Hacker News Reader

> However, the driving footage is from after release of the story. The moon on the 21st had illumination of ~17%. The night of the accident was the day after the new moon with an illumination of 1%.

Most of the illumination from the driving footage clearly comes from the street lights, not from the moon. In the Uber dashcam footage, the street lights barely illuminate their own foot, some seem to not even reach the ground. The phase of the moon is essentially a non-factor.

Not to mention LIDAR is not affected by external light source. So the phase of the moon is doubly irrelevant.

Most, but not all of the illumination. Especially between street lights. That first video is from a section of the road where the accident did not happen. You have to go past the stop light where the first video ends to get to the section where the accident happened (second video).

I understand LIDAR should have handled it, and visible spectrum may not be much different with and without moon lighting. But I would like to see a similar conditions comparison to see exactly how bad the video is compared to actual lighting conditions.

One video pointed out a black splotch over the pedestrian. If you pause and the video you can easily see it. A very unusual artifact to say the least.

EDIT: see masklinn reply -- the accident occured before the stoplight in the much more lighted area, not later area where the news crew was filming and was commented on in the second video. It was near the stoplight with the parking deck in the background rather than only the long spaced overhead lights. Incredibly misleading video from the dashcam. Maybe good to know for defense of the true conditions, but it probably won't make that much difference. look at the slideshow

> That first video is from a section of the road where the accident did not happen. You have to go past the stop light where the first video ends to get to the section where the accident happened (second video).

You don't need the videos, the slideshow has a comparison of the exact spot on pictures 1 and 2 (you can see the shoulder bend on the right). Picture 3 (a few frames back) shows the victim was struck pretty much under the streetlight.

> One video pointed out a black splotch over the pedestrian. If you pause and the video you can easily see it. A very unusual artifact to say the least.

Standard artefact from the sensor bottoming out, as soon as you're below threshold there's literally no signal anymore.

Oh damn. Thank you for pointing out the slideshow. I hadn't really noticed it. It was in the well lighted area before the stoplight, not in the darker area afterward where the news crew was filming and comment was made on the second video.

Not only does it have brighter street lights, there are illuminated signs and parking deck lights in the background.

LIDAR is about the self-driving technology. The illumination of the road is about the operator being able to react quickly. Agreed about the impact of the moon relative to streetlights, but OP does have a good point simply in regards to proactively addressing any pushback.

OP's comment is pushback, the point is so weak it is barely existent.

It wasn't pushback. It was a reasonable line of argument where the second video claimed the accident occured (in the much less illuminated area after the stop light).

Wish I had read this before I made my edits to the two original posts.

Just to quantify this side of the argument, street lighting is by design 50-1000x brighter than the moon, depending on the kind of street. For example the intersection of two major roads with pedestrians present is supposed to be lighted to about 30 lux, and a full moon is .1 lux.

I'm almost certain that the footage was modified to ride the public opinion of "it's too dark to see her, so it was her fault!" even though the car should be able to operate in complete darkness thanks to LIDAR and radar sensors. Uber never plays by the rules, I wouldn't expect them to suddenly be super honest about this.

Modified? We would never! No, we simply, uh, compressed it. So that more people could view it without bringing down the server. Standard practice.

"So what does the uncompressed footage look like?"

Intern lost it.

I'd be surprised if it was actually modified seems too big of a liability.

If it was, someone should be going to jail as a result, since that's the footage they seemingly handed over to the police as evidence in the investigation.

There are some odd artifacts in the video[1]. Like that the brightness is very low and there is black blob over part of her which then disappears.

[1] https://www.youtube.com/watch?v=sKBy4_3azBg

Moon or not, the Uber video shows the headlights projecting less than half the distance they're rated at. They're functioning so badly, in fact, that it wouldn't be safe to drive with them, human or bot.

Interesting, I wonder if they where on, maybe those are just the daytime running lights.

I'm pretty sure the multiple streetlights present on that section of the street are the main thing illuminating the road; the phase of moon is an insignificant factor.

Not to mention, the big selling point with AVs is that they use LIDAR which is superior to human vision. So where's the LIDAR data? Why didn't the car see this person using its non visual sensors.

I think its a given that visual only data is pretty lousy for AV purposes. You have too many issues with it.

There are extensive street lights along that stretch of road. The moon isn't going to be a big factor.

While I completely agree with your assessment, I feel compelled to point out that LIDAR is an active sensing method and so ambient illuminance actually makes things harder for the sensors.

I really wanna make excuses for the Uber vehicle here because self driving vehicles will be awesome, but in this particular case it just failed. :(

Yeah and killed an innocent person!