Lander's Lobby - Isn't It Called A Railway..?
THERE HAVE BEEN SOME WORRYING developments on the ‘autonomous vehicle’ front...
In his recent budget, the UK’s Chancellor, George Osborne, announced that the government’s ‘Intelligent Mobility Fund’ will be used to sponsor tests of selfdriving vehicles on British roads. Osborne claims that this will put us at the forefront of “an exciting new technology” - despite the fact that the vehicles themselves will be manufactured elsewhere. Ironically, while the Treasury is busy chopping money from the Motability scheme (which provides cars for disabled drivers), there’s evidently enough spare cash lying around to fund computer driven lorries on the M6 motorway.
That project - starting later in the year - will see up to ten lorries, linked only by computer, running together just a few feet apart. One driver, in the lead vehicle, will control them all (although there will actually be drivers in the others, ready to take over in emergencies). The plan is to cram even more vehicles on to our overcrowded roads, and, at the same time, to make fuel savings by exploiting the aerodynamic efficiency of close spacing. (I can’t help thinking that we’ve already invented this - isn’t it called a railway?)
Britain’s motorways, with their relatively frequent entry and exit points, are undoubtedly the worst in Europe for such ‘road trains’. There are well-founded fears about other vehicles getting stuck on the outside of one these convoys, unable to pull across safely to access a slip-road. The response to that concern has been a suggestion that the convoys would either use a dedicated lane or would be limited to travelling in the middle lane of a three-lane motorway. Now; the chance of extra lanes being added to all UK motorways must be next to zero - so, unless our ‘rules of the road’ are modified to allow passing on both sides, the inside lane would effectively become redundant. Even fewer vehicles will use it if nose-to-tail lorries are permanently occupying the middle one.
Anyway, it is said that the only stretch of motorway suitable for these experiments will be the top end of the M6 - north of Lancaster - where traffic is less dense and there are rather fewer entry/exits. Therefore, it would appear that the government is promoting a potentially dangerous excursion into a totally unnecessary form of automation, which (even if successful) won’t actually be feasible in this country! Well, apart, that is, from one short length of existing road near Carlisle...
Meanwhile, over in the USA, Google has admitted that one of its self-driving cars has been involved in an accident that (for the first time) was caused by the car’s computer programming. Apparently, the car had entered the correct, right-hand lane for negotiating a set of traffic lights. The sensors then picked up an obstruction ahead, and the computer decided that the car had to move into the left lane. A bus was detected, already in that lane, approaching from behind. Remarkably, the computer assumed that the bus would slow down and allow the car to pull out in front of it.
The passive human ‘driver’ also made the same assumption and therefore didn’t over-ride the automatic controls. But the bus carried on regardless at about 15mph. Result: one badly dented car and several angry bus passengers.
There are some interesting points here. First: the human driver also thought that the manoeuvre was legitimate, and would presumably have done exactly the same thing in an ordinary car. Second: Google (who must, presumably, have accurate data) say that the car was doing 2 mph at the time of impact.
So did the car pull out much more slowly than the average driver would have done? (There’s evidence to support that theory: almost all previous accidents to Google’s fleet have been when their cars have moved away slowly from a stop, and have been ‘rear-ended’.) Most interesting of all, though, is that the computer seems to have been programmed to drive badly - to think, in effect: “I’m sure it will be okay; that bus is bound to let me in.”
Google’s response included the following statement: “From now on, our cars will more deeply understand that buses (and other large vehicles) are less likely to yield to us than other types of vehicles, and we hope to handle situations like this more gracefully in the future.”
Even ignoring the strange use of the English language, this is extremely alarming. The obvious inference is that Google’s computer algorithms will be tweaked to avoid accidents with buses, but that confrontations with smaller vehicles are still considered acceptable. Chicken runs with cyclists... But in point of fact (in America as in Britain) no-one has to ‘yield’ to a vehicle about to make a lane change - unless, of course, it is a special filter lane with signs indicating priorities. Under any other circumstance, the vehicle already in its lane has right of way - which can, though, be relinquished purely as a matter of courtesy.
Anticipating the behaviour of another driver is almost subliminal: speed; road positioning; signals; make and condition of vehicle; age and gender of driver. Plus, checking to see if the driver is monitoring road hazards or chatting to passengers - even noting whether or not there’s a distracting ‘dangly’ hanging from the interior mirror. The human brain is capable of assessing all that information instinctively, and the likelihood of a computer being programmed to respond to those subtle nuances seems very remote.
That’s not to say that it won’t happen. Although such comprehension can’t be pre-programmed, a computer may be able to learn it - just as we do. Concurrent with the Google collision (caused by ‘old’ computing, if you like) came a triumph of ‘new’ computing - of artificial intelligence. A computer has beaten the world’s best ‘Go’ player, and, in order to do so, it had to learn increasingly complex levels of move and counter-move as the game progressed.
Perhaps it’s just a matter of time before autonomous cars learn to assess the probable driving traits of other autonomous cars.
David Landers - [email protected]
Lander's Lobby, from p. 10 of issue April 2016
Subscribe now - http://classicmotor.co.uk/classic-motor-monthly-subscriptions