The myth of the driverless cars revolution

It seems that we are on the verge of a driverless car revolution. George Osborne recently announced that a “train” of automatic trucks controlled solely by the leading vehicle would be tested in Cumbria this year, one of eight driverless vehicle projects to receive £20m in funding from the government.

Trials of autonomous cars are already under way in Milton Keynes and Greenwich and far more extensive testing is being undertaken in the US. In recent testimony before Congress, Google claimed that its fleet of 56 autonomous cars had racked up 1.4 million miles over the past six years. The company predicts that autonomous vehicles will significantly reduce the 94 per cent of road accidents that are caused by human error each year and free up much of the 3,000 square miles devoted to parking across the United States.

The revolution, it is often claimed, is imminent. The Tesla Motors co-founder Elon Musk, who wants to build trains with a top speed of 760 miles an hour in California and central Europe, said in December that his “fully autonomous” car would be ready to launch by 2018. All of this has created enormous hype, with the press suggesting that we will soon be able to sit in our cars – or “pods” – while being driven to work, where we will instruct the vehicle to park itself, or take the kids to school.

Yet even the most cursory examination of the present state of driverless car technology and the potential obstacles to its success – legal, social, economic, political and practical – suggests that a takeover of the mainstream transport system is about as likely as the long-awaited arrival of the futuristic jet packs of 1960s comic books.

Take the sensors, for instance. They may work fine in sunny California but could struggle to combat fog, snow and driving rain. Google’s autonomous cars have a somewhat patchy safety record: as of last month, they recorded fewer than 20 accidents, all minor (mostly rear-end shunts) and all, bar one, not the fault of the Google technology. This is slightly above the comparable rate for conventional cars and quite possibly the result of dangers – what if a mouse ran across the road? – that a human being would generally ignore.

However, it is the single incident caused by driverless technology, in a car that collided with a bus, that is most telling. According to Google’s accident report: “The Google AV [autonomous vehicle] test driver saw the bus approaching in the left-side mirror but believed the bus would stop or slow to allow the AV to continue.” This implies that the test drivers are still the ones making such decisions and that the cars are not “autonomous”. It raises a crucial problem for the new technology: risk.

Consider an autonomous vehicle trying to move into traffic, a manoeuvre that is impossible without some element of risk. Human beings often do it through some sort of interaction – say, flashing lights, or eye contact – and so an element of risk-taking would have to be introduced for autonomous vehicles, too. Once that happens, accidents will occur, and what parent would send his children off to school in a vehicle with risk hard-wired into its software?

The driverless car does not stand up to scrutiny. When pressed, Musk conceded that the “fully autonomous” car that he said would be ready by 2018 would not be completely automatic, nor would it go on general sale. There is a pattern. Whenever I ask people in the field what we can expect by a certain date, it never amounts to anything like a fully autonomous vehicle but rather a set of aids for drivers.

This is a crucial distinction. For this technology to be transformational, the cars have to be 100 per cent autonomous. It is worse than useless if the “driver” has to watch over the controls, ready to take over if an incident seems likely to occur. Such a future would be more dangerous than the present, as our driving skills will have diminished, leaving us less able to react. Google notes that it can take up to 17 seconds for a person to respond to alerts of a situation requiring him or her to assume control of the vehicle.

What is this technology for? The widespread assumption that driverless cars will be a shared resource, like the London Santander Cycles, is groundless. People like owning their personal vehicle because it is always available and can be customised to ensure that the child seat is properly in place and the radio tuned to Magic. Google may be right that a few parking lots will become redundant but it has no answer for the possibility that autonomy will encourage more vehicles on to the road.

The danger of all the hype is that politicians will assume that the driverless revolution obviates the need to search for solutions to more urgent problems, such as congestion and pollution. Why bother
to build infrastructure, such as new Tube lines or tram systems, or to push for road pricing, if we’ll all end up in autonomous pods? Google all but confesses that its autonomous cars are intended to be an alternative to public transport – the opposite of a rational solution to the problems that we face.

So why is George Osborne so interested in this technology? One possibility is that he considers it a way of weakening two of the last remaining bastions of union power: lorry and taxi drivers. Another is that he hopes it will provide a boost for the UK car manufacturing industry. Either way, the revolution, when it comes, will not be driverless.

Christian Wolmar’s new book, Are Trams Socialist?, will be released next month by the London Publishing Partnership


  • RapidAssistant

    Great article. I was reading somewhere that even Audi – who are one of the pioneers of autonomous cars (or piloted driving as they call it) say that there will never be such a thing as 100% autonomous driving, rather it will be used mainly in isolated situations such as in traffic jams, or for automated parking. Personally I think the technical obstacles such as the developing the necessary “fail safes” to an adequate enough level in case sensors or computers fail is/are overwhelming.

