This weekend I drove a 2006 Toyota Prius into a ditch.
The snow began falling around 11:00. About an inch inch lay on the road, atop ice. The first few flakes had melted and frozen, or perhaps a drizzle of rain had come before the snow. I wasn’t entirely aware of the conditions before setting out in search of lunch. And I certainly wasn’t aware of how a Prius performs in the snow. That is to say, it doesn’t handle as expected.
The hazards of driving in snow are in turning and in controlling speed, both applications of a loss of friction. Without traction, hills present a particular difficulty. To control speed going down a hill, one usually, in dry conditions, slows the turning of the wheels by impeding the movement of the wheel with the disc brakes. On snow (or ice or water or wet leaves) this causes the car to slip, so it’s something to be avoided. Alternately, one downshifts to a lower gear and uses the engine to brake. This method is not possible in a Prius, unless you read the manual. Thirdly, one may use gravity to reduce momentum as one reaches the crest of a hill, in order to minimize peak velocity on the descent. There may be other techniques of which I’m not aware. Conversely, the momentum of descending the hill is needed, again because of the lack of traction, to ascend a hill.
In the movie Cars (2006) a race car encounters impossible situations. He drives down a dark road at night, by the light of the moon because he has no headlights. He falls of a cliff cornering on a dirt track. He pulls a paving machine. He tips tractor cows. (C’mon! One can suspend disbelief only so far.) The missing headlights are a mechanical design decision: he doesn’t have them because the chance he’ll need them is so improbable. They aren’t necessary. But he fell off a cliff because he doesn’t know how to drive on a sliding surface. He has to learn.
The story is that a Toyota engineer drove a Siena minivan through all 50 states to get a sense of how it was used and the conditions one might expect. This resulted in features such as the ability to lay a 4×8 sheet of plywood flat in the back, all-wheel drive, and passenger windows which opened. (Though why it took until 2003 for vans to let their passengers breathe is beyond my comprehension. Did no automotive engineer ever suffocate in the heat of the back seat as a child?) Given the poorly functioning windshield defroster, one might suspect that the Prius was only tested in southern California, but it does have an anti-skid feature which, as far as I could tell at the time, consists of flashing a light and beeping at you — and slowing the rotation of the wheels. Along with traction control, it’s intended to keep you on the road and in control of your vehicle. Mostly. The computer is to assist in handling the skid; actually handling it is up to the driver.
It’s said that experience is the best teacher. I’m not a fan of Mario Kart and other racing games. I’m inexperienced, uncomfortable driving with my eyes my only sense. I lose control and crash. In a car there’s gravity. You can feel the weight shifting and move in concert. There’s more to the road than the speed limit, the angle of the curve, the pitch of the pavement. Cars 3 (2016) is a lecture on the limits of simulator design. There’s more to racing than going fast in a circle for a long time: You will encounter unexpected situations and must adapt to them. Though perhaps even the best simulation, limited only by a lack of imagination, cannot adequately prepare you for the Real Thing. One becomes accustomed to the simulation, prepared for the apparently probable and unable to adapt to the unlikely. In this context, the news that DeepMind’s AlphaGo Zero taught itself chess is important and disturbing: it learns; it adapts. After the novelty of autonomous automobile racing wears off, NASCAR fans may dress up in fancy hats and fondly recall the storied heritage of the sport.
But can they drive in the snow?
Researchers at Stanford’s Dynamic Design Lab noticed something.
[T]hey solved a sliding problem when going around corners at high speed by using data gleaned from the minds of racing drivers.
“We discovered that for the drivers it was an automatic reaction that kicked in as soon as the car started to slide,” [Joe Funke] said, “They knew what to do from experience and just did it.
“The car, on the other hand, used a stabilizing algorithm. When we changed it so that it had a set automatic command when it started to slide it definitely seemed to work.’
They encoded the practice that racing drivers had done.
Number One Daughter (17) has been driving, carefully, for almost a year. But she hasn’t encountered a skid yet. Watching videos on how to correct a skid is useful, but it doesn’t replace experience: the correction for a skid is not like linear driving. Where can human drivers get more experience in edge conditions? Why aren’t these techniques taught to new drivers? Why don’t we teach more than the bare minimum needed to operate a vehicle? For that matter, why don’t we teach high school physics on race tracks? Why is it easier to teach a robot?