× close
Michael Everett, assistant professor of electrical and computer engineering, is director of Northeastern's Autonomy and Intelligence Laboratory.Credit: Matthew Modono/Northeastern University
For many people, this is an attractive proposition. Imagine a world where humans no longer need to own cars and can instead commute by robotaxi.
Self-driving car evangelists say the potential benefits are vast. Having fewer human drivers on the road could reduce greenhouse gas emissions, reduce vehicle accidents, and reduce traffic congestion.
Self-driving companies such as Waymo, Cruise, and Amazon's Zoox have been developing their technology for more than a decade and are deploying and testing robotaxi services in some U.S. cities, including Phoenix and San Francisco.
Progress has been made over the past decade, and these companies continue to expand into more cities. For example, Alphabet subsidiary Waymo last month began offering a safety-less, chauffeur-less robotaxi service in parts of Los Angeles.
However, even as technology advances, its deployment is not without controversy. A select number of cars have been recorded to 'glitch out', meaning they stop in the middle of the road, make illegal turns and cause accidents. Self-driving cars also continue to struggle with driving in snow, rain, and other difficult weather environments that cloud their sensors.
Tesla, which is developing its own self-driving technology, looks set to soon throw its hat into the ring and unveil its own robotaxi on August 8th.
When will we see mass adoption of these robotaxis services?
Michael Everett, an assistant professor at Northeastern University in the School of Engineering and the Cooley School of Computer Science, says the technology still has a long way to go before it's good enough to reach the mainstream market.
“To me, that technology doesn't seem like it's here yet,” said Everett, who directs Northeastern University's Institute for Autonomy and Intelligence. “The reality is that these self-driving cars are still quite specialized equipment.”
The autonomy kit on these self-driving cars consists of LIDAR sensors, GPS navigation systems, and numerous cameras, Everett said. These self-driving cars must make many decisions every second as they navigate the road to understand their surroundings.
And they're far from perfect, he points out.
“There are several parts of the autonomous decision-making process that the car has to go through,” he says. “We're trying to understand a lot of different things in the world, and even that isn't that obvious.”
By leveraging lidar sensors, which emit pulses of light to help identify objects in the environment, self-driving cars can create an approximate map of their surroundings, Everett said.
The challenge comes when the artificial intelligence in these self-driving cars needs to determine what is safe to drive and what is not, he says.
“You would think there would be a surface and it would be safe to drive, but I think in one of those cases, the person on the ground was hit by another car,” he said. “after that [the self-driving car] You might think, “Oh, is that a pothole?” Or is that something really important to avoid?'
“While these rare events may seem mostly innocuous, it is important to consider whether they are truly safety-critical situations or are borderline situations that can be ignored and continued. “That's one of the toughest questions,” he said. I would add.
Everett said it was difficult to assess how far the broader industry had progressed because companies “had an incentive to make their technology seem more advanced.”
Tesla is currently at the center of a class-action lawsuit for misleading users about the capabilities of its self-driving technology.
Over the past few weeks, the National Highway Traffic Safety Administration has also launched investigations into Waymo, Zoox, Tesla, Cruise and Ford, all of which are testing self-driving cars and advanced driver-assistance systems.
Everett emphasized the importance of these companies being transparent as they implement these technologies.
“Transparency is very important because these are not like self-driving cars where nothing goes wrong on a private test track and everyone agrees to accept the risk,” he said. says.
“These are in fact experiments that have been going on on our public roads for about a decade, and the people walking on the sidewalks may or may not have explicitly consented to being part of these companies' experiments,” he added.
So what needs to happen to significantly improve these technologies?
That artificial intelligence needs to evolve further, Everett said.
“The hardware is pretty great at the moment,” he says. “Lidar sensors far exceed human capabilities in terms of accuracy, range, and building a representation of what's around us.
“I actually think a lot of it is in the algorithmic software. There’s still a lot of innovation that needs to happen there,” he says. “Given the hardware that currently exists, a Waymo-type vehicle provides more than enough data to solve the problem.”
So how do these algorithms find out?
“It's a billion-dollar question,” Everett said. “We have a very large engineering team that focuses on different subsets of the problem: thinking about planning, thinking about perception, thinking about predicting other agents in the world. It exceeds human ability and then continues to exceed it.”
