If you’ve read all the headlines about self-driving “smart cars” over the last few days or months, you’d be excused for thinking that mainstream autonomous cars are just around the corner. Unfortunately, it just isn’t so. (Yes, even in spite of Tesla’s announcement last week.) It’s not that they’re not on their way—they certainly are—but expectations about what we’ll actually get and when seem to be out of whack with reality.
Part of the problem seems to be a definitional one. The US National Highway Traffic Safety Administration (NHTSA), as well as its European equivalent, lay out several different levels of autonomous cars. Level Zero is no autonomous features and Level One is for things like adaptive cruise control, which has been around for years. Level Two is for multiple systems working together, such as adaptive cruise control with lane centering, automatic parking and other features we’re starting to see enter the market now.
Critically, in all of these cases, the driver is expected to stay in complete command of the car. The jump to Level Three is a big one because it enables limited autonomous driving where, in certain situations, such as sitting in bumper to bumper traffic, the driver is given the ability to officially cede control of the car to the autonomous system. So, for example, you could legally read email and tweets on your smartphone while sitting in the driver’s seat (instead of doing it illegally, as so many currently do). Even here however, according to the NHTSA’s definition, “the driver is expected to be available for occasional control…” Giving complete control over to the car for the entire trip, what I believe many people think of when they hear about autonomous cars, doesn’t happen until we reach Level Four.
The announcement from Tesla last week about adding some autonomous driving capabilities to certain of its Model S sedans via a software update this summer does appear to be a Level Three-type announcement. But, as others have pointed out, whether it will actually be legal to use those features any time soon is far from clear. That’s at the heart of the problem.
The real challenges here has more to do with liability and legality than anything else. Who’s responsible, from an insurance perspective for example, if an autonomous car hits someone or something? What level of safety can governmental agencies (and car makers) guarantee? These are some of the hardest problems to solve and the ones likely to take the longest time to resolve. They are also the reason why no states have officially allowed the use of autonomous cars in anything but a testing mode (and typically only with a special license).[pullquote]”The technology of making cars function autonomously is obviously coming much faster than the legislation.”[/pullquote]
The technology of making cars function autonomously is obviously coming much faster on prototype cars and other test systems. Not surprisingly, it’s these efforts that have started to generate so much press attention. In addition to Tesla’s announcements, there have been Google’s autonomous car experiments and Audi’s 900+ mile autonomous drive from Palo Alto, CA to Las Vegas just before this year’s CES show. Last week, nVidia announced their nVidia Drive PX smart car platform tools, which provide the ability to connect up to 12 cameras to a development board with two Tegra X1 CPUs and let virtually anyone (with $10,000) start working on their own automated cars. Throw in provocative comments like Tesla CEO Elon Musk’s at last week’s GPU Technology Conference where he predicted that, in the future, non-autonomous cars would be outlawed because of how unsafe they would be perceived to be, and you have the perfect stew of unrealistic expectations.
But you also have to consider more practical issues. As Musk pointed out at that same event, there are roughly 2 billion cars on roads around the world and the auto industry’s maximum output level is roughly 100 million per year. That means, even if the industry started producing nothing but autonomous cars tomorrow, it would take 20 years to completely replace the installed base.
The real problem is, because we’re seeing tech industry companies start to get involved with the auto industry, we’re applying tech industry development speed expectations to autonomous cars. New iterations of smartphones take 6-12 months, so a few years should be plenty to tackle something like autonomous cars, right? Well, no.
The reality is we probably won’t be seeing usable Level Three-types of features in cars we can buy off the lot until the beginning or even middle of the next decade, and fully automated Level Four could take until 2030 or later. By the way, this should also put a new perspective on when (ignoring the even more important “if” question) Apple might enter the car market.
I’m certainly excited about the possibilities of what autonomous car features can bring, but I think we need to keep our expectations in check.
10 thoughts on “Smart Cars Accelerating Slowly”
Thanks for pointing out the insurance and legal side of things. In all the discussion I’m reading on automated cars, that to me seems to be the biggest road block – pun intended.
