There won’t be many driverless cars in this century

There are so many misunderstanding of self-driving cars that I thought a full blog post instead of just a tweet is necessary.

How it all began

The creators of K.I.T.T. created also the dream in many, now grown-up engineers to let people do something else while driving. Obviously they weren’t the first with the idea of autonomous cars (“auto-cars”) or driverless cars, but probably the most prominent especially in the US and here in the EU reaching millions of children at least. Always keep in mind that “autonomous” or “self-driving car” does not necessarily mean “driverless car”. And as K.I.T.T. many cars of the future will have a control mechanism (like the steering wheel and breaks) to let the human take over from the autonomous mode.

A K.I.T.T. replica. Universal Studios – via Wikipedia

Why driverless cars and not e.g. driverless trains?

Why pick the most complex problem first and not choose trains or airplanes instead for the autonomous revolution? The society cries for more secure solutions on the road so it is clear that computers could be better, but then why are there still people in a train or airplane if people are insecure?

Time will show this in more detail, but the two facts aren’t necessarily contradictions. I think that easier to automate situations instead of fully driverless cars will be introduced early on especially where the expensive driver should be replaced or should be able to sleep. For more complex situations like urban centers the “driverless car” will take probably decades. For example we’ll see early on auto-trucks and auto-cars on the motorways (again: not necessarily driverless) but maybe driverless shorter trains. For longer trains the driver is probably not that expensive to make sense to be replaced, the advantages like social acceptance might outweigh the cost factor of the driver.

Another reason why driverless trains are not yet wide-spread and everyone is talking about the more complex driverless cars instead could be that the trials with cars are just not yet done widespread enough. And that e.g. in 3 or 5 years it will be shown that the society would accepts autonomous cars and driverless cars only, if they are not only a lot more secure but virtually flawlessly. The acceptance limits will be extraordinary high after the first crashes and problems (like they are for airplanes) and so we might see an extremely decreased speed in adoption for driverless cars probably leading to earlier driverless trains then. For example in Nuremberg and elsewhere there are already driverless tubes and the one in Nuremberg operates since 7 years. More about this example below.

The lie about the “high precision map”

Thanks to Nokia HERE all people seem to know that without high precision maps you’ll loose the future game of auto-cars (not only Nokias fault). The most famous German TV news (Tagesschau) cited uncritical that precise maps are necessary for this revolution, mashable is doing the same: “GPS mapping in cars has existed for years, but that currently only scratches the surface of the data needed for an autonomous self-driving car.” and of course, they are not alone. The truth is: that was a big and successful marketing campaign from Nokia, which insisted that HERE is able to do auto-car and auto-cars are obviously the future, so HERE is the future and needs to be bought.

But if you think a second about this (or certain people are allowed to think a bit longer) you’ll find out that precise maps are not necessary for auto-cars to work. Imagine that such maps would be necessary. What will the auto-car do if a tree is fallen onto the street or an accidents happens just in front of the auto-car? Will it continue because that is not on the map? It should not. Nearby changes or problems have and are actually tracked using cameras and various sensors but no high precision digital map should be necessary. Maps are necessary to efficiently guide any car from A to B in the road network, this further reduces riscs for all passengers of course. There is one place though, where more precise maps will be used: at the beginning of autonomous driving when simulating rail-like behaviour or to avoid too many processing while driving. But at the end this cannot be the foundation – only a tool to make a self-driving cars safer.

What would a computer need if it drives the car entirely on its own?
More precise maps to crash more precisely into passengers crossing the road?

High precision maps are not necessarily the future of navigation from A to B. But frequently updated maps with a global coverage are. Only then you know new roads and smaller ones in the ‘outback’ and more. This is the reason I think the community based OpenStreetMap project is the map of the future involving thousands of active daily contributors already. Just look at this comparison or use it for your region.

But there are real problems.

Problem 1: Driverless Laws

The biggest problem will be, especially here in good old Europe, that auto-cars involves a lot of acceptance from society and specials laws. Some U.S. states already passed laws for this, other states and countries will follow but certainly it won’t be easy and identical making the introduction very slow. Every country will handle the question differently of “how to handle compensations if a driverless car is involved in a crash”? As discussed below the transportation network will stay always heterogenous and no one can solve this question easily by just referring to “once every car is driverless no crashs will happen anymore”. There are many ethically unsolvable problems. What if the cars needs to decide between two disasters e.g. killing a child versus killing a group of persons? Does one need to program a killer instinct into the car like technology review suggests?

Problem 2: Security

When you hear about the lately released security problems car makers have, then you wish (and bet) that the introduction of driverless cars will be delayed until they fix this first. Because security problems increase if you remove the driver. What if an attacker places a stop sign nearby the road and the driverless car then is forced to stop and all passengers are robbed? Surely you can do device-to-device communication but only if you make it again secure. Which is far from simple as you’ll not only need to secure communication but also authenticated traffic signs like shown earlier with keyless systems.

Now again the driverless tube example from Nuremberg. Researchers discovered security problems with the tube, making it possible to even reach the control center of it and manipulating speed, light and doors. Security researchers think it is just impossible to discover intrusion and making the system ‘perfect’. You can see that security will be the second bigger problem that the introduction of driverless cars will be delayed. Here digital companies having ‘fresh software’ could have a big advantage over the old software grown over the years from existing car makers.

Problem 3: Heterogenous Transportation Network

The third bigger problem is the transformation from 100% manual driven cars to a more autonomous world and the resulting heterogenous transport network. There will be always a heterogenous network involving pedestrians, bikes, cars, auto-cars and driverless cars. Manual drivers will still be necessary e.g. for the city center with so many quirks to handle and also extreme weather scenarios probably won’t be solved fast. And so driverless cars will be rare, especially when taking into account the complexity of law necessary in heterogenous transportation networks.

Conclusion

Autonomous cars will come, but I do not see a fast way to a driverless transportation network, even not for developed countries like Germany, just look at ‘easier’ to automate systems like airplanes or trains. Cars will run autonomous where this make sense like on highways, but even the first 10% coverage of fully driverless cars will take decades.