It certainly looks bad for Uber

http://ideas.4brad.com/it-certainly-looks-bad-uber

by sohkamyung • 8 years ago

86 68 8 years ago

ideas.4brad.com

bwang29• 8 years ago

Good summary of technical evaluation from equipments. I think the other thread https://news.ycombinator.com/item?id=16643056 delved pretty deep in various subject scenario too. It would be the best for Uber to release the raw sensor data during the event, and it might be worthwhile for the community to see exactly what happened.

dnomad• 8 years ago

This blog post, like much of the HN discussion, is full of baseless speculation. There's really nothing here except "LIDAR should've seen it." Duh.

The idea that Uber should release more data which won't be analyzed Donnelly but will only fuel even more speculation doesn't make much sense.

The NTSB is investigating. They are extremely through. Most importantly once all the data has been carefully analyzed and the failure mode had been understood the NTSB will be able to devise guidelines and protocols to prevent it from happening again.

Until then I don't see how all this speculation helps except to fulfill some sort of misguided anti-Uber fantasy.

buvanshak5• 8 years ago

is full of baseless speculation...

Can you list all the "baseless speculations" in the discussions/blog post?

Twisell• 8 years ago

If LIDAR can’t do, car shouldn’t speed.

Either way is a fail.

CobrastanJorji• 8 years ago

It would be the best for Uber to release the raw sensor data during the event.

I agree with that statement only if we define "best" as "for the common good." The best thing in Uber's own interest is very likely to lock that shit down and hold onto it until a judge makes them give it up. Somebody died here, and it very well may have been Uber's fault. Uber could very well be on the hook for a tremendous amount of punitive damage.

px1999• 8 years ago

They'll probably fight that as much as they can, because it's not in _their_ best interests. Raw data will likely make what happened appear easy to stop, particularly if they were capturing LIDAR at the time and (assuming no negligence), at best will make Uber look incompetent.

People won't accept the nature of how image recognition tends to fail - ie seemingly randomly and catastrophically. Assuming a driver-less future, we'll definitely see more of these random tragedies.

foobar1962• 8 years ago

Based on the video shown there is no way a human driver could have avoided that collision.

umanwizard2• 8 years ago

Such a human shouldn’t be fucking driving.

If I’m driving fast enough that I can’t see pedestrians with enough time to react, I slow down.

Yes, this means in near-total darkness, slowing to a crawl if you can’t turn on your high beams.

JustSomeNobody3• 8 years ago

This is exactly what so many people are missing about this story. Posted speed limits are the fastest you can go, not the recommended speed.

ocdtrekkie2• 8 years ago

Someone pointed out elsewhere that human eyes are better in low light conditions than cameras. It's a lot less likely a human couldn't see her than that camera, but the human wasn't looking at the time.

buvanshak5• 8 years ago

Human eyes have much better dynamic range than most of the cameras out there. Yes even in differently illuminated areas. If the vehicle was on High beam, (which you would be on if you have trouble seeing stuff in distance), then A human should have spotted her from a very good distance..

jon_richards2• 8 years ago

The video is extremely misleading because the camera is overcompensating for the bright lights illuminating the road. A human driver could have seen the person and/or bike from a long way off.

Our eyes are extremely good at adapting to high, non-linear contrast. That's why "pictures" of total solar eclipses have to be composite images to really capture what people see when they look at one.

Pilfer3• 8 years ago

A human driver could have seen the person and/or bike from a long way off.

Not true at all, the person was not standing under lit streetlights and was in the shadows. Humans have problems detecting objects in the shadows, especially at night.

3M made a video illustrating how difficult it is to see people at night. In this video, people wearing bright-colored clothing don't even show up until 250 feet away. According to the national safety council, a human driver traveling at 30mph at night, may take up 500 feet to react to and stop when there is an object in the road.

