In public discussion of the future of robocars, the following statements are often seen, but the reality is somewhat different.
Making a vehicle that can safely drive the roads is indeed a major challenge. But all the teams working on it are fully committed not just to driving, but even more to safety. What this means is they won't release a vehicle until they are convinced that it's safe. In addition, regulators are looking to define safety standards to make sure the companies can prove that their vehicles are safe.
This doesn't mean the vehicles will be perfect, nor should it. However, it's much more reasonable to work towards making vehicles that are safer than human drivers, and when those vehicles take over the task from us there will be fewer accidents on the roads.
There are indeed issues around liability, but they're surprisingly less important than you might think. Today we understand the liability of car crashes remarkably well and there is a huge industry that deals with them. Insurance is the central concept and the legal and insurance industries have immense experience in the field.
The dollar cost of accidents today is generally paid for one of two ways:
In other words, it's always the motorist -- or rather groups of motorists -- that pay for the financial cost of accidents, and it's just a question of how the money flows. Lawyers and companies will be very interested in that question, but the effect of the answer on the whole system is small unless huge punitive damages -- higher than what would happen if a person caused an accident -- are awarded. The larger damages question is the real important issue.
It is also worth noting that at least in the US legal system, the makers of a robocar will get sued in any accident, whether they are responsible or not. In the first few years, everybody involved will get sued, and no debate will change that. It's also interesting that unlike most car accidents, these will come with complete recordings of events, probably in 360 degree 3-D LIDAR-vision and more, and it will take about 2 minutes to figure out which vehicles broke the vehicle code and thus get legal responsibility for the accident, and there will be little arguing over that.
The companies making robocars won't make statements on who is liable. Their lawyers will strongly advise them that any statements on that question can only hurt, and so we won't see anybody actually involved in selling the technology answer this question.
I have an extensive article on the questions around accidents you may wish to consider.
Some early experiments at self-driving cars included making closed-off lanes or roads, in some cases with special magnets embedded in the road which could be followed by the car. Several PRT projects also involve closed guideways, or tracks with lines or magnets to guide the vehicle.
This is an easy and reliable way to have an automated vehicle -- essentially just a step up from trains, which have been automated in certain ways for many years. It was a good first step, and a few still feel that the challenge of driving on ordinary roads is too hard.
Many teams around the world are proving that it's not easy, but it's not impossible. The payoff is huge -- if you can only go on a special road your vehicle is highly limited, and growth of the network will be very slow. A car able to travel on ordinary roads can change the world quickly.
There is a principle of such importance here that I have called it the first law of robocars: You don't change the infrastructure. If you break this law it means you have cars that can only operate where you've made those infrastructure changes, and such changes are usually expensive, slow and bureaucratic or political.
A few of the benefits of robocars require very high penetration. Things like automatic intersections without stop lights and very tightly controlled management of traffic (for the maximum increase in road capacity) are best done without any (or too many) of those erratic human drivers on the road.
But a large fraction of the benefits accrue one car at a time, and that's important, because otherwise we would never get there.
Each robocar:
And much more. In fact, the bulk of the benefits are incremental. While it's very interesting to speculate about the cool extra things that can be done in the all-robocar city, there is also a clear path to get there step by step.
There is a massive effort underway, as a subset of the efforts in ITS (Intelligent Transportation Systems) to make a data radio protocol for use in cars. The cars can talk to other cars (V2V or Vehicle-to-Vehicle) and can also send and receive messages from infrastructure elements, like traffic lights or other roadside transmitters.
A special protocol, similar to WiFi, was created with its own frequency band and many other protocol layers have been defined. Planned functionality includes having cars transmit constantly their position (from GPS) as well as their speed, direction, brake status and other useful info. This allows other cars to know what these cars are doing.
(Note the FCC has been unimpressed with V2V up to this point and has made plans to convert the spectrum allocated for this to unlicenced wifi-style use.)
On the infrastructure side, useful planned functions include traffic lights transmitting their state and when they plan to change.
All this could be very useful. Of particular value is the ability to know about vehicles that are hidden from view because they are behind trucks or coming around blind corners.
