AV is a common acronym used to refer to self-driving cars, but, while typically it stands for “autonomous vehicles,” that’s actually a misnomer.
“[…] Autonomy implies a kind of decision-making [like for example] ‘where do I want to go today?’ And we don’t expect automated vehicles in the foreseeable future to be making that kind of judgment,” said Marjory Blumenthal, Senior Policy Researcher at think-tank RAND Corporation, who, as lead author of a study entitled “Safe Enough,” accordingly referred to the vehicles in question as “automated.”
Higher Self-Driving Safety Standards
Discussion of safety is ubiquitous in the field. In her interview with 6GWorldTM, Blumenthal said the bar has been raised for automated vehicles in a relatively short amount of time.
“When we did [a] previous report, which was published in late 2018, we had listened to people talking about comparing to the average human driver and by 2020, when we did the new round of research, that was not good enough, and so people talking about automated or self-driving vehicles are talking about comparison to a better-than-average, a safe human driver, with some people pushing that to professionally trained and safe, adding all these qualifiers on.”
Mitch Tseng, Chair of the Testbed Council at the Industrial Internet Consortium, spoke separately to 6GWorld. Discussing the future of Intelligent Transportation Systems (ITS), Tseng pointed out how automated vehicles such as unmanned buses factor into many such deployments, including in New Zealand and Singapore. From his perspective, the technology is effectively ready, but he agreed there’s a way to go before it meets consumer expectations.
“Just imagine this: You’re a human driver and you’re involved in an accident, people will say it’s bad luck,” he said. “The problem is, when you have an unmanned vehicle, whenever there is an incident, people have a different way to assess that and say, ‘Oh, autonomous vehicles are not safe,’ but they forget that even with human drivers cars on the road are not safe either.”
As a result, Tseng argued, it can be a very hard industry to succeed in, due in large part due to legal liability. He recounted an infamous incident in which a Google car was broadsided by a manned vehicle running a red light. He called the human-error-caused accident unavoidable at the time, but suggested advances in LIDAR and camera technologies could be made in the future to prevent an incident under similar circumstances.
“If, for example, in the intersection, you also have some kind of message repeater […] [Vehicle-to-Infrastructure; V2I] would tell the car on the broadside ‘I’m traveling at 40 miles per hour and I’m not engaging the brake […] even though the light may be changing,’” he said. “The Google car on the other side, if they have V2I technology, would receive this information [and can engage in] defensive driving to avoid an accident.”
An Automated Vehicle Technological Roadmap
On the subject of advances, Nick Harris talked to 6GWorld on the potential benefits of photonic computing in the automated vehicle space. Harris is the CEO of Lightmatter, which bills itself as the “photonic (super)computer company,” with the computing platform in question purpose-built for artificial intelligence (AI) applications.
“We can build a faster processor that can handle even more advanced AI functions and it’s using less battery range,” he said. “I think, if you want to build a car that doesn’t have a steering wheel and no pedals and you’re truly just a passenger and you say to the system, ‘Take me to wherever,’ there’s a big gap between what electronics is going to be able to do, because of these fundamental challenges that we run into with transistors, and what we need. That’s where photonics plays a role. We’ll let you do more computation and use less battery.”
Of note, Apple has long since entered the fray as well, with its Project Titan having started operations in 2014. However, the classical computer company just announced a 2024 target to produce a self-driving car, potentially with new battery technology itself that would reportedly increase vehicle range.
Ashley Nunes, a researcher and consultant who specialises in transportation safety, addressed Apple’s announcement in an interview with 6GWorld. He expressed skepticism regarding the overall projected benefits, in part because of the apparent shift from the robotaxi model other companies are pursuing to a personal-car approach.
“If everyone tomorrow went out and bought Apple’s driverless car for their own personal needs, you would see some reduction presumably in emissions… and some reduction in terms of congestion on the road, because these vehicles are supposed to be better at platooning, but it’s not necessarily clear you would see a reduction in the total number of vehicles on the road, which has long been one of the selling points of driverless technologies, that we reduce these inefficiencies in car ownership,” he said.
The Need for Education and Oversight
Nunes argued it’s unrealistic to expect that self-driving technology will be rolled out on a widespread scale anytime soon. Asked about Uber recently selling its self-driving unit to start-up Aurora, he said it was a good example.
“What’s interesting about [Uber’s] sale is the fact that Aurora, which is headed by Chris Urmson, who I debated last year on this very topic, for the longest period of time was talking about perfecting their technology specifically for robotaxis and now they have pivoted to say ‘We think a better application is trucking,’” he said.
Nunes argued it may be a sign of acknowledgement that robotaxis aren’t a recipe for financial success. He conceded trucking could work, with less oversight required due to trucks generally running on rural roads or highways with less congestion and chances of things going wrong. Even so, he argued, less oversight doesn’t mean no oversight at all. It doesn’t mean financial viability, either.
“The regulator will say, ‘You can’t have people on the road for longer than a certain number of hours.’ Fair enough… but the same is true if you have someone remotely monitoring a truck,” he said. “There’s only so much time I can spend looking at a monitor, which means at some point you need to bring someone else in, which is a cost. So, it’s unclear whether or not the cost savings associated with removing an individual from the truck and replacing them with some sort of teleoperation service is a fiscally viable pathway to boosting the margins for trucking companies.”
From a technological perspective, Harris said not all is lost. However, regulatory concerns are just one aspect preventing widespread public adoption.
“I would say that the neural nets that we’re using right now to do analysis of objects that are around and just sensing the environment, it’s in a state where it works pretty well, but it doesn’t work all the time and that means that you need a lot of redundancy and back-up checks. It adds complexity,” he said.
“If you can run much bigger and more powerful neural networks, I think you’d have a chance at reducing complexity and increasing safety and helping convince regulators… So, I think that there’s obviously a big road towards getting self-driving cars ubiquitous, but a lot of it’s technological. A lot of it is [also] educating people,” he added.
Tseng brought up the education aspect himself, arguing it’s key especially in an age when cellphones are everywhere. In the context of ITS, he said the private and public sectors need to work together to improve communications between one another and to the public. He used a hypothetical example of automated valet parking.
“When you need to retrieve the car, you just press a button and it will come and get you at the exit, something like that. For some reason, a glitch [may] happen and a car runs into another car. Somebody just happens to be there and they take a picture and post it [online]. That would have an impact. During the millions and millions of trials, only one makes it [online], but that one gets spread millions of times, not the success stories,” he said.
Nunes believes the sector could rebound following an incident like that, pointing to the airline industry and Boeing 737 Max getting certified by the Federal Aviation Administration after a defect resulted in several fatal crashes. He said a financially viable AV business model is possible, but that the available evidence doesn’t necessarily support the notion of success to the degree AV manufacturers are suggesting.
“Systems do fail and we find ways to make those systems better, but I think asking a question as to how safe is safe enough is secondary to this notion that high degrees of automation, regardless of their perfection level, still necessitate human oversight,” he said. “As long as you require oversight either for technology-based reasons or for regulatory-based reasons, you’re introducing the very cost the manufacturers said they were trying to eliminate; and then this requires some sort of cost-benefit analysis to say ‘Well, do the benefits associated with having remote oversight outweigh the costs associated with it?’”
Feature image courtesy of metamorworks (via Shutterstock).