MikeyMK

''That is the sound of... inevitability.''

McBTC

It's all academic-- any company fielding cars that run over pedestrians will be sued out of existence and not just by victimized walkers but also by occupants who were forced to participate in a killer business model and powerless to stop the wonton carnage... about like finding yourself in the wake like a deranged religious fanatic bulldozing about without a conscience. At least on trains, passengers could have pulled an alarm chain...

RFEngineer

This. If the "safety driver" is already treating the car as infallible, what is the point? This just shows human nature. Human's will either be "engaged" or not. There is no in between.

njkayaker

Bizarre.

You still discussed it and then complain about other people doing it.

How do you distinguish "the very valid points" from ones that aren't?

Presumably, he though all of his points were "very valid" (he said it "made perfect sense").

It doesn't help him (or anybody) to just ignore things that don't make sense.

Maelochs

Ummm ... "wonton carnage" is what happens in Chinese restaurants .... and in WWII Japanese soldiers didn't charge into battle yelling "Bonsai" (unless they were very attached to their pruning abilities.)

njkayaker

There were three points of failure:

1- the pedestrian didn't yield nor did she appear to be looking.
2- the AV didn't take reasonable action (for unknown reasons).
3- the "backup driver" wasn't paying attention.

1- The pedestrian has a requirement (practical and legal) to take due care. She was at risk by her own actions whatever the vehicle was.
2- The AV should have (it seems) been able to deal with the less-than-careful pedestrian.
3- The "backup driver" should have been paying attention.

njkayaker

There were three points of failure:

1- the pedestrian didn't yield nor did she appear to be looking.
2- the AV didn't take reasonable action (for unknown reasons).
3- the "backup driver" wasn't paying attention.

1- The pedestrian has a requirement (practical and legal) to take due care. She was at risk by her own actions whatever the vehicle was.
2- The AV should have (it seems) been able to deal with the less-than-careful pedestrian.
3- The "backup driver" should have been paying attention.

noisebeam

Exactly. I nearly posted the same. #1 is primary root cause, 2&3 contributing factors in that order.

AlmostTrick

Be careful when looking at time stamps. There are several clips per second. I saw four distinct steps "in the car lights". The woman was not walking fast. While timing and walking I get two steps per second, so at least two seconds to react here. I stand by my previous comment because...

While motoring at 40 mph in the dark many years ago, I had a deer leap out right in front of me. It happened so close and fast that it was over before I had any chance to realize what I saw. I braked on instinct and lightly bumped her at maybe 15 mph. I saw her run away, there was zero damage to my car. There was no steering on my part, because as you pointed out, that takes more time.

I highly doubt a skilled, attentive driver would have hit this victim without braking at all, and certainly a machine that is supposed to be better than a human shouldn't have either. Which may have allowed Elaine, who made a mistake, to survive.

I-Like-To-Bike

The apologists want to ignore the issue that the Uber AV cars are allegedly equipped with sensors more effective and capable for gathering data and driving the vehicle safely on the public roads than the apparently cheapo dash cam cameras used to record the two videos released by the Tempe Police Dept. The apologists also choose to ignore that it is quite possible, if not probable, that almost any reasonable driver paying attention to the road would be able to see better than the crappy dash cam used by the Uber vehicle and react better than the Uber vehicle which did not respond at all to a hazard in the middle of the road.

It should also be noted that the two videos were released by the Tempe Police Department, not by Uber, since the police had physical possession of the vehicle and its dash cams and Uber did not have access after the collision. Whatever data Uber has on this incident has not yet been made public but I assume that the NHTSA investigators will obtain it and use the detailed data gathered by Uber from its sensors as well as its other investigative tools to come to a conclusion based on all the available facts and data.

The apologists presumably will continue in their attempts to whitewash Uber/AV failures and promote their dreams regardless of the evidence.

I-Like-To-Bike

He does if he wants to reach a bad/erroneous/biased predetermined conclusion that no human or AV controlled vehicle could have avoided this collision.

Maelochs

Ummmm ... yeah, I am mystified by such hi-tch stuff.

She was about fifty feet from a vehicle traveling at about fifty feet per second when she became visible and was hit about a second after she was visible.

I just did the math.

njkayaker

The video might not be a good representation of how visible the pedestrian was and for how long.

02Giant 

I will continue to believe there are only two valid points of failure. #'s 2 & 3
The vehicle failed to observe what was clearly in it's path, for whatever reasons. Was this an anomaly or does it point to a larger fault?

The woman could have easily been an animal. If it had been, would Uber have even reported it?