I think the biggest roadblock is the implications of having driverless and human-driven cars share the same roads. The big problem isn’t technological or legal, it’s one of human behavior. In a nutshell, it’s highly likely that we end up with a situation where human drivers have basically ‘bullied’ driverless cars off the roads.
Bob, what level of safety do governmental agencies and car makers guarantee for Level Zero cars now? Are there specific standards, such as “Brakes and steering shall have X probability of failing within the first 100,000 miles”?
I honestly don’t know, but I’m pretty sure there are standards for numerous components. However, the thing with Level 0 is that there’s no additional risk beyond just the basic elements of the car. Once you start moving up the chain, the risk of systems interconnecting components becomes more of an issue.
I think both issues are not *that* intractable:
– validating an automated drive system is certainly a lot of work and money. Teaching and validating tens of thousands of human drivers is a lot more work and a lot more money. I’m not sure that argument works the way you intend it ^^
– I don’t change houses every time I change desktops. If a car has the required space (for cables+sensors+processing, which I’m sure most recent cars do) and I/O (for automated control, ditto), the renewal and evolution cycles for steel and silicon do not need to be in any more sync that those for concrete and silicon are.
@bobodonnell:disqus , i agree that development and legalization of self driving cars will take some time. Personally i think that we’ll first “automate” transportation using human being and something like UberPool(uber’s ride sharing service), which will make car ownership expensive and unnecessary , and when we’ll have driverless cars , we’ll bolt them into UberPool’s service ,which means that even with the limited production capacity we’ll get to full deployment faster than 20 years.
And if i we’re to guess – i’d say Apple might be developing the ideal car for an UberPool kind of service, because Apple is all about the user experience, And you can’t offer a great experience if you force someone to drive.
Developing a self-driving car that is 99% accurate on a pre-mapped road is not beyond the wit of man. However, such a car would, of course, be very dangerous and hit something or someone every few hours.
Making the car 99.9%, 99.99%, and then 99.999% accurate (i.e. only or out of a 100,000 cars would be involved in an accident every day) is going to be a long drawn out process. 2030 seems very early for that autonomous car, but I do see a lot of potential for driver-assist technology that can make, say, driving on the freeway in slow rush hours traffic a lot more bearable.
Consider me a skeptic on driverless cars. We can’t even deploy fully autonomous ships or aircraft on a large scale even though the technical demands for those are much lower than for autonomous cars.
I think autonomous cars are easier to deploy if all vehicles on the road are autonomous. Then the computing problem is much easier because the action of all cars on the road are predictable. Throw in unpredictable, unreliable, and sometimes intoxicated human drivers in the mix and the computing problem multiplies.
And we all know what’s going to happen, the human drivers will game the system and before you know it driverless cars will end up constantly yielding the right of way to intentionally reckless human drivers so travel times on autonomous cars will fall far behind human driven vehicles and so the only people who will ride driverless are those who have no other choice.
No, autonomous cars will only work if human driven cars are banned from the roads or if full-time automated monitoring and enforcement of traffic regulations is imposed on human drivers. I don’t know about other countries but I think none of those two options are even remotely acceptable in the U.S.
Actually DHL in Europe has started using fully autonomous drones for parcel delivery. I’m not sure flying is that much easier though, if only because you cannot just pull up and stop. Human pilots get a lot more training than human drivers, which probably means flying is harder and riskier.
As for bad and reckless drivers, one can assume if the problem becomes as dire as you envision, cars that are allowed to drive us will also be deemed worthy of issuing citations, or at least of calling for The Law. Long term, the idiots will probably darwin themselves out of the roads anyway ^^.As a user, I’d still rather work or relax for an extra hour in my car than shave an hour of my 2hr automated commute but have it fully take up my time and attention for the remaining hour.
Aside from those 2 issues and the main effect of being able to do something else while in a car, there are huge side effects re. parking (who cares where the car park is if your car goes there and back by itself ?), accidents and insurance (do we even need insurance if we’re not even the ones actually driving ?), even ownership (cars can follow the same model as buses and trains… car as a service), and lots more.
I don’t think driverless cars issuing citations will go down well at all in a country where people won’t even abide remote traffic light enforcement via cameras.