A £1.4m EPSRC-funded research programme led by Oxford University will aim to give vehicles the ability to navigate roads autonomously with data from onboard sensors, cameras, radars, lasers and downloaded information such as aerial photos and on-the-fly internet queries.
Principal investigator Dr Paul Newman from the university’s robotics-research group said the idea will be to develop a car with a central computer that is malleable like the human brain and capable of continuously ‘learning’ from experience and interpreting situations. This will require the development of probability statistics for analysing information from onboard technology and using machine-learning techniques to build and calibrate mathematical models that can explain the vehicle’s view of the world.
Current navigation systems in cars such as GPS, Newman said, do not give vehicles the accuracy needed to know how and when to move safely or make autonomous decisions.
‘GPS only gives you knowledge to maybe five or six metres,’ he said. ‘I wouldn’t want to be going around a roundabout with a car that only knew where it was to the nearest five metres.’
The new autonomous vehicle technology developed through the five-year research programme will be tested on a car driven around a privately hired race track. With safety being a top priority, Newman said a number of issues will need to be addressed. Some of those on his list include: Can the vehicle navigate autonomously at night? Can it deal with changing seasons? How would it handle a change in the road structure? How well does it detect pedestrians?
In addition to Oxford University, the research will receive support from partners including BAE Systems, Nissan, the Department for Transport, Navtech Radar, Guidance Navigation and MIT.
Newman said the concept of autonomous cars is not such an unrealistic idea, considering the trend for intelligent vehicles. ‘Cars are going to have infinite computation and access to infinite storage because they are going to be connected to the internet the whole time, which is the cloud,’ he said, referring to cloud computing.
There would also always be the option for the driver to take over if a vehicle could not handle a particular situation. Newman envisioned the car being there only to help the driver and it could be deployed into autonomous action by sending a message via mobile phone.
‘You’re sitting on the Embankment trying to get to South London and, quite frankly, I lose the will to live,’ he said. ‘I think “I’m sitting in traffic, why can’t I watch a movie, why can’t I answer my email?” It’s because I have to sit here watching the car in front and yet all the cars know what’s happening around them.
‘Why couldn’t a licence be given over mobile phone to the car to say “OK, you can drive yourself a few miles because all the cars know what’s happening. It’s very slow traffic on a well-known road.”’
Newman also hopes that, with advanced machine-learning techniques, it might be also possible to communicate with the autonomous vehicle as a back-seat driver sometimes does to the main driver.
Newman said the legal issues regarding autonomous road vehicles are unclear at the moment.
‘I’m starting to talk to the DfT about what to do about this,’ he said. ‘Insurance is obviously a big issue. Who is culpable if the machine goes wrong?’
Newman said there are also possible moral dilemmas that could arise. ‘Imagine you have ‘n’ thousand people killed a year on the roads at the moment and you could show the machine was 80 per cent better than a human,’ he said. ‘So only one in five of those people are getting killed but they are getting killed by a machine. Many will say “No, I’d rather get four times more people killed than have any of them killed by a machine”’.
Newman predicts that the moral dilemmas surrounding more autonomous technology will not be limited to roads.
‘I think this debate is going to come everywhere with machines,’ he said. ‘I think it’s going to be in the military domain. What military decisions do we want machines taking in our name? What driving decisions do we want machines taking in our name? What data-protection issues? What medical decisions do I want this computer to make on my behalf if you’re a consultant? It’s everywhere. We’re going to have this debate.’