Police Say Uber Is Likely Not at Fault for Self-Driving Car Fatality in Arizona

Besides, some commenters already pointed out that the google maps of the road in question says that the speed limit is actually 45 northbound (which was the direction the car was going), whereas the 35 number is the speed limit southbound.

I'll ignore ACC for a moment and talk about the modern implementation of regular cruise control, however. You'll have your target speed stored in a control unit, which cares about 3 to 4 things: your set speed, your current speed, the flow of fuel to the engine, and in newer models, access to the brakes. So, how do we determine the current speed? The drive shaft will have a magnet mounted on it, and each time it spins around there is a sensing coil that sends a signal to the control unit. Based on the frequency of these signals, it determines what your speed is, and compares it against your set speed. If it is low, it will send a signal to a servomotor that is connected to the accelerator, sending more fuel into the engine, hopefully increasing your speed. If it is going to fast, it will cut fuel, and on some models, apply breaks if you are more significantly over the set speed.

However, this is a pretty dumb system. It only knows how fast you want to go, how fast you're going, and whatever the formula is that it follows for increasing the flow of fuel to the engine. It doesn't know about millions of other factors that affect how fuel efficient that acceleration is going to be. It is constantly readjusting based on the new speed readings. To my knowledge, all manufacturers also go for something that provides a "smooth" driving experience for their adjustments, rather than how to most quickly achieve the set speed. When you take the limited data available for a cruise control system and combine it with built in smoothing to keep the ride comfortable, you end up in a situation where the cruise control is making frequent adjustments that end up on one side or the other of your target speed. The lack of perfect information means they purposefully do not attempt to perfectly match the target speed - attempting to due so would require so many adjustments that it would be pretty terrible for fuel economy.

Now, hopefully, the Uber vehicle is going to have even more data available than my ACC equipped car. But even with my ACC equipped car, almost all of the work has gone in better determining what the set speed should be, and significantly less on making sure it hits it exactly. I expect this is probably the same on the Uber vehicle. Because quite frankly, it's just not that important. To be quite blunt, the difference between being hit at 35mph and 38mph is probably just how far your corpse flies.

As for saying they should drive 3mph slower than the speed limit, then they might end up going 6mph under the speed limit instead of 3mph under.

I suppose my fundamental argument in the previous post and now this is: What outcome is expected from them being able to stay perfectly at 35mph instead of having enough variance to hit 38mph? What benefits are there? A few mph makes the most difference in damage from an injury standpoint at very low speeds, and these are the speeds that variance is almost always significantly lower - there's no evidence here that Uber cars are going 8mph when they're meant to be going 5mph, for example, and this would not be the type of variance we see in cruise control systems in general - and very little difference to the outcome at high speeds, where the greatest variance is likely to be shown.

My understanding, which may be incorrect, is that the 38mph number came from GPS recording, and that "cruise control" in Uber's self driving vehicles is not based on GPS speed, due to the impact poor/loss off GPS signal would impact it.

>Modern cruise controls are perfectly capable of maintaining set speed, even at a considerable downhill.

I'm not particularly handy with repairing cars, but I have a pretty decent understanding of how cruise control systems work due to nerding out on it when I bought a car with adaptive cruise control. Everything I have read on the design of cruise control says this simply isn't the case, because it isn't even a design goal. Much more important in the design is keeping close to it while maximizing fuel efficiency and driver comfort, and even in modern cruise control not all systems have access to the brakes, which would be required for keeping it at the set speed going considerably downhill. Anecdotally, all of the cars I have owned fit this and do not keep my vehicle at exactly the set speed, including the most recent one with ACC.

In theory this may be true but it's absolutely not in practice. It's not uncommon to see people at least 20mph over the limit on motorways. On the M4 that is practically the norm.

A speedometer's reading is measured on the gearbox, at the output but there are a myriad of little things after that including variance in the final drive, tire size and tire pressure all of which can make the "calculated" speed and the real speed to be within a few % of each other. All of these components are also affected by wear too.

At the moment you need a GPS to conclusively measure a car's speed with >99% accuracy but GPS signal isn't available/sufficient everywhere. One could argue that we should create some sort of highly accurate local system that measures speed and we certainly can but we'd be trying to solve a 0~10% speed measurement problem: we'd get no real value from it.

In other words: a car measures how fast a specific cog is rotating, adds a few variables and estimates how fast the car is going based on that.

edit: and all of this assumes the car is going straight, calculating speed when the car is curving is more complex.

Oh, certainly, and I apologize if my argument came across as "It's not possible for people to not speed" or "People only speed because of the properties of speedometers or cruise control"

My point is that speed within a few mph is already inaccurate, and that 3mph is pretty much within the current margin of error.

>speedos universally over-read in my experience

I don't know that there's a common leaning one way or the other in actual variance due to speedometer construction, but over-reading is the most common occurrence due to tire pressure being low causing it, and a huge amount of people drive with low tire pressure. Overinflated tires, or moving to larger tires in general, will result in under-reads.