This, and other functions would be quite valuable to the robocars, but they face a major problem. Only a portion of the cars on the road would have these transmitters -- a fairly small proportion during the 2010s. Even in the distant future, if all cars must legally have the transmitters, there will still be cars with malfunctioning transmitters, as well as pedestrians and cyclists. As such, it will never be acceptable to rely on the radio signal. A robocar must be fully capable of detecting any sort of vehicle, person or obstacle on the road, and acting safely with it whether it has the transmitter or not.
Once the robocar has sensors that good, the additional value of the radio signal is modest. As noted, it becomes truly valuable only for those hidden vehicles. Alas, the number of collisions which occur due to a vehicle being hidden is real and serious, but comparatively modest. And the number of collisions with hidden vehicles that had transmitters will be even smaller for some time to come.
While we can assume any robocar would have the radio, in general if 10% of cars are equipped, the odds of preventing any given radio-preventable accident are just 1%. It's 10% for the robocar, but 10% of a fairly modest number is an even more modest one. The value is real, but the cost of equipping all cars is immense, and doing it quickly is untenable. Robocars can't wait for this, and will be on the road long before this technology reaches useful deployment.
(It should be noted if you have a high penetration of cars with advanced sensors like robocars, and they can be trusted to report on everything they sense around them, you get a different story, and can greatly increase the awareness of each car -- when you are lucky enough to have the right configuration of other such cars around you.)
Some authors have written that robocars won't be practical without V2V systems. That's not true, and it would be a wait of decades if it were. Rather the two technologies will probably develop orthogonally, and eventually prove useful to one another many years hence.
To top it off, there are many problems with this technology. GPS is neither accurate nor reliable. It's a challenge to tell if a car is in the same lane as you if they used a different model of GPS, and GPS fails in many urban canyons. Security remains a major and unresolved issue, and others are hungry for the radio spectrum -- the FCC has recently asked to have it back. Without a law forcing all cars to include the radios, they will not be deployed for a very long time, and even with such a law, they would not start coming out until late in the 2010s and would not be common until a decade later.
There are a number of useful things that can be done with data networking in the car, but most can be done with ordinary mobile WAN network, ie. cell phone data.
Prototypes are here now. Google and many academic labs have demonstrated vehicles on city streets full of cars and people. Major car companies have product announcements for highway lane driving in 2014 and 2015 models. Neither of these is the complete car that can go on any street without supervision, but the course is clear -- it's coming this decade. Already, Google alone has run their cars a collective half-million miles on regular roads -- that's more a trip to the moon and back.
What we won't see for a few decades is very high penetration of robocars in the market. Human drivers will remain on the road for some time to come, perhaps generations, though we may see some roads or parts of town dedicated to robocars in the future. While there are some special robocar benefits that come most strongly when they dominate the roads, there are many other benefits to drive adoption that improve the world one car at a time.
Surprisingly, state governments have been very amendable to clearing the legal pathway for these vehicles. Already Nevada, California and Florida have passed laws and some regulations outlining the methods to test, certify and eventually licence consumer use of robocars. Several other states and areas have proposed laws. The National Highway Transportation Safety Agency, which sets motor vehicle safety standards (such as requiring airbags) is also doing its own investigation of how to write regulations.
(It should also be noted that generally, because nobody thought to forbid it, analysis shows that operation of these vehicles, particularly with a licenced human inside and responsible, but not holding the wheel, is probably already legal in much of the USA and many other places.)
While this might not be true everywhere, once there is safe and legal deployment in some places, particularly the USA, other nations will follow suit. Enlightened regulators see this as a technology to make driving safer. Driving is such a dangerous activity now that the desire to make it safer is overcoming fear of change and fear of robotics.
Ideally, a robocar should never deserve a ticket. If it does, its programmers have made a serious mistake. But more to the point, once they become aware of this mistake, they will work hard to fix it, and they'll deploy that fix reasonably quickly with radio updates. As such, no model of robocar should ever get the same ticket twice. That should make the chance of a ticket fairly rare. Other robocar markers will quickly make sure they also don't make that mistake. It's as though once the first person got a ticket for an unsafe lane change, immediately no other person ever did one again.