  • Thanks for that. Have you any link to the Audi reference? It would be good ammo for my next piece on this

  • DrPlokta

    All human-driven vehicles already have “risk hard-wired into their software”, and so we have an answer to your question: any parent would send their child to school in such a vehicle.

  • RapidAssistant

    Christian – I can’t find the exact article (typical….), but these four links do have some good quotes from Audi engineers on the technology.

    Here’s a good YouTube film of the Audi A7 parking itself – this is what they are referring to as a practical application of the technology in today’s world:

  • This is the usual guff that fails to distinguish between driver assistance and full automation. There is no ‘road map’ explanation of how you get from aiding a car to drive along a motorway to full automation in crowded urban environment.
    ERTMS is a good example of just how long it takes to introduce such technology.
    The other interesting point is when the Audi guy says it improves safety. Well, the opposite could be argued. It will deskill drivers so that they are no longer as safe as before.

  • No, you miss the point. It is about perception. We all know that human drivers may have crashes, but we always think that we won’t because we are better than the average or whatever. Even if completely safe automatic planes were developed, people would still want a pilot!

  • DrPlokta

    I’m happy to agree that people are very bad at assessing risk, and this may be a problem for driverless cars. But you were talking about risk, not about perceived risk.

  • RapidAssistant

    Automation has been eroding driving skills for years – when you see a lot of the aggression on the roads today, I think it has fed the “I’m invicible” attitude created by modern safety systems. If, for example we forced everyone back to an era of drum brakes, crossply tyres, no airbags, ABS, traction control and all the rest of it would it make people slow down a bit???
    Taking it a step further – look at America – hardly anyone knows how to drive a manual transmission anymore because they are 95% automatic, and youngsters today in this country will soon forget how to do a hill start after they’ve passed their test – because so many modern cars these days have automatic handbrakes!

  • Mike Sharp

    Aren’t you assuming that not only is our understanding of human risk pretty good, but also that our expectations of risk avoidance are the same for a risk-taking machine as for a risk-taking human?

  • RapidAssistant

    You could take that one stage further and say that software in a driverless system in itself has risk “hard wired” into it because a human still has to write the code at the end of the day, so in that respect – engineering out human error is effectively impossible.

    How can you program a computer that has to reliably deal with any of the infinite number of scenarios that can happen out there on the road. 99.9999% reliability is not acceptable, because that 0.0001% could result in someone being killed.

  • c1ue

    Nice article.
    I’d add another data point: autonomous cars won’t make parking lots obsolete. They will just shift them onto the roads.
    I’ve done ride sharing research by driving for ride sharing for 2 months. After 800+ rides delivered, what the data shows is that the length of the average ride delivered was, in fact, matched 50%+ by the road traveled to deliver said ride. Or in other words, vehicle sharing – whether autonomous or not – involves 50%+ distance per ride than individually owned vehicles.
    This was in a major US city, by the way, which already suffer massive traffic problems. Adding 50% more distance traveled to even 10% of the total vehicle miles traveled in said cities will result in truly astounding congestion.
    I’d also note that the ride share practice of stopping any damn place the passenger desires is equally non-conducive to efficient traffic flow.

  • Excellent point — thank you

  • c1ue

    Another note: a number of people have touted ride sharing as “saving the environment”. This isn’t very clear, especially given the extra distance of travel noted above. As a generalization, half of a car’s total pollution profile is associated with its manufacture and half with its usage. A taxi, for example, will drive 3 to 4 times more than a “regular” car.
    However, taxis average 5 rides per hour at peak, whereas ride share can only do around 2 – because of that extra time driving to pick up the passenger.
    The math on the pollution associated with the extra distance thus is far from clear whether overall pollution per mile traveled is significantly reduced or not: if the ride share is traveling 100% more distance per ride, then the pollution savings of more rides per new car manufactured is totally cancelled out by the greater distance per ride.
    On the other hand, if all the ride share cars are Prius hybrids or electric vehicles in California or Sweden, maybe there can be some savings although lithium batteries are almost all made in China from coal belching power sources.

  • Actuaries will work out that cars are safer when they are driving themselves, so the insurance companies will offer lower premiums to drivers who use autopilot as much as possible. Paul Whitehouse will appear in adverts telling us he lets his car drive itself and saves money.