> In 10 of the 13 simulated contact events, the SDC’s predicted behavior would have, in simulation, caused contact (though 2 of these involved simulated contact with traffic cones).
The 13 events happened in the 15 months from September 2014 to November 2015 (both incl). 424,331 miles driven.
Can the average driver drive 30,000 miles without making contact with a traffic cone or another car?
The broken pylons at my closest bike lane and highway exit indicate that very few drivers can achieve this.
Detail:
I'm speaking in generalizations here, but it seems that small and even less traveled roads that have pylons often have these damages.
Scarily, on a run today, a "lightly" traveled road in a residential area had multiple "stop for pedestrian" signs that had been knocked down. I find this particularly scary on the (non major side street) that I live - where other than residents (300ppl) and 1 business (< 300 ppl) few people should choose to drive - loses these road pylons in a matter of days.
I am asking, if anything, that we take a more critical look at the statement being made every time Chris Urmson opens his mouth. Because when he brags about the safety of his cars, he's ignoring or failing to mention these incidents where humans saved him from having a very, very bad day.
I agree with some of what you have said, but this is the most bizarre statement I have read today.
There are probably hundreds (or more) traffic cones hit for every small child hit by a car, just because there are way more cars per day passing close to traffic cones than to children.
I personally have had my car hit by others 5 times without reporting it to my insurance company (car was parked, hit and run), as the damage was less than my deductible, and I opted to just live with a few dents. Estimating the number of times a car collides with anything per mile driven is very hard.
My intention in comparing a traffic cone to a small child is the actual safety issue. If a SDC could hit a cone, it could also hit a small child. They're small road hazards which are not generally going to be mapped by the car in advance. I don't want to dismiss the fact that the SDC would've hit a road obstacle, just because in that case the obstacle was a cheap plastic object.
I think the point is that if you counted these kinds of incidents in human driver statistics, your miles per accident numbers would come way, way down.
I suspect those average numbers are being skewed by the rural drivers who rack up lots of miles without interaction with other cars.
San Francisco averages 6.4 years between accidents. https://www.mainstreet.com/slideshow/most-car-accident-prone...
That matches roughly with my experience. And, while the Google car isn't driving San Francisco, the Santa Clara area isn't that much more benign.
But then, we can also go into their statistic basing on a single insurance agency, unreported incidents not factoring in, there's definitely a lot of margin of error here.
I'm glad we're talking about it critically though, and not just glossing over the figures.
One of the other potential points is something Tesla said about Geohot's garage-built car. That it's easy to build a car that performs correctly 99% of the time, but it's exponentially harder to build a car that performs correctly 99.9999% of the time. The number of miles may be less material than even I've focused on here.
Google likes to tout the mileage figure, but it ignores the times that it's human drivers take over. The reality is, driving on straight stretches of road is easy. 99% of driving is easy. Figuring out how to evaluate how safe you are in the other 1% of that time is perhaps an even more challenging problem.