A car that is both intelligent and safe, that navigates traffic entirely independently: this is the vision of autonomous driving. But what is actually already possible with our current technological capabilities? A team put together by Mercedes-Benz tackled this question across five stages, each on a separate continent. The team was on the move from September 2017 to January of this year, and when team leader Jochen Haab talks about Intelligent World Drive today, he can barely contain his excitement. With its assistance systems the series-produced S-Class was capable of a level of automated driving that Haab and his colleagues were amazed by. But they know there is still much work to be done. In the years ahead, developers will have to answer myriad complex questions.
“The idea was to show that we could put the car into typical situations and gather experience,” says Haab. “We can then use the information obtained from doing that to improve our systems.” Their journey began in Frankfurt, at the IAA, and took them to the United States by way of Shanghai, Australia and South Africa. Jochen Haab describes the challenges that the S-Class encountered.
Frankfurt, Germany: “There was rain and congestion; construction and speed demons on the road: all everyday occurrences on the German Autobahn.”
A unique peculiarity of German driving can be found in the large speed variations on the Autobahn. You can pass a lorry travelling at 130 km/h while a car comes up behind you at 200 km/h. The on-board radars scan the road 80 metres behind the car and 250 metres in front of it. The stereo camera sees 500 metres ahead – and 90 metres of that is in 3D. Is that enough? How much time does the system have to react after registering an approaching vehicle? And what will it do to prevent an accident? Customer opinion is already clear: they expect the system to respond only when absolutely necessary.
Traffic in China is characterised by high traffic density, large numbers of two- and three-wheelers, what appear to be zebra crossings on the motorway (though in fact these markings are for showing safe driving distances), and speed limits for individual lanes. More than anything else, driving through Shanghai showed us that maps need to be absolutely precise, right down to the lane. In densely meshed motorway networks, the car is not currently capable of determining its position; in addition, the lanes at many major junctions are unmarked, making it impossible for the cameras to detect them.
Without surrounding traffic, at the moment the car cannot know where it is. Without this information, it would just drive ahead. People here also expect autonomous cars to take curves in the way Chinese drivers do: continuing to brake in the curve and accelerating afterwards. European drivers, on the other hand, begin accelerating before leaving the curve. Taking these cultural differences into account is imperative for the acceptance of autonomous driving.
Melbourne, Australia: “Many Melburnians avoid the notorious hook turn, but our systems have to be capable of finding a solution.”
Australia: speed limits that flash on electronic displays, road trains in the outback, kangaroos unexpectedly hopping across roads, and then the Melbourne “hook turn” – a right turn from the left lane, which crosses tram tracks and can only be executed when the cross traffic has a green light. It’s a manoeuvre that can test a driver’s nerves – even locals would sooner make three left turns – and like so many other difficult traffic scenarios, it poses complex challenges for driving systems. The human brain can process the blinking LED rings that indicate the speed limit, but the vehicle’s systems are not yet powerful enough. And the question of how to reliably respond to animals without endangering passengers is just as tricky. It requires the dimensions of an object to be detected early, and an exit strategy to be adapted to the situation in order to avoid a potential accident.
Traffic in South Africa is confusing: pedestrians walk on the clearways and unexpected manoeuvres are just an accepted part of driving. In Cape Town, our test drivers encountered an oncoming bicycle as they drove at 100 km/h on the motorway: the cyclist was riding on the hard shoulder, which was just a few centimetres wide. And on a serpentine road south of the city, the cameras could no longer detect the route because of sand covering the asphalt. On the winding coastal road, however, the car could drive almost autonomously.
Cape Town, South Africa: “It’s important for the technology not to get distracted, but it also needs to be able to quickly and accurately analyse perceived chaotic situations.”
Las Vegas, USA: “The smooth traffic is practically ideal for autonomous driving. From a technical perspective, we’re not far from being able to implement this for everyday use.”
It’s smooth sailing on America’s freeways and highways: keeping up with traffic here is easy, and there are rarely large variations in speed. However, it can be difficult to simultaneously gather and process the multitude of information an eight-lane freeway provides, and where passing is allowed on both the right and the left. Nonetheless, the US market is certainly suitable to be among the first taking further steps towards autonomous driving. Americans’ reactions to the experiment were similar to those seen around the world: a mixture of scepticism and enthusiasm. Taking their concerns and questions seriously was one purpose of the journey.