I reccomend watching this video as it strongly changed my views about night driving.

https://www.youtube.com/watch?v=XMvM7-9lgeg

jbkrule• 8 years ago

Did you read the writeup? He directly addresses the points of when a human would be able to detect the person as well as the fact that the Uber was driving at the upper speed of safety for the situation which a human driver may not have done.

viburnum2• 8 years ago

Bullshit. Do not overdrive your headlights.

newnewpdro2• 8 years ago

The video is misrepresenting what's visible vs. invisible.

There was no human driver in that car. There was a person in the driver seat focused on something other than the road.

Anxuiz• 8 years ago

If only looking at the posted dash cam video, yes it does look difficult to avoid. But that video is incredibly dark and not at all what the environment looks like to the naked eye.

Here's a video posted by a local on Reddit showing better how the street looks at night: https://youtu.be/1XOVxSCG8u0?t=31

jankotek• 8 years ago

Could someone post stats: mileage covered by self driving cars, mileage in bad weather, accidents per mile....

Also is it true it was going over speed limit? How is that even possible for self driving car?

ocdtrekkie2• 8 years ago

Self-driving cars are meant to coexist with humans so they tend to be programmed to drive like humans (slightly above the speed limit).

viburnum2• 8 years ago

There are 12.5 deaths per billion vehicle miles driven. This is the statistic that self-driving cars have to beat in order to be "safer."

Based on this, Uber has had about 3 million test miles (under mostly safer than average road conditions, but I'll leave that aside). They're now at 333 deaths per billion miles driven. Gonna have to go a few more days without a death before they're "safer." (This is not all robot cars, of course, but not all robot cars are meaningfully the same right now).

http://www.eschatonblog.com/2018/03/the-faith-is-strong.html

extralego2• 8 years ago

One lesson from this accident might well be to map "illegal, but likely crossings"

Good write-up but I don’t understand this part. I would not be comfortable with these cars making moves based on calculations of the legality of observed behavior.

EDIT: Nevermind! I thought about it some more and realized they mean mapping the geographical locations where pedestrians are likely to be crossing the road. Much more humane than I previously thought! But I remain uneasy about any system with room to benefit from selective inquiry like such.

rcthompson2• 8 years ago

I didn't read that as a suggestion to take the legality of a crossing into account. In fact, the suggestion is to do the opposite: instead of only mapping legal crossings, map all likely crossings without regard to their legality.

TomMarius4• 8 years ago

It doesn't mean that it will take legality into account, it means that they should have a database of illegal crossings as well (they already have one of legal ones - a map).

garyfirestorm• 8 years ago

How do you propose on building a database like that? Any patch of road which is not a legal crossing is inherently a potential* illegal crossing. I think illegal crossings are non deterministic and hence unpredictable.

Edits

Reply

dtech• 8 years ago

The meat of the article is about how LIDAR, Radar, or any number of systems could have detected the pedesetrian/obstacle on the road and should have stopped the car.

If those worked, considering likely crossing locations in addition to certain crossing locations like zebra crossings should only improve safety, not be required for it.

Reply

extralego2• 8 years ago

Makes total sense! I would just be tempted to apply the extra safety measures which would be applied to the crossing zones, to every zone.

Reply

TeMPOraL• 8 years ago

Still, IMO it's a workaround, not a solution.

Reply

thisisit• 8 years ago

This can get interesting in Asian countries. As pointed out by wiradikusuma:

https://news.ycombinator.com/item?id=16646415

people tend to cross ignoring the traffic laws.

Reply

developer2• 8 years ago

The technology should be good enough that every single meter of road is considered a "likely crossing". There shouldn't be an iota of difference between a supposed "likely crossing", versus any other section of road. If the sensors and decision engines aren't good enough to be driving under normal conditions, then perhaps they should not be on the road (yet).

I think it's absurd these vehicles are on the road today. The process is being rushed for the financial gain of corporations. No matter how far up the list safety of these cars is, it's not priority #1. The greed of companies vying to be the first to stake a majority share of this new market is the only real priority.

