On Valentine’s Day this year, a Google self-driving car ran into the side of a bus as it negotiated roadworks on a busy road in San Jose, California. No one was hurt, but it resulted in the bus being taken out of service. To date, Google’s autonomous cars have been involved in eighteen accidents since the company started testing the concept in 2010.
The problem is autonomous cars need to be able to detect hazards, which mean they have to be able to pick out obstacles from a static background. The only way they can do this is by detecting a change in that background. Essentially this means they need accurate up-to-date maps.
This should be easy for Google. After all they have their own Google Maps and Street View services. But there is a problem with this: Google only surveys routes occasionally; indeed many roads have only been mapped once.
But help is at hand from an unexpected source. In August 2015, a consortium of big car makers, including Daimler, Volkswagen and BMW, bought a rival mapping service called Here Maps from Nokia for a reputed $3 billion.
So what do the car makers want with a mapping company? The clue for this comes from a statement on the Nokia website: “HERE is developing a location cloud that harnesses the power of data generated by vehicles, devices and infrastructure to deliver real-time, predictive and personalized location services”.
In other words they plan to put Here Maps’ mapping sensors in their new cars. Instead of a few Google cars on the road, we will see thousands of ordinary cars building maps as they drive. This will help driverless cars deal with the unexpected. Instead of swerving round roadworks and crashing into buses, they will have their maps updated in real time so they can safely negotiate obstacles.
And it’s not just the Germans who plan to crowdsource mapping data. General Motors and the electric car company Tesla have teamed up with the Israeli company Mobileye who make image recognition software. As it says on their website “Using sophisticated vision algorithms, Mobileye collision avoidance technology is able to ‘interpret’ a scene in real-time and provide drivers with an immediate evaluation based on its analysis”3. Indeed, Tesla Model S cars are already equipped with this software.
Meanwhile, over in Japan, Toyota is getting in on the act. At a cost of $1billion, they have set up an institute to research into the use of artificial intelligence in autonomous cars. The institute is run by Gill Pratt, who used to work for DARPA (the agency that invented the Internet, by the way). During an interview in 2015, Gill was asked “Where do you hope Toyota’s robotics program will be in five years?” in reply, Gill stated “Our long-term goal is to make a car that is never responsible for a crash.”
Science fiction is full of stories about robots taking over the world. Now it seems to be coming to pass that these robots won’t be ray-gun toting androids, but artificially intelligent cars.
So what does this mean for Google? Well they have responded to the accident in San Jose by programming their cars to realise that large vehicles like buses are less likely to give way to cars. Who knows, they may start programming their cars with common sense next.
Based on an article from New Scientist magazine 16 January 2016.