Now what? Problems and the bigger picture

Before sleeping behind the wheel becomes acceptable, let alone commonplace, we must make all the necessary

pit-stops to grapple with what we’re really building.

 

Be it Mercedes-Benz S-Class, the BMW 7 Series or Volvo S80 – the top line offerings of manufacturers are primed with enough features to help you relax on the highway, not collide with pedestrians or lane-shifters in city traffic and park into that tight spot. A lot needs to be worked out though to achieve “full self-driving automation” – level 4 as defined by the National Highway Traffic Safety Administration (NHTSA) of the USA – a time when driving to somewhere would be as simple as entering the destination into the navigation module à la Google Maps.

 

Professor Dr Hermann Winner has taught automotive engineering at Technische Universität Darmstadt in Germany since 2002. He was responsible for developing adaptive cruise control at Bosch in the early 1990s and holds more than 100 patents in the field of automotive engineering. But it’s his work with others at the Daimler-Benz Foundation – on the book Autonomes Fahren: Technische, rechtliche und gesellschaftliche aspekte (Autonomous Driving: Technical, legal and social aspects) – that has contributed immensely to the discussion on driverless cars.

 

“It’s state-of-the-art technology, but not something that can do magic”“Apart from being entirely accurate, maps and current location strategies can develop availability problems. Hence, sensors and cameras must provide primary information. GPS and maps can help you perform better, but you can’t rely on them,” he says.

 

But sensors aren’t enough in themselves, as BMW's

Dr Werner Huber explains: “It’s state-of-the-art technology, but not something that can do magic. There is a limitation on lane detection due to snow on the road. There is limitation on cruise control due to visibility owing to rain or fog. A sensor is very limited in its knowledge when the environment is not optimal.”

 

Peter Lehmann, head of research cars at Mercedes-Benz, concurs and adds that more development is needed. “For steering, we use stereo cameras but will need to upgrade to infrared to operate in bad weather conditions and at night,” he says. Infrared light can cut through fog and help the car spot the oblivious deer it’s about to crash into.

 

While stressing on the drawbacks of the sensors, Prof Dr Winner points out that apart from continuous work on making them better, we also need to make them cheaper. But it’s not just the car that needs work. Highways and urban areas mostly have well-painted lane lines and road signs, but sensors are ill-equipped to handle off road routes.

 

“If you cannot detect anything, then it’s hard to control a car. We can compensate a lot of elements missing in the infrastructure but it’s not really a smart idea to support the driver based on too less knowledge. It’s better to throw off the systems than pretend you know everything when the car is more or less blind,” says Dr Huber.

In a world of limited autonomous driving then, the transfer of control from machine to human is a dicey situation. Dr Huber believes the transition is a problem: “When it is necessary to get the driver back into the driving task, you need a foresight of 150-200 metres.”

 

A world of fully autonomous driving brings its own set of challenges though, uncertainty and ethics being most prominent Prof Dr Winner notes. The topic of ethics for a robot is a story in itself, as to how it must be programmed to choose between two wrongdoings or give priority to life inside and outside the vehicle. In the book co-authored by the professor, the chapter on ethics concludes that the automotive industry cannot ignore the larger implications of robotic cars: “Without looking at ethics, we are driving with one eye closed.”

 

But it’s the other aspect – uncertainty – that Prof Dr Winner stresses on throughout our conversation. “With autonomous driving as opposed to driver assist systems, the car must always make a decision. With adaptive cruise control and braking, there’ll be no action if the uncertainty is too high. But you have to take a decision with autonomous driving even if you have uncertain information. No one has figured out the uncertainty management until now,” he says.

 

Globally, road traffic is responsible for the death of1.24 million people every yearFor the time being, automakers are happy to invest in driver assists. Continental believes it will be a step-wise approach, starting with partly automated driving in 2016 and ending in fully automated driving by 2025. But others such as Chris Urmson, head of Google’s self-driving car programme, think otherwise.

 

At the annual TED conference event in Vancouver, Canada this year, he delivered a 15-minute speech on how Google’s cars see the road and the challenges that lie ahead. In his conclusion, he said: "We think the right path is to go through the self-driving rather than driver assistance approach because the urgency is so large. In the time I have given this talk today, 34 people have died on America's roads.”

 

And that’s just a fraction of the issue we have on our hands. Globally, road traffic is responsible for the death of 1.24 million people every year. That amounts to the loss of someone’s life every 25 seconds.

 

For decades, companies have invested billions of dollars to protect the car’s occupants from metal and glass in case of an incident. But now they must work on what Urmson labels the least reliable part of the car: the driver.

 

Home

© copyright akhil arora