Slashdot Mirror
Police Release First Video From Inside the Uber Self-Driving Car That Killed a Pedestrian (recode.net)
An anonymous reader quotes a report from Recode: Three days after an Uber self-driving vehicle fatally crashed into a pedestrian in Tempe, Ariz., police have released video footage of what the vehicle saw with its cameras moments before running the woman over, and what happened inside the vehicle, where an operator was at the wheel. The video footage does not conclusively show who is at fault. However, it seems to confirm initial reports from the Tempe police that Herzberg appeared suddenly. It also showed the vehicle operator behind the wheel intermittently looking down while the car was driving itself.
I don't understand people like this. It's dark, you aren't lit, you're crossing a road with a large, fast moving, well lit hard to miss cars. And you can't be bothered to look for oncoming traffic. Only saw the video twice on the news but it looks like she never knew the car was there.
That said, I'm glad I was correct in my knowledge of what those "safety drivers" actually do all day.
This is a good example of why visual sensors are insufficient for autonomous driving.
Yes, this was hard to see using passive techniques with visible light (ie, your eyes), but WTF, the person wasn't sprinting or jumping off the curb, something active like LIDAR should have had no troubles spotting this.
I've seen this happen on huge college campuses as well. Legions of kids crossing streets while paying zero attention to the potential for oncoming traffic. Usually it's because their face is buried in their phone, but sometimes it's not, and they literally step right off the curb into traffic for seemingly no reason. It might make me sound like an old guy but my generation had a healthy fear of death by car instilled into it (by our parents and guardians) which seems to be sadly lacking these days. It's amazing that more people aren't routinely run down.
No human driver could have seen that woman in time to stop, but a car equipped with infrared lidar should be able to. Time to update the sensors on the test fleet.
I had heard reports that the video showed her popping out of no where. Absolutely that is not what it shows. It shows here suddenly coming into the headlights lit region, not appearing from behind a bush.
What's the cardinal rule of driving at night or snow storms? NEVER outdrive the range of your headlights. That is, your stopping distance abolutely positively has to be within your range of sight. Anything else is completely irresponsible.
So that's clearly what happened here. The woman in the video just appears like magic in a couple of frames from dark to in the head light. That means the leading edge of the headlight zone was something less than 1/2 of a second from impact. No way can you stop in that time.
This is defacto outdriving your headlights. Uber is guilty. Case closed.
Now moving on to technical details this also shows. I think part of this is that the dymanic range of the camera sucks. I am fairly sure my own eyes would have been able to see further into the dark. Those black pixels ar not just dark they are completely saturated on the dark end. Nothing is resolvable in them which is why the appearance time is so short. This is a serious problem for all systems as the dynamic range of most cameras is very limited, especially when were dealing with 1/R^4 light fall off ( 1/^R^2 light outbound and then 1/R^2 reflected. Thus a 256 bit sensor is effectively a 16 bit dynamic range sensor. And if you were to account for glints and such then it's even less. No wonder she pops out.
Secondly, where the hell was the lidar here? Shouldn't that have spotter her?
Uber is flagrantly at fault.
Some drink at the fountain of knowledge. Others just gargle.
Exactly what I expected to see....
Someone walking a bike.
At night.
No streetlights.
No backlighting at all.
Wearing black top and dark pants.
With no lights at all on the bike.
No lights on the person.
Not in a crosswalk.
Apparently not looking.
About 2 seconds of visibility.
The pedestrian is almost 100% wrong in every possible way. I don't see how this could be ANY human driver's fault, had a human been driving. As for autonomous, I guess it depends on what sensors. Could their system have had an infrared camera or other sensor that could have seen the wreckless pedestrian sooner than was evident in [human] visible light? That would have been nice. But does that make the pedestrian less at fault? I think not.
... however the test driver did not really pay attention.
Being test driver is obviously a fucked up job. 99% is killing time and 1% is killing time.
In Germany there is not one test driver but 3 ... one who would react if something goes wrong and 2 to write protocols about notable stuff.
In this case it is notable that the lights are configured incorrect. They barely shine 15 yards ahead, that is definitely wrong, and a driver or the automatic driving system should adjust speed to about 1/3rd of what it was driving.
Cost free eBook I read (by iBook/Kobo/Amazon/ObookO/Gutenberg etc.): "The Green Odyssey" by Philip Jose Farmer.
True she comes out of nowhere on the video, but that's a really crappy video. She was walking slowing and already in the car's lane when the headlights hit her, even if she had been stationary the result would have been the same.
Of course a human driver could have hit her as well, but I suspect that most often a human driver would have seen her far enough ahead to stop or at least swerve enough to avoid her (of course most Ubers might have as well).
I'm curious if that's the only video available since decent cameras are not that expensive, and I'd expect the car to have several cameras at different contrast levels.
I stole this Sig
It is astounding that LIDAR failed to see that person. I can't even imagine how they went undetected. I really want answers as to why.
