I read somewhere that Intel, which has a division called Mobileye developing the technology for self-driving cars, tested the video that was released by the police with their system. It spotted the pedestrian one second before the collision. Another indication that the accident was avoidable.
I’m not suggesting we empirically determine how deadly the cars are, just noting the problems with extrapolating.
It was in this article which someone linked to earlier, which said the standard collision-avoidance system in the Volvo was disabled on the Uber vehicle.
Actually, I read the news about the Intel technology in a different article.
Indeed, it’s a classic spherical cow solution. The main applicability to the current situation may be to the speculation that Uber’s AV team might have had a “spherical cow”, quick-n-easy mindset.
FWIW, I suspect AVs will have more success in “limited access” situations, like autonomous long-haul trucks sticking to the interstate. Far fewer pedestrians there.
One of the earlier links in this thread mentioned other Uber collisions. One really confused me. The Uber AV collided with a car making a left turn. The driver said he did not see the AV coming (thought it was clear), which was moving at about the speed limit (38mph). So this raises a few questions. If you are approaching a messy-looking intersection situation, you are probably going to be on your guard and ease up a bit on the throttle, which does not at all seem to be what this Uber AV did.
Additionally, how is it that the driver could not see the AV? I understand that assertions made in these kind of events may be unreliable or confused, but it sounds like these vehicles absolutely need to be highly visible. Humans seem to sense when someone is looking at them (some strange combination of perception factors), but when the car has no driver, this human capability is thwarted.
The Uber AV ethos seems to prioritize getting the passenger to where they are going. Forward progress is crucial to them, even if it means shaving the edges of caution off the system. Personally, I think this corporation needs to fail, go under and be replaced by a better model.
Sure. I could imagine the alternative, perhaps under the influence of a (mythical) perfect truth serum, being “I figured he’d yield once he saw I was in the intersection.” I certainly see enough of that kind of intersection chicken without any intelligence (artificial or otherwise) involved.
Well, the Uber AVs would be no less visible than any other Volvo XC90, or any other mid-sized SUV for that matter. It’s not like Uber invented stealth at the same time. However, the issue of “no driver looking at me” would make me more cautious: there have been many times I’ve said “that guy doesn’t know I’m here, and he’s going to do something on that basis”. Stuff like accelerating straight ahead through an very stale yellow light when I was thinking of turning left, or almost changing lanes into me. To other drivers (paying attention, and (as you say) sensing the lack of human attention in the other car), the lack of a human driver would make them more cautious, not less.
Yeah. They’ve already demonstrated ample “first at any cost” mentality. I used to think that stealing technical trade secrets and proprietary information from competition is the lowest they would go. I was wrong.
Can you tell me where that statistic comes from? There were around 6,000 pedestrian deaths in the US in 2017, and roughly 3.2 *trillion *vehicle miles, traveled, so (very roughly) one pedestrian death per 500 million miles driven. Total AV miles logged aren’t even remotely close to that.
Correct. Exact numbers for miles driven on public roads by all AVs aren’t available, but by adding up the miles reported by the biggest players, it is less than 10 million miles. With one death per (less than) 10 million miles driven, the AV rate is 50 times worse than regular vehicles.
Ah, got it. I was misreading. :o
Okay, completely unrelated, but I hadn’t really thought about this statistic before.
The US drives 3.2 trillion miles in a year. A light year is “only” ~5.9 trillion miles.
That means that if you total up all the speeds of all the cars at a given time, they are moving a pretty significant fraction of the speed of light. In fact, given that most driving happens during the day, the “total speed” may even exceed it.
Like I said, not related, just thought it was interesting.
That’s not the way motion works, but it reminds me of a story Richard Feynman told when he was on a board reviewing math and science textbooks for the state of California, and one book had a problem that summed up the temperature of stars of different spectral classes (including a “purple star”) to get the “total temperature”. Apparently he was not amused.
Stranger
Of course not, and I was just making what I found to be a humorous observation. If you have 2 cars traveling at 60mph, their “combined speed” is not 120mph. OTOH, over an hour, their combined distance traveled is 120 miles.
I just hadn’t seen miles driven expressed that way before, and my first thought was, “Wow, that’s the best part of a light year”.
I see pedestrians on the highways around here every day, and I only spend 1 1/2 to 2 hours on the road, total.
These involve road construction crews, Highway Patrol traffic stops, motorists on foot because their vehicle ran out of gas (or hit another car), tow service personnel, and so on. That’s just the human’s in the road.
I would respectfully request that self driving cars to stop for other obstacles, too. Moving or non-moving.
Just looking at the case here, in our OP, and watching the released video, it seems to me that the human safety driver realized a bad situation had developed once they actually looked up to the road.
If the safety driver had not been trying to monitor/work while distracted, he would conceivably have stopped the vehicle in time, IMO, and no injuries would have resulted. Thus, we might never have heard of this incident. Uber might not have discovered that [whatever this flaw is] exists as a bug to be squished. That’s actually a tiny bit sobering, for me. 
I would hope that they have logs of anytime the human has to take over to see what the car was not capable of doing.
Dammit. I forgot about those. Sigh.
The day I see an AV perform satisfactory in, say New Delhi, I’ll buy into the idea.
Even New York City!
Also, an AV that will work well in Canada probably won’t work in Mexico. An AV that will work in Germany probably won’t work in Italy. AVs will have to be specifically tailored to each country, and even to individual cities.
Of course they’re doing that. It’s the entire purpose of putting these cars on the road in the first place.
The one reason I’m gung-ho about AVs is the fact that what they learn about safety can be applied to all the cars. When I was a teen driver, I had my share of near misses and minor accidents. I learned from them, but nobody else could, and I had to go through that risk before learning it in the first place.
However, 13 miles per intervention? GTFOff the road, a 16 year old with a brand new learner’s permit should do better than that. That’s your target? Cars that crappy belong on a test track, and governments should demand far better performance before allowing these things on the road.
Uber settles with family of victim.
So the homeless woman had a husband and daughter but was still living in the street? Some family (not that it gets Uber off of the hook for running under-tested AVs around in the public).