Bumps on the road: Why driverless cars are further off than we think
Published Thursday, March 8, 2018 6:00AM EST
This past January, a Tesla Model S, travelling in Autopilot mode, slammed into the back of a stopped firetruck on a freeway in Los Angeles. Amazingly, no one was hurt.
But the crash highlighted for many that for all the hype in recent years about how self-driving vehicles are going to revolutionize driving, there are still lots of problems left to figure out.
To hear automakers tell it, automated vehicle technology is advancing so quickly, it’s only a matter of years before we can all begin to sit back and let computers, radars and cameras take the wheel
“Fully self-driving cars are here,” John Krafcik, the CEO of Waymo, told attendees of the Web Summit technology conference last November.
Waymo, the autonomous-vehicle company spun out of Google, is already beginning to unleash a fleet of fully autonomous cars and minivans that are logging millions of test miles on the streets of Arizona and California. The company has also begun testing totally driverless freight trucks.
General Motors is testing self-driving Bolts on the streets of San Francisco, and the company expects to launch self-driving vehicles for passengers and package deliveries in big cities sometime in 2019.
Lyft recently began testing its own self-driving cabs in Boston, while Uber is running a testing program of its own in Pittsburgh.
And Tesla CEO Elon Musk is so confident in his self-driving technology, he has predicted Teslas will be better at driving in the next few years than any human could hope to be.
Change is happening in Canada too, with the first trial run of a driverless vehicle on a Canadian street occurring last October and a large, autonomous vehicle demonstration hub recently opening in Stratford, Ont. -- one of several testing locations across the country.
So it might seem the automated vehicle revolution is imminent.
"It’s not a matter of when, but how quickly they will roll out to our cities." — Paul Godsmark, CAVCOE CTO
If it is, Paul Godsmark, the chief technology officer at the Canadian Automated Vehicles Centre of Excellence (CAVCOE), which provides analysis and consulting services about automated vehicles, believes it’s a shift that is going to change not just the face of transport, but almost every industry in Canada.
He thinks the kind of change coming is similar in scope to how much the internet has changed our daily lives.
“... It’s not a matter of when, but how quickly they will roll out to our cities. Because the technology is just about ready for prime time,” he told CTVNews.ca.
Yet plenty of others are not so confident that the self-driving era is upon us.
Ray Tanguay, the former chair of Toyota Canada and now the federal government's auto sector adviser, told the North American International Auto Show in mid-January that he didn’t foresee the widespread deployment and use of automated vehicles until about 2040.
Even some of those working directly with the technology are trying to rein in expectations.
Bryan Salesky, the head of ArgoAI, an autonomous vehicle startup that has US$1 billion in backing from Ford, recently penned a piece lamenting that the imminent arrival of self-driving vehicles’ imminent arrival is being overplayed.
“Those who think fully self-driving vehicles will be ubiquitous on city streets months from now or even in a few years are not well connected to the state of the art or committed to the safe deployment of the technology,” he wrote this past fall.
So what will it take before self-driving vehicles hit our streets en masse and what will be the hurdles along the road? Here are just a few:
1. Overall reliability
Sebastian Fischmeister, the executive director of the University of Waterloo’s Centre for Automotive Research and an associate professor of computer engineering, says while automated vehicles have been taught how to handle roads, highways, and traffic signals, they are still lacking dependability when it comes to an element of the road that’s much more unpredictable: humans.
"While a human can reflect on a new situation, computers don’t have that capability..." — Sebastian Fischmeister, Waterloo University
Automated vehicles will likely have to share the roads with human drivers for many more years to come and while it’s easy to “teach” smart cars about stoplights and roundabouts, it’s more difficult to teach them how to “read” human drivers.
“The fundamental problem is that computers do what you tell them to do and nothing else,” Fischmeister recently told CTVNews.ca.
He says there are so many possible scenarios that an automated vehicle needs to know how to handle that it’s difficult to prepare them for every one. What should an automated vehicle do, for example, if it approaches intersections in which human drivers fail to obey the rule to yield to the right? How should it navigate an unexpected construction site or highway detour? How can vehicles be trained to read the body language of pedestrians or nervous drivers?
“While a human can reflect on a new situation, computers don’t have that capability unless you program that into them,” says Fischmeister.
While driverless vehicles can be “trained,” with so many itching to get them on the road soon, they are unlikely to have learned how to handle every scenario by then.
Nissan is working on one possible solution: the teleoperation of autonomous vehicles. That would see humans act like air traffic controllers, monitoring vehicles from afar and stepping in to operate a vehicle if it gets confused, loses its connection to GPS, or simply loses power.
But the feasibility of the widespread deployment of that kind of system has yet to be explored.
As with all new technologies, the first products off the assembly line are going to be hugely expensive.
The key technology currently in most autonomous vehicles is called Lidar, or Light Detection and Ranging technology. The system involved dozens of rotating lasers that sit in a hub on the roof of the vehicle and emit multiple short pulses of light. The systems’ computers then measure the return time of those pulses to create a virtual 3-D map of the vehicle’s surroundings.
