The bugs need sorting but test drives show they already work and within a decade they could be on our streets.
Self-driving cars are no longer from the realm of science fiction
Imagine being able to read a book or watch a movie, check your emails or enjoy a meal while your car automatically ferries you to your destination. Then when you arrive and climb out, you push a button on your smartphone and it finds itself a parking space. It could even take your children to school for you.
To those who enjoy the thrill of driving, the prospect of a machine taking control of this symbol of personal independence may be nightmarish. For the rest of us who just want to get from A to B, it could be a dream come true within a decade.
Self-driving cars are no longer science fiction. Engineers and software developers around the world are already test-driving them, and although they still have some key problems to sort out, they work. The sensors, cameras and radar technology they are based on are already in use as standard features on cars, such as parking assistance and cruise control systems.
The big challenge lies in the software programming of onboard computers that must enable the car to respond to any eventuality on the road.
Daimler is among the leaders in developing self-driving technology. Its main rivals include Google, which is testing driverless cars fitted with its Google Chauffeur software in California and Nevada.
Google has modified a number of models including the Toyota Prius, Lexus and Audi TT and they have clocked up more than 650,000 kilometres without accidents.
This summer, Daimler successfully completed a 103km journey in a modified S-class limousine equipped with eight radar systems and three cameras.
It mastered city traffic and busy roundabouts without “the driver” having to use the steering wheel, brake or accelerator.
“There were no dangerous situations,” says Daimler board member Thomas Weber, the head of group research. “A lot of research work is still needed, but it showed us how extremely far we’ve come.”
He conceded there were some situations during the journey when the test driver had to intervene. At one point, an elderly lady at a pedestrian crossing beckoned the driver to pass before she walked over the street.
“He had to intervene manually and drive on because the car would otherwise just have stood still until she was gone.
“The driver also had to step in when the car entered a narrow road partly blocked by a luggage removal van. He had to steer the car on to the pavement to pass it. Using the pavement wasn’t programmed in to the system.”
Computers have yet to be taught to master situations of human interaction that are part of everyday driving – one driver gesturing to another to let him enter a lane, or flashing headlights to warn of danger.
There are other drawbacks. “It is very difficult for sensors to ascertain the softness of obstacles, like a plastic garbage bag by the road,” says Christian Gerdes, a professor of mechanical engineering at Stanford University.
At present, a self-driving car can’t distinguish reliably between a bag, a cardboard box and a pram in the road and would perform an emergency stop in each case, thereby possibly causing a rear-end collision. Poor weather conditions are a further problem. Snowflakes can block or confuse sensors, for example.
Developers are divided on how long it will take before self-driving cars go on sale. Google believes automated vehicles could be commercially available in five years.
Daimler says that is too optimistic. “We don’t think the Google story is that realistic,” says Markus Maurer, a professor of electronic vehicle systems at the University of Braunschweig, who heads a research team sponsored by a Daimler foundation.
The German company expects a gradual process of automation whereby cars will be able to do more things themselves. “I think in 2020 we will be very much further and that cars will offer autonomous driving in very many situations if the driver wants it,” says Professor Weber.
Convenience is a major factor, but the overwhelming argument in favour of self-driving cars is safety. Some 1.2 million people die in traffic accidents worldwide each year, and in the overwhelming majority of cases, the accidents are caused by human error.
The automobile has, in fact, caused more deaths than any other non-military invention such as aircraft, electricity or nuclear power.
As car ownership surges in emerging economies, the death toll is destined to keep on rising. Aircraft and trains are thousands of times safer than automobile transport because they are steered by trained experts – not by millions of individuals who may be too distracted, too tired, too reckless, too short-sighted, too aggressive or too drunk to drive safely.
Trains and aircraft are themselves largely automated. When the pilot takes off, he switches to autopilot. The drivers of modern high-speed trains do little apart from occasionally pushing a button to inform the locomotive they are still conscious.
The cars modified by Google require human intervention only once every 80,000km on average, says Sebastian Thrun, the German-born Google engineer leading the project. That’s still too often. The average has to fall to one in millions of kilometres for driverless cars to be commercially and legally viable. But according to Mr Thrun, robot cars are already better than human drivers at spotting traffic lights and avoiding rear-end collisions.
They are also too polite. For safety reasons, they are programmed not to weave into lanes occupied with dense, fast-moving traffic, even though that’s a standard manoeuvre that has to be performed by any car seeking to enter a busy motorway.
The market launch of truly driverless cars will mark such a deep change in transportation that it will have to be preceded by public debate.
Legislation will need to be changed to permit driverless cars – at present, European Union laws call for drivers to control their vehicles at all times. Fundamental legal questions will need to be answered, the main one being who is liable for accidents involving self-driving cars – the owners or the manufacturers. And what death toll would society be prepared to tolerate in a world of autonomous vehicles? The first dead child may provoke a public backlash powerful enough to stop the trend in its tracks.
Three American states – Nevada, Florida and California – have already passed laws permitting driverless cars for testing purposes. The California state legislature, keen to help the industry, has passed a law requiring the Department of Motor Vehicles to create standards for autonomous vehicles by January 1, 2015.
The software will somehow have to be programmed to think ethically– that is possibly the biggest technological challenge. After all, the onboard computers will at times need to take life-and-death decisions. Should they be programmed to protect themselves from damage at all costs, or should they factor in the need to protect the lives of people outside the vehicle too?
A human driver may take a split-second decision to collide with another vehicle rather than with a pedestrian. Will onboard computers be programmed to take such decisions?
“Autonomous vehicles don’t eliminate human error – they shift it from driver error to programming error,” says Prof Gerdes.
One thing is clear – driverless vehicles will have to be able to drive safely without any supervision at all. “Humans are very bad supervisors,” he adds. “If you put them in an autonomous car the chance that they will pay proper attention is very small.”
Given all the challenges, a “big bang” introduction of autonomous vehicles seems less likely than the gradual transition Daimler expects.
The German car maker, which spearheaded automatic safety features such as anti-lock braking and electronic stability control which prevents vehicles from swerving during sharp braking, pledges that in the foreseeable future the drivers of its cars will always have the opportunity to override any onboard automation and take the wheel themselves.
“The opportunities are incredible,” says Prof Weber. “People get tired and make mistakes, but the system doesn’t get tired. We’ve just got to make it fail-safe. ”