BTW, the concept of a "safety driver" on a Level 3+ autonomy system is just window dressing. Distraction is bad enough on Level 2 systems that mandate hands on the wheel and sometimes involve attention monitors. With a level 3+ system, where the person isn't driving at all, distraction is essentially guaranteed (this has been studied; it doesn't matter who you are, you will get distracted sitting behind the wheel for long periods without actually doing anything). A person simply cannot transition reliably from "not at all driving for hours on end" to "emergency driving" in a split second.
Is your job to sit under bridges and jump out at unsuspecting travellers?
"Operator inattention" is absolutely the ultimate goal of automated cars.
If the human has to pay attention, then the human might as well drive, and the automation is pointless.
I don't care if it's 90,000 hectares. That lake was not my doing.
No freaking kidding. It doesn't matter that it was dark - LIDAR should have seen her from hundreds of meters away, watching her slowly step out into the road and watching her steady march across the road. Instead, it maintains speed without any braking whatsoever, straight into her. I mean, what the heck,Uber?
If you have kids anywhere near where Uber is doing automated driving tests, keep them inside. Seriously... This is just ridiculous. If it can't detect an adult slowly walking across the road, holding a bicycle, even when it's about to plow into her, what is it going to do when a little kid suddenly darts out in front?
Are they even using the LIDAR, or is it just a decoration to make passengers feel better?
Is your job to sit under bridges and jump out at unsuspecting travellers?
Sometimes people do things for which technology... has no way to compensate
Especially when that technology isn't fucking ready, yet.
Suppose that this was not a self-driving car. You see a video of a driver spending 50% of their time looking down at a (phone, book, video game, etc.) and 50% looking ahead. They look ahead, and suddenly get an OH SH*T look and plow someone down. What would the law say?
1) The pedestrian was negligent.
2) The driver was negligent.
This is contributory negligence, and I don't think the driver would get off with no penalty just because the pedestrian was negligent. This cannot be allowed to continue.
So back to the self-driving part: either the driver thought "Oh, it's a self driving car, I'll play a video game" or Uber said "Monitor this status console here on your lap and just look up every now and then to make sure that you don't plow over someone." The police need to figure that out. If it is the former, the law should do whatever they normally do in cases of contributory negligence. But if it is the latter, then Uber needs to lose their license for testing these cars, and face a big fine.
If you framegrab the images and then histogram the light curve it's hard edged at zero. Someone deliberately made the blacks blacker so it seems like no one could have seen her. Perhaps this is an artifact of the video compression algorithm or the camera itself.
Some drink at the fountain of knowledge. Others just gargle.
The radar and lidar certainly SHOULD have seen the pedestrian and it certainly appears that the driver was NOT paying attention. I also will say that in the video I could see the pedestrian while still a way out where the car should have started braking and it could have avoided killing her. While not nearly as noticeable as they would have been had the bicycle had reflectors on the wheels I could still see it when pausing the video.
The pedestrian should have had reflectors on the bicycle wheels. Just the other night I barely saw a bicyclist crossing the street in front of me at a crosswalk until they were in my lights due to the lack of any reflectors and dark clothing. I don't know what the laws are in Arizona, but where I live bicycles are required to have reflectors, a headlight and a taillight at night.
If the driver were paying a lot more attention to the road than the phone then this also could have been prevented.
This post is encrypted twice with ROT-13. Documenting or attempting to crack this encryption is illegal.
No human driver could have seen that woman in time to stop
Had the headlamps been aimed properly, they could have. In the video, when the car is traveling at 38 mph (56 feet/second), it takes about 1.5 seconds between the time the pedestrian came into view and when the collision occurred. That means that the headlamps are only lighting up an area 84 feet in front of the vehicle. If the vehicle's headlamps are about 2 feet off the ground, then when they're properly aiimed, they should be lighting up an area about 285 feet in front of the car (VOL headlamps where the left half of the horizontal beam cutoff is 2.1 inches below headlamp height at a distance of 25 feet from the front of the vehicle).
If the pedestrian was visible at 285 feet, it would have taken 5 seconds from the time the pedestrian came into view till when a collision could occur. That would have given the driver a second to react and 4 more seconds to slow down and/or change direction to avoid a collision.
1) It's a shitty video, human eyes may have seen a better picture
2) A second and a half of breaking from 38 mph may not have been enough to stop the vehicle, but it would have been enough time to slow down and swerve. The pedestrian may have been hit regardless, but they also may have survived.
On the contrary, almost any human driver would have spotted that women from far away. Don't be fooled by the artificially darkened video, as others have noted in reality the lighting conditions on that road are pretty good, and she was also not jumping on the road but crossing it slowly and under perfect weather conditions. As some autonomous driving expert on HN as commented, the sensors should have had no problem picking her up from far apart and this looks a lot like a problem with the LIDAR software.
This accident was fully preventable.