Reply

bradtemp• 8 years ago

Neither humans nor robots can treat everywhere like a crosswalk. We human drivers see pedestrians standing on the edge of the road all the time, often getting ready to cross. If we slowed down every time we saw that, roads would flow poorly. On the other hand, we do slow down when we see people poised to cross at an actual crosswalk, and the law demands we do that.

Right now, live humans are better at reading the body language of pedestrians than machines are, and this is an area of research.

My suggestion is that mapping places where people are likely to cross allows you to dial up your caution in those locations. Not to dial it all the way up to what the law demands in legal crosswalks, but not to have it at the most basic level. Of course, once a ped enters the roadway, you will dial up the caution.

On 45mph roads, peds know to expect cars not to keep them safe. You wait for a long clearing in the traffic and you run if you have to. On low speed roads, we cross even with cars coming, expecting them to see us and react.

Reply

wiradikusuma• 8 years ago

From the video, it seems that the woman didn't even look at the street (the direction where the car came). Actually, she didn't even seem to look at the street before she crossed.

I do believe that autonomous vehicle should have prevented this kind of accidents (since they are 100% aware, unlike humans), but why did she act like that? Is it common there to just cross the road and expect cars to stop?

The reason I ask is culture difference. From my experience, in some countries (e.g. Indonesia), people cross the road like they have superpowers. They use their hand to signal the car to stop while taking time to cross. In Malaysia, even if you pay attention to the road, the drivers _seem_ to want to run over you when you cross (they go faster).

Reply

rcthompson2• 8 years ago

I can't speak to Arizona in particular, but in all the parts of the USA I've been to, the common wisdom has always been to look both ways before crossing the road no matter what, because regardless of who has the right of way, there's always the possibility of some idiot, drunk driver, or other hazard that ignores right of way rules.

Reply

uluyol• 8 years ago

Common wisdom isn't necessarily followed. I've see many people cross the street blind especially on college campuses. Worse, where I currently live (Ann Arbor) I have seen people do this when they very clearly do not have the right of way.

Reply

dragonwriter• 8 years ago

From the video, it seems that the woman didn't even look at the street (the direction where the car came). Actually, she didn't even seem to look at the street before she crossed.

IME, at least in places I've driven, this is not at all uncommon for apparently-homeless people in urban and often suburban environments, at points that are not legal crosswalks and at points that are controlled crosswalks but against the controls (also, of course, at proper crosswalks without or consistent with controls, though that's less remarkable.)

Reply

newnewpdro2• 8 years ago

From the video, it seems that the woman didn't even look at the street (the direction where the car came). Actually, she didn't even seem to look at the street before she crossed.

You saw the pedestrian _enter_ the street in that video? I saw nothing of the sort when I watched it.

Reply

Pilfer3• 8 years ago

I think the point the poster is trying to make is that if the pedestrian was looking while crossing the street, they would have ample time to react to the oncoming car.

Car headlights can be seen from over 1000 feet away, and at the speed the Uber car was traveling, the pedestrian had 20 seconds to see & and react to the vehicle coming at her.

From the video, it's clear they were not looking for oncoming vehicles while crossing the road.

Reply

nikkwong• 8 years ago

Tangential but I find it interesting that countries so geographically similar would have such a different custom in handling that situation.

Reply

hiyer• 8 years ago

In India itself, Hyderabad is like @dragonwriter described Indonesia - the traffic keeps flowing and pedestrians walk across anywhere - the vehicles will just weave around you. Bangalore is like Malaysia is described - if a driver sees you attempt to cross, he or she will actively speed up.

Reply

whyever• 8 years ago

Funny, I have seen what you describe about Indonesia in Malaysia.

Reply

ojosilva2• 8 years ago

The good news about self-driving car accidents is that improvements to the system will be eventually shared by all cars saving many lives.

