From the perspective of the autonomous vehicle industry, I hold two iconoclastic opinions that would seem contradictory.
Autonomous cars are not happening in the foreseeable future.
The technology for successful autonomous vehicle deployment in real commercial situations was ready about ten years ago.
That’s very confusing isn’t it? How can autonomous cars not have a plausible future if the technology was ready a decade ago? Yes, that is exactly my question. I have covered why I (and many industry leaders) do not believe that autonomous vehicles are coming soon. How exactly then can the technology be ready now and have been patiently waiting for a decade?
In my article, Autonomous Car Plan Without Supernatural AI, I propose a very valuable economic and technical model for an autonomous car venture that could have started years ago with no special undeveloped technology.
Recently I thought of another obvious application of autonomous cars that is long overdue - parking.
Amnon Shashua, cofounder of Mobileye, recently gave a brilliant talk at MIT on the state of the art of autonomous vehicles. His talk is called "The Convergence of Machine Learning and Artificial Intelligence Towards Enabling Autonomous Driving". Pretty fancy! He talks about all kinds of very tricky problems. At 8:15 he divides the problems facing autonomous driving into three categories: sensing, mapping, and driving policy. None of these things are required for autonomous driving. None. Let’s explore how (even 10 years ago) cars can drive by themselves without complicated driving policy issues, without mapping, and incredibly, without any kind of sensing whatsoever. Perhaps the first rule of moving forward with autonomous vehicle technology is to look at what AI researchers are finding difficult and specifically avoid that problem.
Back to parking. If you drive to a store, sporting event, whatever, inevitably one of the most annoying aspects of your whole day will be parking. James Bond’s cars sporting machine guns is far more plausible than some of the parking spots he lucks into. The thing that is very special about the parking problem is that it is not on public roads. Many people in the business believe that changing the infrastructure to accommodate automotive autonomy is absolutely unthinkable. It simply can not be done they believe. But (in addition to too many obvious examples to mention) it obviously can; if it could not, by the same arguments, there could be no parking lots.
If some private developer can build acres of drivable asphalt in the first place, then I’m pretty sure they can make minute adjustments to help out autonomous cars. And it wouldn’t really take much. All that is needed is a drop off area and a dedicated roadway (one lane could suffice) to some special parking area. All the cars need is drive-by-wire and wifi. They don’t need mapping or complex policy AI to interact with humans in parking lots. Once the owner of the car gets out, the car begins its unmanned autonomous journey on private roads that are guaranteed to be clear of anything complicated. See how I just got rid of 90% of problems in autonomous driving?
Let’s get rid of another 9%. (The last 1% we can do!) We don’t even need sensors. Seriously. The cars can be blind. All we need are stationary surveillance cameras. You could say that putting cameras in parking lots would be too expensive, but I’m not talking about 1906, I’m talking about 2006 when such cameras were so cheap that they showed up in $30 flip phones. The cameras can cover, redundantly if you like, the entire journey the car is expected to make on its own. All of the computation for the system can be in proper computers that don’t need to be squeezed into a wattage cars can sustain. The car makes an SSH connection to the central control. The central control plans a trip and starts giving the car low speed instructions on getting to a parking spot. The control watches to make sure everything is going well. If it doesn’t, we have an insurance claim and not trolley philosophers.
At this point, the people in the autonomous car business may try to make up some stories for the purposes of confirmation bias. They will posit—if it can be done so easily and we haven’t done it, there must be a reason. There is only one just-so story I can think of: nobody cares. Because the auto industry and computer researchers have not solved this problem, it must not a problem that needs solving. But this is extremely wrong.
This DOT report tells us this astonishing fact about car accidents that occur in parking lots.
…the information on [nontraffic crashes was not] available until 2007, when Congress required National Highway Traffic Safety Administration (NHTSA) to start collecting and maintaining information pertinent to these events.
Wow. That must be because such accidents almost never happen and parking lots are so safe, right? Keep reading that report. There were, on average, 1621 annual deaths in parking lots between 2008 and 2011. And we’re just getting around to keeping statistics on this? Throw in the 91,000 injuries per year and I’m pretty sure there’s going to be a lawyerly amount of money floating around to clumsily mop up the mess post hoc. That expense is passed on to consumers in the form of insurance and legal costs. Even if you are in favor of boosting for lawyers and insurance agents, surely society would be functioning better without 91k unnecessarily broken people per year.
But even if that sort of medieval warfare kind of horrific violence doesn’t bother you there are other things to consider about parking. It sucks not just for drivers. It sucks for everybody. This article in The Economist is a great place to start when thinking about the parking problem. There are so many externalities and hidden costs to subsidizing parking (like we do). Just the pollution from people driving around looking for parking is quite non-negligible.
If you want to tell me that a city taxi driver’s job is still essential, fine. But a parking valet? Come on! If so, then we’re never going to see any real autonomous cars ever. So let’s start with this fruit that’s not just low-hanging; it has dropped right into our shopping basket!
The real reason I’m keen to see it isn’t that I give a toss about parking personally. My family understands that I like to proudly park in the usually empty section reserved for athletes (sort of the opposite of the idea behind handicapped parking). What I think is important about pursuing this is that it can be a success story. It can be real, not vapor. It can make a real ROI for people working on autonomous vehicles. It can comfortably create a track record of safety. It can acclimate people to the idea. It can spur car makers to get on with some progressive hardware. It can generate new ideas about what is possible and where improvements can be made in real commercial situations. Before climbing Mt. Everest, try something hard but easier like Mount Cook; Sir Edmund Hillary wisely did.
Update: The very competent civil engineer points out that this kind of thing is commercially viable.