While dozens of companies are refining Lidar for automated vehicles, the most refined systems being tested at the moment cost more than the vehicles themselves – in the range of US$70,000. Scaled-back versions in the $4,000 range have been unveiled, but it’s unclear whether those would create images detailed enough for city driving.
“That’s a big stumbling block still, and the price of the sensors will have to come down” before automated vehicles can ever be widely adopted, says Fischmeister.
A single sensor may not even be enough; backups might also be needed in case of breakdowns, adding to the costs, he said.
Godsmark believes that with so many players in the automated-vehicles field competing against one another, the price of the sensors will eventually drop. Personal computers and cellphones were once absurdly expensive too, but once mass production began, they became affordable. But for now, costs remains an issue.
Tesla, it should be noted, is one of the only players choosing to nix Lidar altogether, in part because of costs. It hopes that radar, ultrasound and a powerful onboard computer will be enough to gather data about the vehicle’s surroundings. It still remains to be seen whether that decision leaves Tesla the Betamax amid a sea of VHS.
At the moment, most automated vehicles are being tested in favourable driving conditions, on dry roads and during sunlight. But the vehicles have yet to really be put through their paces to see if they can handle the worst road and weather scenarios, such as a nighttime drive in a snowstorm on a road covered in ice.
“In a blizzard where snow is basically coming straight at the car, the vehicle’s sensors can’t see that,” says Fischmeister. “...The last thing you would want is for the car to just suddenly stop when it can’t find its way anymore, leaving passengers essentially trapped.”
And what about durability in less-than-ideal weather? Will the cameras and radar sensors on automated vehicles freeze up during extreme cold, or overheat during heat waves?
CaVCOE’s Godsmark is confident these are issues engineers can eventually address. He says he’s been on tours of technology companies in Israel, and has seen sensor technology that has learned how to navigate through any kind of weather -- reading the road during snow and fog even better than a human, and using “gated time of flight” technology to actually “see” between raindrops.
He agrees it will take more refinements to make sensors that are robust enough to withstand extreme weather, but with so many players in the field working on readying automated vehicles for challenging weather, finding solutions might be “only a matter of time.”
What makes autonomous cars so revolutionary -- that computers do all the work and control everything -- is also what could make them so dangerous.
Fischmeister notes that if someone wants to sabotage your current vehicle, they have to physically approach it and manipulate it. But once that vehicle connects to another, or receives map and traffic information from “the cloud,” it will become open to hacking.
“As vehicles become connected and reachable through the internet, they become vulnerable,” he says.
Every player in the autonomous vehicle game is aware of the stakes, particularly if hackers figure out how to attack multiple vehicles at once. The risks have consumers worried too, with one recent survey by the AIG insurance group finding that three-quarters of drivers are worried about hackers taking control of their vehicles, or recording their private conversations or personal information.
Though automakers, software developers, and security experts are working on ways to stymy hackers, it’s a concern that will likely begin early in autonomous vehicle deployment, with the sophistication of hackers evolving in lockstep with the technology.
“It’s definitely a big topic that will arise as soon as these vehicles get deployed in large numbers,” Fischmeister says.
5. Consumer acceptance
Getting consumers to actually trust driverless vehicles could be one of the toughest hurdles for automakers to cross.
"These things will crash. The hope is they will crash less often ..." — Paul Godsmark
The AAA (American Automobile Association) has been gauging Americans’ attitude to self-driving cars annually for years, and recently found some reason for hope. In late 2016, 75 per cent of drivers said they would be too afraid to ride in a self-driving vehicle; that number dropped to 63 per cent in the latest survey in January, 2018.
But though the numbers of skittish consumers are falling, most of us still feel unsafe about a vehicle in which a human could not take over the wheel.
The irony, of course, is that computer-driven vehicles with cameras and sensors all around it will almost surely be safer than human-driven ones. Though the vast majority drivers think they are excellent drivers, the fact remains that more than 90 per cent of crashes are caused by human error, according to the U.S. National Highway Traffic Safety Administration.
Godsmark expects that consumer reticence to change as soon as drivers start seeing autonomous vehicles out on the roads.
“The more people can see and touch it, the quicker they’ll accept it,” he says.
“...Remember, people didn’t trust autos at first. They didn’t want to get into elevators. But once they try (a self-driving vehicle), they’ll be won over.”
Other consumers might be reticent to buy self-driving vehicles not because they are afraid of them, but because they just enjoy driving too much. Will automakers continue to make “manual” vehicles available to appeal to these consumers, and if so, how will those vehicles integrate into an automated vehicle world?
Godsmark knows there will be bumps on the road to fully automated highways and streets. There have already been accidents that have made headlines and there will likely continue to be more, particularly as automated and human-driven vehicles learn to share the road.
“These things will crash,” he concedes. “The hope is they will crash less often and when they do, it will be less severe.”
More seniors dying on Canadian roads than any other group
Made-in-Canada solutions to seniors dying on the roads
How a refresher course helped me become a more confident driver
How advanced driver training can make anyone a better driver
10 tips from the pros to improve your driving right now
How new safety technology might actually be making our driving worse
Driver’s quiz: Put your driving IQ to the test
Can you train your brain to drive longer into your golden years?