Uber Autonomous Vehicle - pedestrian death

Johnboy1951

Sad to hear of such an incident.
Now there will be a huge amount of investigation etc which will surely colour whatever regulations are put in place.


https://techcrunch.com/story/uber-self-driving-car-strikes-and-kills-pedestrian-while-in-autonomous-mode/

mordeith

Johnboy1951 wrote: »

Sad to hear of such an incident.
Now there will be a huge amount of investigation etc which will surely colour whatever regulations are put in place.


https://techcrunch.com/story/uber-self-driving-car-strikes-and-kills-pedestrian-while-in-autonomous-mode/

No pedestrian has ever been killed by a human driven vehicle of course.

jimmycrackcorm

The investigation result will be very telling. There was a safety driver in place so I'd be curious why they weren't in a papering to slam on the brakes.

I don't know why there is an expectation that an autonomous car will not kill anyone.

colm_mcm

jimmycrackcorm wrote: »

I don't know why there is an expectation that an autonomous car will not kill anyone.

You’d assume they’re be designed not to.

peposhi

I'd would be very interested to know how many safe km have all autonomous cars travelled before the accident. Then compare that to an ICE...
Also the footage of the accident (if it's ever released).
Sad to see it happen though. May the person that died rest in piece.

jacknorris

Has anyone heard or read of a reason as to why the safety driver was unable to prevent the accident?

Bob24

Sad news for sure, but impossible to draw any conclusion on the possible impact on the future of self driving cars before we know more about the circumpstances.

The key question will be how much the automated system is at fault? It could range from not at fault whatsoever (the person threw themselves in front of the car from a very short distence with not enough time to beak even with immediate reaction time) to fully at fault (the system would have had plenty of time to react but failed to identify the danger).

Bob24

colm_mcm wrote: »

You’d assume they’re be designed not to.

Of course they’re designed not to, but the car doesn’t control its environment and there are times when accidents can’t be avoided by a car a lone. If let’s say someone who is suicidal is standing by a high traffic road and throws themselves in front of a car from a very short distance - there is little way to avoid impact in a safe manner, if at all.

And that is just one exemple.

Billgirlylegs

jacknorris wrote: »

Has anyone heard or read of a reason as to why the safety driver was unable to prevent the accident?

Preliminary accounts seem to point to the lady stepping out quickly, from the dark at a "non crossing".

It really is the issue with automated driving systems - can they deal with human eccentricity?
If they can't, who is to blame if one of these cars kills someone?

bren2001

Bob24 wrote: »

Of course they’re designed to, but the car doesn’t control its environment and there are times when accidents can’t be avoided by a car a lone. If let’s say someone who is suicidal is standing by a high traffic road and throws themselves in front of a car from a very short distance - there is little way to avoid impact in a safe manner, it at all.

And that is just one exemple.

Well that's the point, self driving cars have the potential to reduce the number of deaths on our roads. They will never eliminate them. They also have the advantage that every time it happens, the code can be updated to ensure it won't again (granted, very difficult in this situation).

Dardania

jacknorris wrote: »

Has anyone heard or read of a reason as to why the safety driver was unable to prevent the accident?

One of the articles this morning mentioned that police believe the woman just walked into traffic. Possibly homeless. Car was doing 38mph in a 35mph zone. Driver first knew about it when they heard the bang.

doolox

Once you have to have somebody vigilant at all times in the car to watch out for other peoples mistakes then the whole concept of autonomous cars goes out the window.

The whole key is to get rid of the need of a human driver and its costs. People would also be attracted if riders in an autonomous car could be drunk while riding in the car, a marketing game changer which would re-open the freedoms enjoyed by previous generations before drunk driving laws made spontaneous medium distance socialising next to impossible.

Just as you do not allow people enter a lift shaft at will and at any time they feel like it, autonomous cars will need a completely separate infrastructure to pedestrians and other road users, just like railroads and elevators have.

This will be so costly that the idea of autonomous cars for short, social trips will be next to impossible to implement but might be possible for heavy, long distance travel something like roadtrains in australia are confined to rural roads only, so that costs can be spread out over a big long distance mission or task.

Confining autonomous vehicles to motorways might be a start but loses a large market in the city taxi sector.

seamus

This will put Uber out of the game.

They've been developing their autonomous software using Agile-like, "minimum viable product", "testing in production" methodologies. These work perfectly well when all you're doing is sending messages between phones. But in applications with life-or-death implications, having your software "learn on the job" isn't good enough.

Several reports months ago stated that Uber were miles behind Google and Tesla in the viability of their product, and their decision to halt all testing will only make that worse.

