TG meets the self-driving S-Class

Car company bosses love to introduce new models by being driven in them onto the stage at motor shows. Mercedes boss Dieter Zetsche, an eminent engineer, likes to drive them on himself. But at this year's Frankfurt show he broke his habit and swooshed on in the back seat of an S-Class. Only thing is, there was no one driving.

This feature was originally published in the November 2013 issue of Top Gear magazine

OK, so it's not hard to programme a car to steer a few metres to a predetermined point on a show stage - even if, no doubt, the Mercedes spin doctors lost sleep at the remote possibility of the car getting it wrong in front of all those news cameras. But this S-Class is capable of far, far more than just a few metres. It genuinely is a self-driving car. It had just done a run between the German towns of Mannheim and Pforzheim, about 60 miles in normal traffic, sensing its way around obstacles and dealing with traffic lights, zebra crossings and junctions unaided.

This was another nifty piece of Mercedes-Benz PR, since it was the same route driven 125 years before by Bertha Benz, the world's first long-distance car journey.

Professor Ralf Herrtwich, head of driving assistance at Daimler's advanced development department, oversees the project. "We are serious about autonomous driving. We need to make it affordable," he says. Good point. The wealthy already have what's necessary for their S-Classes to get about without their having to drive. It's called a chauffeur.

Funnily enough, I'd met Prof Herrtwich nine years ago and asked if people wanted autonomous driving. He said: "We currently cannot sense that the demand would exist." I'd felt comforted by that. No need for cars to take over from us drivers, so the tech wasn't being actively pursued. Anyway, industry people were wary because they were afraid of being sued by customers whose automated cars had crashed. And they thought, back then, that there would need to be new infrastructure - road signs with WiFi beacons or somesuch - to direct cars around. Now it's all changed. Technology will soon be available for cars to read the road themselves. The US Army has devised vehicles capable of driving through enemy cities without having vulnerable soldiers aboard. Google has self-steering Priuses bimbling around California. And the carmakers refuse to get left behind. Mercedes says it'll be first. GM says much the same. Nissan has said "before 2020". Tesla is hiring engineers for the same task.

So self-driving cars are coming. But do we want them? Herrtwich's answer now is very different from what it was in 2004: "Customers are waking up to it. In the past two years Google did a lot of good work, no question. It made people realise what was possible. And if it's possible they say 'I'd like to have it.'"

For Mercedes more than almost any other carmaker, self-driving is a good fit. Herrtwich says: "It's a combination of safety and comfort. And those are two of our brand values. It takes us toward accident-free driving." Sure, the system hasn't attained total reliability yet - Herrtwich admits it's not great in the dark, and rain will upset the on-board cameras' vision. But he adds: "In developing this S-Class we gained lots of confidence." You can be sure when it goes on sale it'll be more consistent than we are, and less prone to lapses of concentration or tiredness or ill-temper. Thus safer. And as to the comfort, call it luxury if you like, he says it's nice to have the choice of driving or not. "More and more people say that driving is fun, but not in every situation." He says you'd let the car take over in traffic, or long hauls, or a routine commute.

It's coming, almost by stealth. OK, the S-Class that drove those 60 miles was experimental, but actually its sensor hardware isn't far beyond what a production car has. I've driven a fully optioned showroom S-Class in traffic. It steers around bends by reading the white lines, and by using radar and cameras, knows the position of the car in front. It draws to a halt when the car ahead does, accelerates to follow it. It can do that all the way to autobahn speed. Sure, if you take your hands off the wheel for more than a few seconds, the car beeps at you to get back on the job. (Cadillac says it's two years away from putting this into production with unlimited hands-off. It's called Super Cruise.)

Just for trucking along in flowing traffic, the S-Class doesn't do badly provided it can see lane markings clearly. Plus it'll brake in an emergency if you neglect to, and warn you if there are cars in your blindspots. It's even got night vision. And it can park itself, as many cars now can.

The experimental car uses the same sensors - long- and shorter-range radars for cruise control and emergency braking, a stereo camera and a normal camera. It adds three more: a radar sensor pointing to each side to detect traffic at junctions and a rearward one too. But what's really new is the computing power and the software, and the "sensor fusion" - the software's ability to look at the output of all these radars and cameras. And then do the job you and I do: figure out what's going on.

Then, of course, the car has to act on it. Decisions about junctions, roundabouts, pedestrian crossings, red lights, and when it's safe to move into the path of oncoming traffic to get around a parked lorry or roadworks. The trip wasn't rehearsed as such, but the car's navigation database had to be bolstered with extra info about lane markers, pedestrian crossings and traffic lights. Plus a database of photos along the route was installed in the car. It compared the images from the camera with those stored ones, and when they matched it knew exactly where it was, far more precisely than by GPS alone.

What's been the hardest thing to overcome? Not the actual driving per se, says Herrtwich. "What's hard is the environment perception - it's amazingly difficult to read traffic lights, to figure out which one among many is the one that actually applies to you. Also proper positioning of the car in the road. We thought GPS would be able to do more than it can. Also the system is bad at social interaction. If it detects a pedestrian near a crossing it stops. If the pedestrian waves it on, it just stands there."

Surely the software struggles to predict other drivers' paths? "Well, our system has the advantage of sometimes detecting a vehicle before we humans see it. The system has a predictive algorithm on top and a separate set of underlying emergency reactions."

Herrtwich says autonomous driving is allowed in the legal framework we now have, the 1968 Vienna Convention. It says the driver must be "in control", and since there's a driver at the wheel one could argue he or she is in control, even if asleep or surfing the web. But an unambiguous amendment is due in two or three years, he says.

Harder to fathom is the way we'll take to it. Broadly, we have got accustomed to human error causing crashes. But the idea of a machine causing a crash, or failing to prevent one, is somehow more upsetting. No doubt autonomous cars will have to pass a far stricter driving test than humans do.

But whatever you and I think here and now, carmakers act globally. Britain and Europe's driving culture has always been more DIY than the rest of the world. America made cruise control and automatic transmissions near-universal when they were still rare here. Why wouldn't they want to go further? Who'd want to do a Tokyo or Shanghai commute if the autonomous system would do it for them? And when you come to an interesting stretch of road, the engineers say, just switch it off.

More places you'll see this technology

GM EN-V concept

Always autonomous, should never crash. Needs its own paths away from other traffic, possible in future Asian cities.

M'ways will be the first roads where autonomy's possible, as traffic is one-way. SC takes full control of steering and speed.

Now testing on Japanese roads with near-autonomous systems, using laser scans. Nissan plans to sell autonomous cars by 2020.