We can finally say this poor woman did not die in vain, which is a step in the right direction which may end up making cars safer than airplanes someday.

The only serious mistake (misunderstanding) to come out of this is to believe the human behind the wheel is at fault. No human being can keep full attention on a road when a computer is driving the car for you.

Reply

buvanshak5• 8 years ago

The good news about self-driving car accidents is that improvements to the system will be eventually shared by all cars saving many lives.

If this was true, how come the car was not trained for this pretty common/typical scenario. Did it take one human life to show that it is possible that a pedestrian can cross the road in front of the car?

Are the people gone so blind with this obsession with new tech?

Reply

ojosilva2• 8 years ago

The system is very early stage, I believe there's still a long way to go until the system is fully trained and has enough redundancy as to remediate for bugs in one component. I'm pretty sure the engineers foresaw that situation to an extent and that their design intention is to prevent or avoid running over pedestrians.

The same happened in aviation throughout the years, until airliners got to a point where systems just don't fail... Or at least not without other 2 systems correcting it.

There's human intervention behind every aspect of this accident, and again, like airliners, there will be many accidents (and hopefully not many deaths) until engineers can get it right. FAA and other agencies played a crucial role in exposing and recommending improvements across different manufacturers and airlines, hopefully government will assume that role here too to ensure improvements are mandatory before sending prototypes into the road.

Reply

bigiain• 8 years ago

Sadly, I suspect all three of those sentences have a fairly high chance of being incorrect.

If Uber think they can gain some benefit from keeping part or all of what they learn here to themselves, I strongly suspect all that data will be wrapped up forever in trade secret claims...

I certainly wouldn't want to be the one going to tell this woman's family that "she didn't die in vain, thanks to her and Uber, self driving cars will now be safer for everyone!" If they didn't punch me in the face and say "Fuck you! She died in vain, while Uber made the most profitable calculations and moves as they risked and callously ended human lives to increase their shareholder value!" then they've better people than me.

The human behind the wheel was not sitting in a known-safe autonomous vehicle with a record of better-than-human road safety, they were at work doing the critical job of overseeing a prototype self driving car with a barely alpha-release software version controlling a machine which is well known as capable of killing people. It was their only job - to keep their full attention on the road.

Reply

zaarn• 8 years ago

I do not believe the human behind the wheel was at direct fault. I would find them guilty of (gross) negligence since they were supposed to pay attention.

The problem here is that the car had superior systems to the human eye and no large lag between perception and action. It was fully capable of seeing the pedestrian and yet it didn't even attempt to break.

Reply

dirtbox2• 8 years ago

I moved to the US a few years ago and have already lost count of how many cyclists wearing dark clothing and no lights on poorly lit roads I've nearly hit at night.

I had to take a cycling proficiency test in the UK, here no one seems to have a clue, they don't even always ride on the right side of the road.

This woman neglected any due care and her death, while tragic, is entirely her own fault.

Reply

rootlocus• 8 years ago

Quote from the article:

To be clear, while the car had the right-of-way and the victim was clearly unwise to cross there, especially without checking regularly in the direction of traffic, this is a situation where any properly operating robocar should have avoided the accident anyway.

I'm not assuming you haven't read it, but I think this is the best answer I can give you.

Reply

bryanlarsen• 8 years ago

"entirely her own fault"

Why does the entire Internet feel the need to apportion blame in this case?

There are four entities who could have and should have relatively straightforwardly avoided this death.

1. The woman shouldn't have crossed the street there and then.

2. The safety driver shouldn't have been looking at her phone.

3. Uber's automation should have caused the vehicle to brake much sooner.

4. That street should have been designed much safer. The design of a lit crosswalk on the median encourages people to cross there, so much stronger discouragement is required. Furthermore, a 35mph limit in an area with pedestrians is going to regularly cause pedestrian fatalities. That's a trade-off most people seem willing to make, but if you make that trade-off you have to own it. If the speed limit was 20mph that woman would be alive today.