InOmaha

Is there any information on how these vehicles are doing in their bad weather testing stages? Driving in Arizona is one thing, but most of the rest of us have large swings in weather throughout the year. For me it's a big swing this week as we had snow flurries Monday, it's 65 and sunny today, and they are expecting 1-2 inches of rain in the next couple of days.


It's one of the big reasons I'm curious how the car "sees", because it's common to have really poor visibility several times a year if a storm opens up on you suddenly while you're driving. That may only occur for 10-15 seconds on the road and you drive out of it. Temporary poor to zero visibility driving capability for rain, or snow whiteouts, will be a big selling point if I'm shopping for an AV. Whether it has a dashcam built in or not to record what I see will not be a big selling factor for me. I can just move the cameras I already own.

Daniel4

The vehicle was driving beyond the lights. At night with only headlights to rely on, the vehicle should have done the 2-second rule. That allows the backup system (the human operator) a chance to respond in case the primary system failed (which it did). We also know the operator wasn't paying attention, so an attention sensor should be added to make sure the backup would work.

I-Like-To-Bike

Your so-called math about when the victim became visible based on the video recorded by the cheapo lo-tech dash cam used by Uber is mystifying.
The cheapo lo-tech dash cam used by Uber may not have even a fraction of the resolution or range of normal human vision.

AlmostTrick

She did not walk (rather slowly) from the center line to almost the other side of the lane in one second. (four steps!) Your math is off.

I-Like-To-Bike

She presumably had to walk, pushing a bike, at least 30 feet from the medium strip across approximately 3 traffic lanes in unobstructed view of the Uber before it struck her.

njkayaker

You'll continue to be wrong in your belief too.

The difference is that, with a small amount of extra care, she could have avoided being killed.

If it had just been a human driver, would the driver even have reported it?

I-Like-To-Bike

I would agree if the headlights actually were as dim to human vision as they appear in the video recorded by the cheap dash cam used by the Uber vehicle. I assume the NHSTA will investigate if the headlights were really as dim as the dash cam portrays.

KD5NRH

Oh sure, try to blame open source. What devalues programming as a career is the "ship it unbelievably broken and we'll fix it piece by piece with weekly updates" mentality. If any other industry recalled their product several times a month as a standard practice, they'd be bankrupt, banished from civilized society and quite possibly imprisoned for fraud.

At least with open source, the end user has the option of finding someone ethical to fix the product.

robertorolfo

Wow, a lot of stuff going on here, and (as predicted), we now have video of the incident.

Starting with the video itself, I don't think it was "doctored" or manipulated in any way, but it does seem particularly dark. You cannot see the victim at all until the beam of the headlights reach her. This seems a little strange, since there were street lights on either side of the road. Yes, she was in a "shadow" between the lights, but in a situation like that you still don't expect total darkness between the areas of light. And although she wasn't wearing any bright clothing, her shoes were white and should likely have been visible sooner.

I know it's impossible to say for sure, but a human driving at that time would likely have their eyes acclimated to the amount of light available and able to see at least something in the area between the streetlights. Also, considering the nature of the road, they would likely have had their high beams on in this situation, which would have made all the difference to an attentive human driver.

The observer is of course an afterthought, as she was clearly looking at her phone, and this only goes to show that putting in observers who aren't required to actively observe is completely pointless.


Moving on to some of the discussion here, it is indeed quite surprising the sensors didn't notice here, even in the dark. This is exactly the situation where we expect the technology to be superior to human perception, and clearly it is/was not.

Do human drivers kill thousands of people on the roads every year? Yes. We all know that. But how is that argument relevant to the idea that testing these AV's on public roads is a mistake?

Most testing is done on closed roads for a reason. No car company would ever get away with testing a brand new braking system for the first time on public roads, would they? Can you imagine if they did, and a fatality resulted, if their argument was that, 'a regular driver with a regular braking system probably would have done the same or worse, so move along...'? Point is, some of you are being way too dismissive here.

With the resources these companies have, it should be very easy for them to test these things in a closed, safe environment. Situations like these could be simulated with numerous permutations, so it would likely even be superior. Surely something like a pedestrian moving perpendicular into the path of the vehicle could, and should, be simulated in a better way then waiting for it to actually happen.

KD5NRH

"May not?" I have yet to find one that shows even a quarter of what's visible to normal sight unless it has an actual low-light mode.

I-Like-To-Bike

I'll take your word on that, I was just trying to be conservative at pointing out the obtuseness of those who assume that the dash cam recording provides an accurate representation of what would be visible to a human paying attention.