SUBSCRIBE TO OUR BLOG
'The car that drives itself' is an old science-fiction favourite, but suddenly it's dawning on the general public that we might soon find ourselves belting in for real. But 'highly automated driving' can be a confusingly broad label. What exactly do we mean?
A good start is the framework laid down by America's National Highway Traffic Safety Administration. This defines five distinct steps, the final one of which is 'full self-driving automation', a car undertaking the entire journey unaided. You, the so-called 'driver', would need only enter the destination. You'd spend the actual journey settled down to other matters – reading, emailing, video and TV, even dozing.
But it's not just about freeing up the driver's time. A key motivation here is safety. Not just for people in the car, but also the more vulnerable ones with whom we share the road – cyclists and pedestrians.
Autonomous driving can also boost road capacity and cut jams, because cars will able to travel closer together even at speed. And there’s an environmental benefit: smoother, more predictable traffic would save fuel.
It's likely the technology will be ready for primetime sometime around 2020. Nissan has publicly committed to that date. Many other manufacturers seem to be just as well-advanced in their prototypes, even if they're being more coy about the timing.
Ah, but will we want it? Most of us think we are better and safer drivers than we actually are. It follows we're probably also delusional in thinking machines will never be better than us.
Every technological step towards safer cars has met resistance. When seat belts were made compulsory, people said "I'd rather be thrown clear of the impact." When airbags arrived, they asked "What if you're wearing glasses or smoking a pipe?" With ABS brakes, it was "I want full control of the car." Such reactions seem idiotically luddite now.
Sure enough, drivers who've tried them have come to rely comfortably on today's adaptive cruise control and lane-keeping systems. There's no reason to think we won't gradually accept the steps that will follow, all the way to full autonomy.
A level-zero car won’t do anything for itself and is always under the driver's full control.
But that's already ancient history in many countries, where there are already mandatory assistance systems that take the car to NHTSA's 'level one': 'function-specific automation'. They include anti-lock brakes and skid-control systems.
Cruise control, even at level one, can be pretty advanced. Commonly, cars can vary their speed to match the one in front by using radar. Forward-facing video cameras can see the white lines and nudge the car into the middle of its lane. A whole suite of emergency systems are already on sale. Cars see into blind-spots, check for crossing traffic as you reverse out into the road, and even detect pedestrians and autonomously jam on the brakes.
Location-specific functions come into play here. Headlamps can steer around curves on input from the navigation system. Adjustable suspensions might configure themselves for upcoming corners. And we see eco-gains: the plug-in hybrid version of the Mercedes S-class is due on the market this year. Navigation data on elevation and gradient, and speed limits, cause it to set the powertrain to work as efficiently as possible, avoiding unnecessary use of the petrol engine and making the most of the electric motor and downhill battery recuperation.
But the defining factor in keeping a car at level one is that the systems don't combine braking and speed control.
Not so in NHTSA's level two, 'combined function automation' where, in certain conditions, the driver is in command but not direct control. For instance, using radar cruise control and lane-sensing cameras, the car will maintain itself on a motorway lane and change speed to suit the traffic. Or it might do the same things in stop-start city traffic jams. You can actually buy these abilities in a Mercedes right now, albeit if you take your hands off the wheel for more than a few seconds it will sternly warn you to put them back.
General Motors calls its similar system 'Super Cruise', and the company says that when it goes on sale 'later this decade', it will let the driver pretty well fully disengage for extended stretches of uninterrupted highway. Experiencing this is a bit of a rubicon: when the driver can take his or her hands off the wheel, feet off the pedals and – most transformationally – eyes off the road, the whole relationship with the car is entirely redrawn. But always, the driver is expected to come back into play at short notice.
Once that notice period is extended and the driver really can set about other unrelated things, the car will have arrived at NHTSA's level three: 'limited self-driving automation'.
The Mercedes and GM systems rely on cars' internal sensors, but there's another way, being researched by the EU Sartre project, Toyota and many others: 'platooning'. This uses vehicle to vehicle (V2V) communication. A lead vehicle will send its speed and path wirelessly to vehicles behind. The leader will probably be a truck, because – contrary to stereotype – truck drivers are highly qualified and skilled. This 'road train' of closely packed vehicles can follow each other closely along a motorway, increasing the road's capacity.
In any NHTSA level-three setup, a well-flagged handover between driver and vehicle is critical. When you're turning off the highway or when the city traffic speeds up, the car must alert you in good time, so you can clear your head and re-focus on driving.
But once we’re at NHTSA's level four, 'full self-driving automation', that shouldn't be necessary. You should be able to trust the car in all circumstances.
Perhaps surprisingly, according to Professor Ralf Herrtwich, head of driver assistance at Daimler's advanced development department, the hardware to do that isn't much more advanced than what's already fitted to the top-spec S-class. Herrtwich was responsible for the recent 60-mile autonomous drive through real cities and rural roads in Germany.
The step-ahead tech in that experimental car wasn't the hardware, but in software, image-processing and navigation. The car could spot all the traffic, see pedestrians and cyclists, and react correctly to traffic lights even at complex junctions. It used both recognition functions and a prediction algorithm. Critically, it knew exactly where it was on the road, to a far greater accuracy than GPS, by comparing the real-time images from its cameras with a new generation of ultra-detailed maps by HERE.
A fear of lawyers, perhaps. Imagine an autonomous car crashed into you. The old adage 'sue to money' comes into play. If you thought you could get compensation from a car manufacturer, among the world's most capital-intense entities, well you'd be more likely to go to court than if a regular human were driving. So you can be sure that before they bring this stuff to market, manufacturers will be doing a whole lot of legal due diligence as well as testing for technical robustness,.
And what about the barrier of driver acceptance? Well it's true many of us greatly enjoy driving ourselves. But probably our carefree fantasy of steering down the open road takes a back seat to the actual daily grind of traffic-choked commuting and endless motorway drudgery. Very well: when the going is dull, or you're tired, or busy with your emails, then let the car drive itself. When the traffic clears and a tempting road opens up ahead, grab the wheel and hit the system's 'off' button.