Eventually only a couple of companies will actually own and maintain this software and all other manufacturers will use it under licence. But in the rush to become one of those companies, others must fall by the wayside. Uber is the first, but won't be the last.

Dardania

I'm wondering why Volvo's pedestrian protection technology didn't kick in, whatever about Uber's...

Creamy Goodness

Why can't America solve this by giving everyone self-driving cars? Works for gun deaths.

doolox

That would be a non-runner.

jimmycrackcorm

seamus wrote:

They've been developing their autonomous software using Agile-like, "minimum viable product", "testing in production" methodologies. These work perfectly well when all you're doing is sending messages between phones. But in applications with life-or-death implications, having your software "learn on the job" isn't good enough.

I wouldn't be surprised. Uber doesn't strike me as a very ethical company and prove to taking shortcuts.

Dardania wrote:

I'm wondering why Volvo's pedestrian protection technology didn't kick in, whatever about Uber's...

No technology can defy the laws of physics when it comes to a fast moving vehicle and a pedestrian that steps out at the last moment.

Johnboy1951

Dardania wrote: »

One of the articles this morning mentioned that police believe the woman just walked into traffic. Possibly homeless. Car was doing 38mph in a 35mph zone. Driver first knew about it when they heard the bang.

An autonomous vehicle that was exceeding the speed limit should be of concern to all.

When a human is driving one of the main things to watch for is the unexpected, such as someone stepping into traffic.

With the technology we expect these vehicles to have on board, such as multiple cameras etc., I would expect an autonomous vehicle to react much quicker to an unexpected event than a human. After all it can monitor all views simultaneously, whereas the human driver cannot, and often relies on reacting to some event caught 'out of the corner of the eye'.

All that being said, there are times when a fatal injury is unavoidable, by either human driver or autonomous vehicle.

Johnboy1951

This says the pedestrian was walking with a bicycle
https://www.nytimes.com/2018/03/19/technology/uber-driverless-fatality.html

The Uber car, a Volvo XC90 sport utility vehicle outfitted with the company’s sensing system, was in autonomous mode with a human safety driver at the wheel but carrying no passengers when it struck Elaine Herzberg, a 49-year-old woman, on Sunday around 10 p.m.

Sgt. Ronald Elcock, a Tempe police spokesman, said during a news conference that a preliminary investigation showed that the vehicle was moving around 40 miles per hour when it struck Ms. Herzberg, who was walking with her bicycle on the street. He said it did not appear as though the car had slowed down before impact and that the Uber safety driver had shown no signs of impairment. The weather was clear and dry.

There are quite a number of questions to be answered it seems. Not least of which is the vehicle not slowing before impact; the speed it was travelling and why the human driver did not react.

bk

The latest it seems is that the Sheriff is saying they have reviewed the footage and basically their was no way that either a human driver or automated could have avoided the crash. And no charges will be brought against Uber:

Based on the footage, Moir said that the driver had little time to react. “The driver said it was like a flash, the person walked out in front of them,” she said. “His first alert to the collision was the sound of the collision.”

She added, “It’s very clear it would have been difficult to avoid this collision in any kind of mode [autonomous or human-driven] based on how she came from the shadows right into the roadway.”

It would seem the lady suddenly stepped onto the road and couldn't be avoided.

BTW while initially the police said the speed limit is 35mph, photos on Google Maps would indicate it is 45mph.

Looking at Google Maps at the precise location where it happened, it looks like she was trying to cross an 8 lane road!! Imagine trying to cross the M50, only wider!

It looks like that area of road is very badly designed for pedestrians and has on unofficial crossing where some people might try and cross to a bus stop on the other side.

Unfortunately someone suddenly stepping in front of a car travelling at a high speed, is something that no car could avoid, no matter human or AI driven.

liamog

seamus wrote: »

They've been developing their autonomous software using Agile-like, "minimum viable product", "testing in production" methodologies. These work perfectly well when all you're doing is sending messages between phones. But in applications with life-or-death implications, having your software "learn on the job" isn't good enough.

Human drivers learn in the exact same manner, they are taken out in supervised conditions and learn how to control the vehicle. At some they pass a bar where they are able to fully control the car on their own.

In acknowledgement that drivers are not 100% educated the moment they pass their test, we now have the "N" plate which carries a couple of restrictions.

The advantage for autonomous software is that each lesson only needs to be learned once for the whole fleet.

Bob24

liamog wrote: »

Human drivers learn in the exact same manner, they are taken out in supervised conditions and learn how to control the vehicle. At some they pass a bar where they are able to fully control the car on their own.

In acknowledgement that drivers are not 100% educated the moment they pass their test, we now have the "N" plate which carries a couple of restrictions.

