By Professor Paul Newman
Machine intelligence is coming to drive us. This robotics-science technology will not arrive unannounced in our cars. We will not overnight have no need to learn to drive. Its just that we won't have to be driving all the time in all places.
Current robotics research will facilitate and induce a competitive struggle between car manufacturers. Their products will offer up an "autonomy option" to the driver with ever increasing frequency. Just like all great technology the trick and reward is giving people a new choice. Future vehicles will frame our user experience as "Drive if you want, but you don't have to right now". The race will be on to increase the duration of "right now".
We will hear about the advent of this technology for years and years but it will come nevertheless. There will be the traditional flagship announcements of both standout and standalone pilots involving breath-taking feats of engineering prowess. Manufactures, researchers and startups will all join in the cacophony. Exciting times.
Smart transport is perhaps an unusual domain to be contributing to. The transport market is vast and will only get larger as our aspirations increase. The mechanical side of the problem is well advanced. For one thing cars exist and cars are so, so cheap and so, so reliable. Furthermore we've already spent billions on building supporting infrastructure (roads) and have a enormous commercial sector set up to support car users. In fact, all that is missing is the software to make the cars suitably smart. We are, in the simplest of terms, just one text file (a computer program) away from self driving cars.
But here, in the midst of all this excitement it is worth sounding adding cautionary minor third in a chord often produced by the robotics research community. These folk have contributed the most to driverless car technology to date and they are accustomed to their creations being taken as evidence of science fiction having become reality. That text file I mentioned earlier is hard to write. The devil is in the detail and the detail is devilish.
We must understand that some fantastic challenges remain in bringing robust and flexible driverless technology to our forecourts. The retail prices must make sense and this impacts sensor and algorithmic choices which in turn influences reliability. The man-machine interface must be right - how does a car swiftly bring a human "back into the loop" when autonomy is failing? How do you form trust between man and machine? And of course underpinning this all is the thorny issue of enduring machine perception.
We will, I'm sure, gradually and iteratively solve these problems, but we should not understand the current hullabaloo about driverless cars to mean "almost in production". Instead we should see it as an opportunity to consider the implications and benefits (congestion, pollution , safety, time saving and fun) as we mature from today's lane-keeping, parking and intelligent braking to full-blown, long-term autonomy.
I do not see a mass produced end-to-end autonomous car - one that has a single "drive-me" button which can be pressed whenever the passenger demands it - appearing anytime soon. I have more faith in a graduated model, one in which the car itself offers autonomy when it believes the process will be safe, achievable and insurable. In the youth of the technology this will be for only short stretches of road at a time in good conditions. But as the technology improves and adoption increases, as insurers get more data and people get busier the duty cycle will increase.
Crucially no dependence on external infrastructure, no cross sector technological co-ordination is needed. The cars will operate by themselves and if needs be, independently, on the roads we already have. No government or state intervention needed. No massive spend on infrastructure or large scale integrated IT project. Let the cars drive themselves, in some of the places, some of the time.
But have no doubt that it is computing that will revolutionise and drive the future of transport - just like it has healthcare, finance, entertainment and communications. As a nation, we need to be the best at writing those text files. We really do.
Professor Paul Newman is the Project leader of the Mobile Robotics Group at the Department of Engineering Science at Oxford University. The team is developing a self-driving car that interprets data from sensors so it can 'see' its surroundings and negotiate them safely.
Hear Paul speak in London on the 4th November at "Smarter Mobility: An evening of debate", which is part of the Switched On series of talks and debates from Intelligence Squared, supported by Shell.
Follow the conversation on #iq2mobility