In a 28 July article at New Scientist, Jeff Hecht says that fully autonomous cars are still in the future. In fact, “Some observers are now openly saying the dream of full autonomy is a mirage: creating robot vehicles able to tackle any kind of road or traffic situation is just too tough a nut to crack.”
What does it mean?
The article cites SAE International, formerly the Society of Automotive Engineers, for its six levels of automation, as shown in the diagram at the top of this post. Fully autonomous vehicles, at level 5, are still in the future, with vehicles working at levels 0, 1 or 2 now.
The New Scientist article does a good job of reviewing problems with getting to fully autonomous vehicles: safety or the perception of safety; the predominance of computer engineers in autonomous vehicle development with their “move fast and break things” style (I can hardly write that phrase without shuddering); and the inability of human to quickly take over driving if needed (the article cites research that it takes an average of 5 seconds – an eternity in a fast moving vehicle).
The article notes that, skipping over level 3 and moving to level 4 or 5 autonomy might make sense so that the human never needs to engage, but then points out that computer vision stumbles over simply situations that humans can easily handle: “We can instinctively tell, for example, whether lane markings are complete or dashed lines even if they are partly covered by snow, or that a stop sign remains a stop sign even if partially obscured, and instantly recognise the implications of an emergency vehicle heading our way.”
A level 2 driving system, according to the SAE classification, GM’s Super Cruise option in premium Cadillacs, is described by Cadillac as “the first true hands-free driving-assistance feature for compatible roads.” When engaged, the technology controls both speed and direction of the vehicle, staying in the lane; it can also change lanes when the driver turns on the turn signal. A Driver Attention Camera system makes sure that the driver is still paying attention. Its ability to drive is limited: “If equipped with Lane Change on Demand, you are able to prompt the system to change lanes for you. However, Super Cruise will not steer to avoid safety situations. You need to take control to steer around a traffic situation or object, merge into traffic, exit the highway, make a turn, or stop for crossing traffic or a traffic light, stop sign or other traffic control device. Super Cruise does not steer to avoid construction zones.”
Then comes the beginning of my “I told you so” moment. The system is available for use only on designated roads. I studied the map (at the same link I gave earlier) and found that most interstate highways are included. Near my hometown of Pueblo, designated roads include I-25 and stretches of US-50 east and west of Pueblo that are divided highways. We drive to and from Columbus, Ohio, at least once each year to visit family, and I found that our first day of driving, which is not on the interstate or other divided highway, is not on designated roads, but I-70, which we join in Hays, Kansas, is, as is the rest of our drive on I-70, except for a stretch of I-70 west of Kansas and a few gnarly intersections with other highways (which I, as a human, know are pretty tricky). I expect that Super Cruise would not be useable in practice on some other stretches of I-70 where I know that construction was underway earlier this summer.
Honda claims to have a level 3 vehicle (recall that the driver must be ready to take over driving quickly). I found the disclaimers disquieting, including a diagram showing, I think, that the vehicle might recognize only the upper section of the cab of a truck pulling an empty flatbed. It might not detect the flatbed and, I think, your vehicle thus might follow the cab and could run into the flatbed.
The CEO of GM, Mary Barra, wants to have Super Cruise available “in 95% of driving scenarios.” New Scientist points out that of the 6.5 million kilometers of public roads in the US, only 300,000 kms meet current Super Cruise requirements. “Of those that aren’t covered by the system, 4.2 million kilometres are paved, ranging from busy city streets and quiet, wide, well-maintained streets in affluent suburbs to lightly travelled two-lane rural byways without centre lines. The remaining 2 million kilometres are unpaved, lacking markings and often signs.”
Dr Missy Cummings is a professor at Duke University, director of the Humans and Autonomy Laboratory and Duke Robotics, and an expert in human-autonomous system collaboration. In an October 2020 interview with Forbes, she was blunt about the problems with sensors, testing, and lack of repeatability in performance of existing autonomous vehicles. She says, for example, “If you can’t get a single Tesla to repeat its behavior in the same conditions over and over again, then why are we letting these cars, in theory, engage in automated driving?”
Finally, in my “I told you so” moment, New Scientist says, “The need to upgrade those roads to be robot-friendly ‘is a hidden cost most people are not thinking of’, says Cummings.” The designated roads for Cadillac’s Super Cruise are designated because these are roads that have been designed and built in a way that is friendly to an autonomous vehicle: divided highways, clearly and consistently marked lanes, and so forth. The part of the system that is not carefully controlled in that environment is all the other cars on the road.
On 23 Jan 2021, I wrote: “One of the difficult parts of autonomous driving is to predict what other vehicles will do, especially ones being driven by humans. Limiting the environment to only automated vehicles provides an easier problem to solve. The word `system’ also suggests coordination among automated vehicles, in which they share navigation information, but also share information about their intentions.”
I will go further now and say that the best application of autonomous vehicles is in long distance driving using designated lanes with only other autonomous vehicles. I think that application will be very useful for long distance trucking. One could counter my argument by saying that I have described a train and I agree.
What does it mean for you?
Don’t be dazzled by the rhetoric. Another quote from Dr Cummings in the Forbes interview: “I’m not worried that China or Russia will pull ahead of us in AI technology. Everyone is really bad at it right now.” She also describes current facial recognition technology as “terrible,” and says about Elon Musk “I’m not bothered by him as a person, I just really want him to reconsider what he’s doing with Autopilot because I think it’s exceedingly dangerous.”
Don’t be dazzled by the big companies throwing money around. New Scientist mentions that Uber and Lyft both sold their vehicle development groups in the last year.
The New Scientist article concludes: “There is a growing sense that the phase of irrational exuberance that often characterises new technologies might be over for self-driving cars, replaced by a more limited vision in which automation doesn’t fully replace human drivers, but helps us drive better under certain circumstances. That’s still a revolution of sorts – just not the one, perhaps, we first thought was coming.” I think this conclusion applies to many other areas of automation.
Where can I learn more?
Because the phrase “artificial intelligence” has almost come to mean any computer application, my skepticism may seem to be countered by useful applications. I am not, of course, saying that computers are useless. But this article has a list of successful applications that includes two I have already dismissed (autonomous vehicles and face recognition) and also includes Gmail’s spam detector – which, I find, worked flawlessly until a few months ago and now fails regularly.
I admire the people working to further the capabilities of computers to make our lives better. I am old enough to be still thrilled by the existence of my laptop, my cell phone, cruise control on my vehicle, Supernatural on my Oculus Quest II, etc. My life is better with these technologies. But those people must believe in what they are doing and they are not trust worthy in their evaluation of the value of their work or in their prediction of the future of these technologies.
The preceding two paragraphs are an apology for not being able to recommend a good source of reliable information on what to expect from automation in the future.
This work is licensed under a Creative Commons Attribution 4.0 International License.