As far as I can see it, all 4 entities are 100% responsible for the death of the pedestrian.

None of those 4 entities passed the "reasonable person" test with their actions, therefore all 4 are fully responsible.

Sure you can argue all you want on whether one entity's misbehaviour is more egregious than the others. It doesn't matter; all 4 engaged in behaviour that regularly kills people at a rate much higher than acceptable.

Reply

dirtbox2• 8 years ago

You can continue to cast the blame for frankly abysmal state of cyclist safety in this country, meanwhile cyclists die with alarming regularity however many of those factors are in play.

Reply

imh• 8 years ago

The police may not have a good way to evaluate the vastly superior dynamic range of human vision compared to the camera.

The solution is simple and even pointed out later in the article:

Note that the streetlamps are actually not that far from her crossing point, so I think she should have been reasonably illuminated even for non-HDR cameras or the human eye, but I would need to go to the site to make a full determination of that.

Seems like the question of whether a driver would have seen the pedestrian is simply solved by going to the site another comparable night and seeing how visibility is to a human pair of eyes.

Reply

Symbiote• 8 years ago

This is what the people who investigate railway accidents in the UK do.

They would go to exactly the site, with an identical car, someone wearing the woman's clothing (or identical clothing) and an identical bicycle, and reenact the scene, measuring light levels and so on as the car approaches.

For example, page 23-24 of this report positions a tram at various distances to check the visibility of the headlights: https://assets.publishing.service.gov.uk/media/547c8fbfed915...

This is much, much more than is done for a road accident, but would seem entirely proportionate for a robot car accident investigation at this stage.

Reply

yread• 8 years ago

It's fascinating how brigaded the reddit videos thread is

https://www.reddit.com/r/videos/comments/86756p/police_relea...

Basically all top root comments are saying that it's the pedestrian's fault! Some math genius even proved how impossible it is to stop.

Especially compared to quite decent ARS article https://arstechnica.com/tech-policy/2018/03/video-uber-drive...

Reply

lmilcin• 8 years ago

To all that say it's not the fault of the driver (be it human or computer) please, remember, that for every collision there is many, many near misses that were avoided because the driver was paying attention and took necessary actions to avoid collision regardless of whether he had or had not the right of way.

The right of way is not a permission to plow into other road users.

What if your kid jumps suddenly on the street, would you be ok if the driver excercised his right of way?

If AIs are allowed to drive they are expected to perform to avoid collisions (and especially with slow moving objects like pedestrians). This was supposed to include devices like lidars which were supposed to perform better than human driver could hope to do.

The added information and reaction time was supposed to offset relative dumbness of the driver.

Now we are learning that the car is only operating on what seems a visual spectrum camera and doesn't seem to be reacting in a simplest situation and the driver isn't even looking at the road.

So the car was supposed to start breaking IMMEDIATELY the threat is recognized. I don't see hard braking on the video. I see a human driver taking her eyes from what seems to be a phone to recognize the situation and NO DECELERATION THAT WOULD HAVE TO BE IMMEDIATELY EVIDENT.

Shame on you, Uber.

Reply

hardly_a_mirage• 8 years ago

Sorry for the off-topic question. I see a lot of people spelling 'breaking' when it should be 'braking'. Is breaking a good replacement for braking? I am under the impression that it is not. However, english is not my first language.

Reply

vevoo• 8 years ago

Waymo's cars, and a few others use long-range LIDARs able to see 200m or more...There is a dark irony that this longer range LIDAR is what the Waymo vs. Uber lawsuit was about<

Prior to the accident, I gave high probabilities that Uber performed a surgery on their LIDAR systems after the lawsuit, aiming to eliminate anything with legal implications. If the above comment is true, and uber is not using any long-range LIDAR, please someone with better knowledge help to stop Uber testing of an unsafe system.

Reply

It certainly looks bad for Uber

Related Stories