A number of methods have been proposed for how police might ask an unmanned robocar to pull over. (Of course, if there are people inside, they'll see the flashing lights and hear the siren like any driver would.) If there isn't a safety issue, there is no reason for the police to pull the vehicle over at all -- just note the licence plate. After all, the cop is not going to have a chat with the empty car.
If safety is a concern, there is a special strobe system that many police flashers already use to inform traffic lights they are approach which might be detected by robocars. Police might be given a special phone number to dial which allows authorized parties to command an unmanned vehicle to pull over given its location and plate number. Or if this is insufficient, the vehicles probably can be trained to recognize sirens and flashers or other signals just fine. Again, the cop won't be having a chat -- if this happens, the owner of the car would be signalled to contact police about dealing with the problem and collecting the vehicle at the side of the road.
People turn out to be remarkably willing to do this. In fact, even though the vehicles are only prototypes today, riders overwhelmingly report positive experiences, and many say their initial fears were quickly reversed.
It seems likely in fact that the problem is not that people won't trust it, but that they might trust it too much. Many systems that will shortly come on the market are clear about their limitations and require the driver to remain alert and watching the road, even if not steering. Some test subjects trust these systems more than they should, and don't pay full attention, so several companies are investigating systems to assure the driver is alert, monitoring their gaze to see where they are looking, insisting they touch the wheel or something else from time to time and so on.
There will for many years be people who don't trust the systems and insist on driving. This is not a big issue, as evidence suggests they are a minority.
Your car today comes with large numbers of computers, and they control some of the most important safety systems like steering, brakes and throttle. These are much simpler computers than you find in your laptop, but you'll also notice that cars drive fine and people are not getting into accidents because of computer failure. (You may have read reports that Toyota's throttle control computer had a flaw that made it accelerate out of control. Those reports turned out to be incorrect; there was no flaw in the Toyota software.)
There are many other small computers managing all sorts of things that you depend on every day, all not crashing. Airplanes, elevators, medical machines -- you name it. All running well and not harming people or failing at significant rates. Machines are not perfect, and the computers in robocars will be more complex than the microcontrollers I've written about above, but the standard of things like Microsoft Windows is not the right standard of comparison.
Some analysts have pondered how computer systems will be able to make complex moral decisions while driving, such as deciding who to run into when there is no choice but to hit one thing or another. That is a difficult problem, but the reality is that humans almost never face this problem in their regular driving, and there's little evidence that they make complex split-second moral decisions in these situations. The numbers suggest it would be an error to insist on this capability before allowing robocar systems to reduce the more real risks of everyday driving.
Designers of robocars will tend to program them to avoid having to make such decisions in the first place, especially when they are operating unmanned and are not in such a big hurry as they need to make the sort of risky decisions humans make. When humans are in the cars, humans can make the morals-oriented risk choices, like how fast to go on a winding mountain road.
Speed limits are set based on what the road engineers and policymakers felt was the maximum safe speed for human drivers. While many human drivers can safely go faster, some can't. In practice we all know that in most countries, the speed limit is rarely complied with and enforcement is intermittent. Indeed, a car traveling at the speed limit on some highways would be an impediment to traffic and present a safety risk to some extent. In some countries, there are no speed limits.
Robocars will probably again let their human occupants judge and decide what speed they wish to go, within the capabilities of the vehicle. Many cars today have cruise control or even automatic cruise control, and these systems let the driver set whatever speed they wish, independent of the speed limit. It seems likely robocars will be designed in a similar way.
It's also possible that in the more distant future, robocars will not be subject to the same speed limit as people are. They will drive reliably and safely, will not follow too closely and will reduce speed in poor weather in a reliable way. They will know the limits of the road, tires and vehicles better than human drivers, and may be demonstrably capable of safe travel at higher speeds. If it's proven safe, there is no reason to legally limit them to lower speed, and they could be allowed to travel at the (usually faster) speed of actual traffic.
This article was in response to the most common misconceptions in 2012. A first round article from 2009 on Objections and their Answers outlines even more myths, with only one in common.
Comments may be left at this blog thread.