sense of place in tricky situations. It’s promising work, but Richardson says it also opens up new terrority for possible hurdles. “Now you’re talking about a car that’s not only a supercomputer, but a sophisticated communications hub. And the cell companies are going to want a piece of the action.” Moreover the communications networks we have today are nowhere near adequate to handle the real-time communications needs of vehicles. Lag time or the equivalent of a dropped call simply can’t happen if you expect vehicles to operate safely. And the challenges just multiply once you have thousands of these vehicles simultaneously competing for bandwidth. Most of the published work in this area has only been proven with vehicle fleets numbering in the dozens, Richardson says. So, bottom line, the driverless car revolution may likely have to be accompanied by a revolution in communications technology.
Beyond these technological challenges, however, Lakshmanan says companies that plan on bringing personal AVs to market still have a more fundamental riddle to solve. “To me, one of the interesting problems of autonomy has always been what is the value proposition to the customer. Is it a convenience feature? Is it a safety feature? Is it cost-saving? I think that’s still not clear,” he says. In other words, the question of why we need or even want autonomous vehicles is not something that’s been broadly proven to consumers, though he says the value case is a little clearer outside the American and European markets. In hyper-dense cities in China and India, for example, using autonomy in tandem with electrification could mean an urban transportation environment that’s far less congested and polluted.
Even without a clearly articulated need for personal driverless vehicles, the fascination remains. And both Lakshmanan and Richardson say this emphasis on fully autonomous personal vehicles can obscure a more interesting question: how autonomy is more clearly positioned to transform our world in the near future. Semi-autonomous robots already work alongside workers in Amazon warehouses, and Richardson says similar technologies could become common in everything from mining to manufacturing to healthcare. Truck “platoons,” where many autonomous semi trucks follow one driven by a human, could definitely be viable for long stretches of lonely Sun Belt interstate. And Lakshmanan says the pandemic has supercharged interest in delivery robots, especially for restaurants, where owners have seen their takeout orders surge but profitability wane due to the high-cost of human-powered delivery services like DoorDash, Grubhub and UberEats.
Lesson being: While it could still be a decade or more before a driverless car is taking you to work, your first AV-delivered pizza may in fact be right around the corner.
Story by Lou Blouin. This is the first story in a series exploring the future of autonomous vehicles. For more on this topic, check out the rest: "Designing autonomous vehicles to be pedestrian friendly," “How we’ll ultimately learn to trust autonomous vehicles,” and "Building hack-resistant driverless cars." If you're a member of the media and would like to contact Associate Professor Sridhar Lakshmanan or Professor Paul Richardson for an interview, drop us a line at UMDearborn-News@umich.edu.