Hacker News Reader

For context, the accident happened under a brightly lit portion of the street. Safety driver could have easily avoided it if she were paying attention to the road instead of her phone.

Picture of where it happened: https://imgur.com/gallery/XQrAB

You can clearly see the shadows on the left that she walked out of are not lit.

Also, the ISO is BLOWN OUT on your image to create the false perception of a well lit scene, you can tell by the significant pixel noise that incredible amounts of digital amplification have been applied to this image.

This is a better approximation of human sight: https://i.imgur.com/djOuoZr.jpg

The image you linked is just so digitally distorted that it's very misleading. Heck, your linked image is so distorted that the sky is literally grey!

I am pretty sure that this accident happened because of the way autonomous software makes decisions on its surroundings. I am almost certain the car knew there was an object but evaluated it incorrectly.

There are lots and lots of false positives a self driving car has to disregard just to make its way along any street. Mailboxes on curbs, random parked cars, bridges, signs, and so on. I think the car unfortunately made a bad decision.

To say the LIDAR/RADAR or IR cameras didn't pick the person up is to say that these cars are woefully under-equipped to sense their environment.

I think the Police report was just working off of current operating Police knowledge and expectations of the law.

If there is a bike in the road, I want my car to stop; whether a prankster kids push it out just in time or it's a woman pushing it across slowly.

I don't think I implied anything to say that a car should treat a bike or any other object moving into its path as a false positive. To the contrary, what I was saying is that it did treat it as a false positive but that was obviously not the intention of the engineers.

My overall point is simple, the car almost certainly saw the person - it just had to have, but the way it evaluated what to do about it was wrong because evaluating reality, as it turns out, is incredibly complex.

> The police reported that the pedestrian, wearing black, entered the roadway randomly and unexpectedly, directly in front of a moving vehicle, and concluded that a human driver would not have reacted in time either

She did not enter the roadway directly in front of the Uber. She entered the road from the left and had to cross multiple lanes to get in front of the Uber, which was in the right lane. There were no obstructions between her and the Uber during all of this, and even if the road had not been well lit by streetlights, she would have been in range of the Uber's headlights for over 4 seconds before the collision (the low beams on that model car have a range of 250 ft, and the car was traveling 55 ft/second).

> The police reported the accident as "unavoidable", and showed that there was a crosswalk very close

The nearest crosswalk was 300 ft away, which I would not count as "very close".

> The police reported that the pedestrian, wearing black, entered the roadway randomly and unexpectedly, directly in front of a moving vehicle, and concluded that a human driver would not have reacted in time either.

That police report is at best sloppy and unwilling to do anything other than take limited evidence at face value and at worst an active attempt at spreading misinformation.

The pedestrian was crossing from the median across two lanes of traffic, and was struck on the right side of the right-hand lane of traffic (at least 75% of the way through crossing the road). Even the video in question (which, as many have noted, is far worse quality than anything a driver would be seeing) shows the pedestrian a good second before impact, which is enough time to at least initiate emergency braking even for a human--and the car shows no sign of having done this.

Despite the Uber-released video being misleadingly darkened, you can clearly see the victim was already in the road when she was hit, she did not "leap in front" of the car. The car made no attempt to avoid a collision, she was simply mowed down.

Yes. And as a pedestrian, I would expect cars to slow down somewhat in advance when I would cross the street like that. That didn't even happen. Even if the car would brake in time, it would still be scary as hell compared to the same situation with real human drivers.

Sorry, but I think this not reflective of what actually happened, you are subtly implying this was a suicide when we have zero evidence of it being anything other than a horrific accident.

The police are biased because of the political situation in the state where the governor persistently courted Uber with relaxed rules.

> Fascinating how narratives and bias work.

> the pedestrian, wearing black, entered the roadway randomly and unexpectedly

People do random and unexpected things all the time, often while wearing dark clothes.

Claiming someone has a bias doesn't reflect well on your argument.

Out of curiosity, what's with the emphasis on the pedestrian wearing black? Was it a night accident or something?

Yes. The video from the internal camera shows a pitch black night with almost no light and a black clad figure appearing out of nowhere. However this footage is quite disputed as this level of darkness would suggest that the car had basically no lights to speak of and that the roadside lightning did not work either, none of which were of course true.

They had Blue Jeans on and where moving a pink bike that had reflective surfaces. The dash cam was very deceptive, a normal driver would have easily seen this a long way off if the driver had been paying attention. On top of that Uber disabled the cars built in safety system that likely would have also worked.

> top of that Uber disabled the cars built in safety system that likely would have also worked.

What safety system? I haven't read anything saying this, so any source would be appreciated!

https://www.bloomberg.com/news/articles/2018-03-26/uber-disa...

Re: sibling comment - they moved to a single lidar on the roof with their new vehicles. It has a ~3 meter blindspot around the car

At 45MPH you can't stop within 3 meter's. The car needed to break 20+ meters away to avoid hitting them.

A 3 meter blind spot around the car doesn't necessitate it can't see beyond 3 meters. Just as you have blind spots in your vision around your body (like you can't see your legs when looking directly forward) but you can see long distances.

The LIDAR is on the roof. I can see the roof blocking some things close to the car depending on how tall they are.

However, my point is what happens within 3 meters is irrelevant, the car needs to look ahead and brake much sooner than the last 15 feet to avoid hitting stuff.

I think we both agree then, I just mis-interpreted the emphasis of your point.

The dash cam doesn't show a figure appearing out of nowhere. It shows a person crossing the street in a slow and predictable manner as if she has no concern that the car speeding toward her will fail to see her and slow down. Her white shoes and blue jeans stand out in the video.

All that's negated by the video showing her looking down at her cell phone.

You guys do realize this is how it works right? If you have any alcohol in your bloodstream, any accident you get into is automatically your fault and you were driving under the influence. The circumstances of the accident simply don't matter anymore, you're the one thrown under the bus for the whole thing.