>I would expect a fully autonomous vehicle to be able to accurately obey the speed limit and I suspect the reason it wasn't/doesn't is because that behaviour has been programmed in or developed.

I don't agree. lagadu covers this plenty well in his comment, however.

As you go faster, your energy goes up exponentially. This is why there are bitter fights about increasing the speed limit on roads; they become exponentially less safe (big caveats apply, naturally).

And this is considering the fatality rate for all human drivers, including those that are drunk, playing on their smartphone, sleepy, speeding, street racing, etc. I would not be surprised if those cases account for a very sizable majority of fatal car accidents. Now imagine you would reduce those numbers by relatively simple measures that don't involve AI. Let the car enforce speed limits, mandate smartphone companies to have a 'car mode' that cannot be shut off, mandate attention tracking capabilities for cars, dead-zone indicators, automatic emergency braking, etc. Maybe even mandate an alcohol lock, or at the very least a black box that records driving behavior so it can be analyzed in case of a crash, with sever penalties after the fact if the black box shows signs of drunk driving, etc.

The list of things you can do to improve road safety without requiring AI cars is endless, yet billions are invested in trying to fully automate driving. This just shows that self-driving cars are not about safety, but purely about business opportunities.

Also of note:

> Arizona has experienced a surge of pedestrian fatalities recently, with more than 10 in a single week of March in Phoenix alone. The state has the highest rate of pedestrian fatalities in the United States.

Why? What you listed seem like conditions that should be easy enough to recognize even with simple algorithms.

At town speeds I would expect it to use the brakes a lot more readily. At 30mph in 3rd (2000rpm) or 4th (1500rpm) there's not really all that much engine braking, and wind resistance won't slow it much either.

(Mine is a 2010 model, but I think this stuff was introduced in 2004-6, something like that... it's not exactly new technology.)

(Once or twice when going downhill without cruise control, just using the brake pedal, it "downshifted" automatically while I had the brake pedal pressed.

CVTs are computer controled and better than manual/standard for gas mileage, but with the downaide that you can't tow heavy loads.)

-- 55% of drivers regularly exceed the speed limit, meaning that the speed limit represents the 45th percentile speed

-- driving between the 50th and 90th percentile speeds is the safest

Taken together, driving at the speed limit is significantly unsafe.

The point of a self-driving car is unburdening the user from the task of driving it, allowing them to engage in other activities if they chose to do so.

That it has the potential of being safer is a nice perk but definitely not the primary goal. Right now the challenge is whether we can make them as safe on average as human drivers do, that way we can avoid effectively sacrificing people for the convenience such a system would provide.

That's a strange way to put it. Surely if someone is not driving their car, they can't be responsible if the car kills people (assuming of course the car was well-maintained and the model was legal). I don't see how that's a bad thing, since they have zero control over the situation.

Do you mean that it could make car users feel less responsible for deaths, even though cars will presumably still kill people?

How about when merging, or changing lanes? The speed limit is not _always_ safe or practical to obey strictly.

That being said in the United States non-human persons such as corporations have no such rights.

https://www.youtube.com/watch?v=UEIn8GJIg0E

It's enormously unlikely that the Uber car failed this badly. The most plausible scenario is that the car did brake but just didn't leave skid marks. The car not recognizing a human in its path would be a shocking failure.

Also I know I wouldn't be programming my self-driving car to jump to a different lane every time it saw a pedestrian in the distance.

But then again I wouldn't be programming my self-driving car to speed either...

(Yeah, I get what you're trying to say, but I couldn't resist.)

"This study examined the effect of brake and accelerator pedal configuration on braking response time to an unexpected obstacle. One hundred subjects drove in the Dynamic Research, Inc. (DRI) Interactive Driving Simulator through a simulated neighborhood 21 times, each time with a different pedal configuration. Each subject was presented with an unexpected obstacle only one time, for one of three previously selected pedal configurations, to which he or she was instructed to brake as quickly as possible. Foot movements were recorded with a video camera mounted above the pedals. Data were analyzed manually, using time and course location information superimposed on the video data. Response times were analyzed using ANOVA to determine effects of pedal configuration and various driver factors. Response times ranged from 0.81 sec to 2.44 sec with a mean of 1.33 sec and a standard deviation of 0.27 sec. There was no significant effect of pedal configuration on response time. Driver age was significant, with increased age corresponding to increased response time. Car normally driven, gender, driver height, and shoe size had no significant effect"

[1] "Braking Response Times for 100 Drivers in the Avoidance of an Unexpected Obstacle as Measured in a Driving Simulator"

https://www.researchgate.net/publication/274973324_Braking_R...

Because if he had seen her before hand (i.e. she didn't "come out of nowhere"), he would have presumably taken control, breaked, swerved, etc.

We already know that the car appeared not to have done that, so the profession from the driver literally adds no further evidence, since that is exactly what we would expect every driver who mows down a pedestrian unwillingly to say...