Hacker News Reader

Comment by mirimir | Hacker News Reader

Right, so now she's crossing three lanes. So that's about 9 meters. At 1-3 meters per second for her, and 17 meters/second for the Uber SUV, it should have seen her 3-9 seconds (50-150 meters) before impact.

That's a pretty clear fail.

I wonder what the Uber's field of view is. At what distance should it have detected stuff about 9 meters off track?

But she did likely walk out from landscaping. So perhaps she initially "looked" like waving branches or whatever. And there must be some mechanism to suppress such false positives, or vehicles would brake when it's windy.

View on HN

> That's a pretty clear fail.

And yet the police were so quick to say that camera footage indicates that Uber was likely not at fault. It was so quick that it makes me inclined to think it's pretty obvious. However, after the SF Chronicle published its exclusive interview with the Tempe's chief of police, the Tempe Police PR person had to issue a statement that, "Tempe Police Department does not determine fault in vehicular collisions."

I'm not one to believe in conspiracies or to automatically suspect shadowy influence, so I want to believe that the the camera footage seems to argue that this was an unavoidable accident. But the evidence released so far argues against that, regardless of whether the victim's crossing was illegal or not. And why are the police making a judgment on this so soon in the first place, especially when the decision will be made by county law officials, and ostensibly after referring to Uber's full suite of sensor data?

https://www.engadget.com/2018/03/20/uber-fault-pedestrian-fa...

It's probably just PR.

> And yet the police were so quick to say that camera footage indicates that Uber was likely not at fault.

I was surprised at that -- it doesn't seem to me that the police would have the technical knowledge of Uber's hardware and algorithms to categorically state that, let alone so quickly.

I fall more into the "AZ wants the SDC companies in state and having a death due to a SDC would hamper that with the public" camp. Better to find the SDC not at fault and save face.

Well, if you saw that video and it really did show a < 1 second time between the pedestrian becoming visible and being hit, then the autonomous systems don't matter at all. That accident is physically impossible to prevent in that car.

So whether or not the police understand self-driving systems is not relevant to the fundamental physics problem. The key question is how long was the ped visible.

"The pedestrian becoming visible" would be to the video cameras. If the vehicle has other sensors (LIDAR, RADAR, etc), how sooner/later would the pedestrian be visible to these sensors? Would it be possible to judge that by looking solely at the dashcam? Would the sensors ignore the pedestrian as a false positive in these circumstances?

That's what I wonder. Yes, it has LIDAR etc. But there was this comment yesterday by arbie:[0]

> Sensor Fusion typically merges LiDAR with stereoscopic camera feeds, the latter requiring ambient light.

So might LIDAR data get de-emphasized when there's too little light for the cameras?

0) https://news.ycombinator.com/item?id=16619917

I don't understand that comment. "Sensor fusion" is the integration of data from two or more different sensors. It doesn't matter what those sensors are. They don't even have to be different types of sensors.

It may well be that the Uber cars only have LIDAR and optical, but that doesn't have anything to do with sensor fusion.

If there’s not enough light for the cameras, LIDAR would get emphasized, not de-emphasized.

Thanks. So in this case, how come the woman wasn't identified based on the LIDAR data?

No idea

I didn't mean just visible as in visible light visible - I mean detected in any way by the systems.

By the time that pedestrian has even traversed a single lane, the LIDAR on top has generated gigabytes of point cloud data showing her moving, assuming constant velocity of both car and ped, on a straightforward collision path.

I don't see how from a single forward-looking camera you want to deduce this crash was "physically impossible". That isn't even half of what we expect from human drivers. We expect them to swivel their head.

Where did I say just "camera"? By visible I mean visible in any way to the systems.

An accident may be physically impossible to prevent, but that's not the only standard. Initial reports (from the police) say that Uber AV did not even brake. Is not braking at all within a 1 second window the performance touted by AV's detection and braking system?

"ped"?

pedestrian

<1 second between becoming visible (on the left side of the vehicle) and being hit by the right side of the vehicle doesn't seem likely.

If it falls to a question of stopping distances, self driving vehicles are not really practically different from any other car, save perhaps for the ability to reduce the reaction time phase of stopping. I can certainly imagine there will be scenarios in which experienced traffic cops can make a reasonable determination, especially if the speed, distances, locations etc of the car and pedestrian are known.

I mean... the police just make shit up all the time. And Arizona was very happy to attract Uber with their regulation-free state law.

The fact that they came out with a statement instead of saying "we don't comment on ongoing investigations" makes it even more shady.

The first bombing victim in Austin was one of the first suspects of the police in his own bombing.

http://foxsanantonio.com/news/local/first-bombing-victims-fa...

https://www.thedailybeast.com/police-treated-first-austin-bo...

The police in Georgetown about half an hour north of Austin went with his gut in one murder case - "it has to be someone the victim knew" and ignored physical evidence at the scene and the eyewitness testimony of the murder victim's son. Then the prosecutor kept working against the accused for the next 25 years and suppressed evidence. https://www.texasmonthly.com/politics/the-innocent-man-part-...

I don't agree with the sentiment that it happens all the time (but don't disagree that it's happened more than a few times), but I do think police are much less likely to make something up if they know (and have seen) the camera evidence. No point in saying something (the day after the incident) that can so easily be contradicted later.

It's all speculation until we get a release of the actual video footage from the car. This is ridiculous. The NTSB/police need to release the actual footage to dispel speculation. We have actual evidence of exactly what happened and yet no one outside of the investigators and Uber can see it.

NTSB investigation has begun and it make take a few weeks:

https://twitter.com/NTSB_Newsroom/status/976215176323194880

I do think discussion is warranted because the Tempe Police have gone out of their way to share details and preliminary judgments about the case. Some of these details have been inaccurate, both for and against Uber's favor. But Tempe police officials are already going out of their way to downplay what their chief claimed:

http://www.phoenixnewtimes.com/news/federal-agencies-investi...

> Glover downplayed Moir's statements, saying some were taken "out of context" by the Chronicle. The chief disagrees with the Chronicle's headline, "Tempe police chief says early probe shows no fault by Uber," Glover said.

My uncle bought a car with some proximity sensor, basic thing for walls during parking. He, like many, was annoyed at the regular bips etc.

I'm sure a system that warns more for the sake of cautiousness would be rejected by people.

Reply on news.ycombinator.com