Dashcam video of fatal Uber collision released


#1

Originally published at: https://boingboing.net/2018/03/22/dashcam-video-of-fatal-uber-co.html


#2

Cabin footage shows the human driver alert but focused mostly on the dashboard or center console, looking up and noticing Herzberg only a moment before the impact.

Distracted driving. (touch of sarcasm here)

Similar to but different from:

http://legacy.wbur.org/2009/05/12/mbta-revamps-cell-phone-policy-after-green-line-crash

Their response was to ban distracting devices from their ‘drivers’.

maybe these uber drivers need a co-pilot to watch all the gauges and keep in touch with corporate, so the pilot can watch the road?


#3

It’s absurd to expect the level of attention from the backup driver that would have been necessary to prevent this.


#4

slightly less absurd than expecting it from a machine, though.


#5

I think the human will be unfairly blamed. This is the scene as a human would have seen it. Eyes on the road might have made a difference. But it also suggests that the Uber autonomous vehicle was perfectly useless: it was looking right at the victim for several seconds before the impact on a simple, empty highway.


#6

A decade ago Mercedes Benz was putting forward facing near infrared cameras in high end cars to improve the ability to spot people and other warm-blooded critters near the driving path.

Are UBER driverless cars not using IR to help spot people?

Sample aftermarket IR cam footage posted in 2009. https://youtu.be/9y3kSvrXYmA?t=1m50s


#7

I mean, I am not necessarily looking forward to our autonomous machine overloads, but no way a human could have reacted to how fast she popped up in the dark. Maybe having the hi beams on would have helped, but I am assuming there is some lidar going on with that car. Also, has anybody done a toxicology test on the victim? Not to victim blame, but she seems oblivious a vehicle is coming down the road. And not in a crosswalk, with dark clothes. Slightly Darwin Awardy.


#8

If uber’s robot driver, with the advantages of computer-fast reflexes, is too stupid to avoid accidents that a human driver would be too slow to avoid, then it must be a pretty bad driver.

I thought the whole point of robot drivers was that they would be better and safer than slow, distractible, falliable human drivers. Unless the entire auronomous vehicle craze sweeping thr tech indistry is just a bunch of overhyped hot air.


#9

You got it


#10

You misunderstand. There are exactly zero companies that are making Level 4/5 autonomous vehicles, anything that doesn’t hit that last level will inherently have giant blind spots and short comings.


#11

That’s fair.


#12

I wouldn’t assume what you see from that footage is what a human would see. Our eyes adjust far better to the dark than most cameras do, especially video cameras.

Of course, again this is largely irrelevant given that the vehicle is equipped with equipment that isn’t reliant on the visual spectrum.


#13

Yes. Relevant questions here include: would an average human driver in this situation without an autonomous vehicle have successfully avoided a collision? With what odds? Would they be held liable if not? My guess is “no, there would likely still have been a collision, and the driver wouldn’t be held liable.” In which case I find it hard to judge an autonomous vehicle too harshly, or to use this as evidence that they aren’t ready for road use.

After all, this doesn’t say anything about whether autonomous vehicles are currently safer than human drivers or not, until we convert it to a rate of fatalities/injuries/collisions per mile driven. I don’t have those numbers, but I’d like to see them.

That said, it is ridiculous, given what we know about how humans work, to expect a “safety driver” to actually be physically or mentally capable of sustaining attention and quickly reacting to changes without doing anything for prolonged periods. That was never more than PR or maybe a legal requirement fulfilling fiction of some kind.


#14

The first thing I thought on watching the video was “Why is this person walking across such a dark road, how would anyone have seen them?!” Thank you for posting a clear image of the actual “human eyes” view.

My remaining question to the cyclist is why are you walking the bike and not riding very quickly across the road and the additional questions around it not being a crossing walk…but any normal driver paying attention would have seen them and slowed/stopped to avoid hitting them very easily. I think it is ok to ask questions of the pedestrian here, but not ok to blame them for this at all.


#15

I think that there is an important point that many seem to miss.
If the car can’t stop for an obstacle appearing in the dark area, this means that it was driving too fast.
Driving at a speed that can prevent any kind of collision is not doable, of course, but in that case the car was moving from light to darkness and that impacts also camera vision, not only human.
Of course, speeding happens also with human drivers, probably more frequently than it happens with self-driving cars, but my reaction as a human would have been to slow down if I couldn’t see at a safe distance.
This did not happen and is probably something that must be improved.


#16

“assuming the video’s shadows accurately reflect the level of darkness a human would perceive.”

Exactly - the apologists so far have mentioned shadows - which shouldn’t matter to a system sending out an active signal - and that there wasn’t enough distance to stop. Which doesn’t answer whether there was enough space to avoid the pedestrian - which there very clearly was.


#17

The only way to save this project is to rename it Blockchain Driving Car.


#18

They actually seem to have regressed.

image


#19

Looks pretty cut and dry to me. The avoidance systems on the car failed to do their job. It failed in much the same way our meat based avoidance systems fail - stupidly and with terrible consequence.
Questions of whether or not a human driver would have fared better or worse kind of fail to take in to account the countless data points we have for human failure. Maybe a human would have done better and maybe not. Either answer is just as valid.
What is more important is that the system in place on this Uber is clearly not any better than a human driver and better is the bar we seem to have set. It seems to me that what we want are autonomous vehicles that can see in the dark and make good choices. That is not what happened here.


#20

In which case, what the fuck is Uber thinking putting their alpha-level prototypes on the road?

So, we should just let the tech industry test their not-ready prototypes out in the real world? And when a drug company comes up with a new experimental drug, we should just let them put it on the shelves in a few pharmacies for testing purposes? Human lives and human well being is of less value than allowing giant corporations to do whatever they deem expedient in testing their new experimental products?