On 11 March 2013 Prof Paul Newman will deliver this year’s Oxford London Lecture. Oxford Today caught up with him ahead of the event to find out more about his talk — and the self-driving cars that he’s currently developing.
What will you be talking about in your London Lecture?
The lecture's called Where Are The Robots? We’ve been promised robots for fifty years, but we still don’t seem to have them. So I’ll talk about why building a robot is so staggeringly and enchantingly hard, and why I’ve found that fascinating throughout my life. Even with all the wonderful sensors available, it’s still incredibly hard to have a robot perceive the world around it, let alone interact with it. But things are changing, and we’re finally getting somewhere.
Which will take me to the field I contribute to: trying to help robots know where they are in the world. I’m interested in creating machines that can navigate — by perceiving objects, understanding spaces and moving accordingly — all using cheap components. We can all walk home from the shops without GPS, so it’s definitely possible to navigate using only on-board sensors. I’m trying to get robots to do the same.
Specifically, you’re developing self-driving cars. What are the problems involved in building that kind of technology?
We’ve worked for the best part of a decade on what we call infrastructure-free navigation, where we try and create machines that just use on-board sensors like lasers and camera to help themselves navigate. While it’s fairly easy to train a car to find its way round a track, getting it to do it reliably — in different weather conditions, at different times of day — is much more complex. Perhaps the major difficulty, though, lies in having the car detect the very small cues that humans pick up naturally — say, noticing a gate is open and then inferring from that fact that something might emerge from it. That kind of thing is easy for humans, but not as simple for a robot.
You’ve recently taken delivery of two Nissan Leaf eco-cars. What are you doing with them?
We want to create technology to support autonomy in cars which uses cheap commercial sensors. We’re not exactly talking about pressing a button that makes the car drive on your behalf here, because that might not be safe. Instead, as I see it, these cars will start driving some of the people, to some of the places, some of the time. So you’ll buy your car and spend a little extra money to get an autonomy pack. You’ll drive home, and be disappointed because the green light doesn’t come on offering autonomous driving. But then, one Thursday evening, the light will pop on, and the car will say “I can drive through this traffic jam for you”, because it knows the roads, and the conditions, and has even obtained insurance for you via the internet. Then you press the button and it’ll drive. Maybe ten minutes later it asks for you to take over again; maybe it drives you all the way home. That’s the way driverless cars will work.
And how does the technology work in practice?
In reality, most of the time you only ever want your car to work in places that it — or at least, another car — has already been. In a city, for example, many people drive through the roads every day. If those cars have cameras or laser sensors, they can create very accurate maps of what spaces looks like. We take that data, and build detailed models of what cities, towns and countryside looks like. If your car knows roughly where it is in the world to a few metres, from GPS data or some other location system, then it can compare those maps to the data it’s getting from its on-board sensors — when the two match up, it knows exactly where it is, and can drive itself. When they don’t match up, a human has to drive.
What are the barriers to driverless cars becoming commonplace?
Insurance is a big issue; clearly insurers need to get behind it if it’s to ever work. But you can already buy a Ford Focus that parks itself, so I think it’s really only a matter of time before they come round to it. There might be some social resistance, too; some people might be sceptical about the idea of driverless cars. But i dare say people were sceptical when combustion engines replaced horses — this time we just have different nouns and verbs. To an extent, though, those problems are there for the car companies to deal with. We’re primarily interested in the academic questions.
Finally, how feasible is it that cars will be driving us around in the near future?
It’s definitely going to happen — it’s harder to believe it’s not going to happen than anything else — but predicting the future is difficult. Eventually, my goal is to provide an autonomy pack, available on everyday cars, for $100. We’re nowhere near that, of course, but I think we could achieve $1,000 fairly quickly. We’re certainly talking years, not decades.