The advantage for autonomous software is that each lesson only needs to be learned once for the whole fleet.

You are talking about something different. The poster was questioning their software development methodology (how they work do develop and test the software).

What you are describing here is your expectation that their software is using machine learning (which is probably correct but an different topic from how they are developing it).

unkel

FWIW a very basic system like Tesla Autopilot (autonomous driving level 2) is involved in about half the number of fatalities per million km, compared to cars that don't have it. This is already bringing insurance premiums down for Tesla owners.

liamog

There is nothing inherently wrong with delivering an MVP. I'd much rather they did that and showed constant improvement than wait for a mathematically proven algorithm that covers all possibilities.

A novice driver is the MVP of drivers

eeguy

seamus wrote: »

Eventually only a couple of companies will actually own and maintain this software and all other manufacturers will use it under licence. But in the rush to become one of those companies, others must fall by the wayside. Uber is the first, but won't be the last.

100% agree with this. The comparison is the early days of mobile phones when every manufacturer had at least one OS.

Now we're down to two, iOS and Android.

Personally I'd say it'll come down to Google and one of the established car manufacturers (Volvo/Renault/Ford) who'll hold the keys to this technology.

Stheno

eeguy wrote: »

100% agree with this. The comparison is the early days of mobile phones when every manufacturer had at least one OS.

Now we're down to two, iOS and Android.

Personally I'd say it'll come down to Google and one of the established car manufacturers (Volvo/Renault/Ford) who'll hold the keys to this technology.

Most of the car manufacturers at the forefront of autonomous tech are in consortiums with a few key tech providers

jimmycrackcorm

It's being reported that the police are already indicating that uber isn't at fault based on the in car video recording.

boardzz

Don’t think this story is related to electric vehicles or hybrids

bertie4evr

Don't think I could react to this any better than the car did (Video in the tweet doesn't show any of the impact).

I can see all this being pinned on the driver now - they weren't looking at the road half the time.

https://twitter.com/brahmresnik/status/976587029390663680

eeguy

bertie4evr wrote: »

Don't think I could react to this any better than the car did (Video in the tweet doesn't show any of the impact).

I can see all this being pinned on the driver now - they weren't looking at the road half the time.

https://twitter.com/brahmresnik/status/976587029390663680[i/URL]

I couldn't have reacted much better. I've said before that I'm a big fan of the cars, but those sensors aren't restricted by visible light.

Im no expert but I would have thought the car should have picked that up on IR or Lidar.

Dardania

eeguy wrote: »

bertie4evr wrote: »

Don't think I could react to this any better than the car did (Video in the tweet doesn't show any of the impact).

I can see all this being pinned on the driver now - they weren't looking at the road half the time.

https://twitter.com/brahmresnik/status/976587029390663680[i/URL]

I couldn't have reacted much better. I've said before that I'm a big fan of the cars, but those sensors aren't restricted by visible light.

Im no expert but I would have thought the car should have picked that up on IR or Lidar.

Totally agree - I remember all those demos on Top Gear about the new (in 2006) Merc S-Class with IR, seeing further than visible in the normal light spectrum.
The other point to make is...the lady pushing the bike across the road - she should have been able to see the car bearing down on her. The road was straight before her...

Dardania

eeguy wrote: »

bertie4evr wrote: »

Don't think I could react to this any better than the car did (Video in the tweet doesn't show any of the impact).

I can see all this being pinned on the driver now - they weren't looking at the road half the time.

https://twitter.com/brahmresnik/status/976587029390663680[i/URL]

I couldn't have reacted much better. I've said before that I'm a big fan of the cars, but those sensors aren't restricted by visible light.

Im no expert but I would have thought the car should have picked that up on IR or Lidar.

Totally agree - I remember all those demos on Top Gear about the new (in 2006) Merc S-Class with IR, seeing further than visible in the normal light spectrum.
The other point to make is...the lady pushing the bike across the road - she should have been able to see the car bearing down on her. The road was straight before her...

Bob24

Agreed with previous posts - purely based on video analysis it was difficult to avoid as the person was in a shadow. But if the car had any kind of IR/radar sensor it does look like it could have picked it up and had time to break. This will need further inquiries.

Also the human driver might be in trouble. Doubtful she could have prevented it due to poor visibility, but still on the footage she doesn’t seem to constantly pay attendtion to the road and seems to be looking at something else inside the car.

Bob24

Dardania wrote: »

The other point to make is...the lady pushing the bike across the road - she should have been able to see the car bearing down on her. The road was straight before her...

Yes clearly her behaviour is very dangerous, she must either not have been able to pay normal attention for whatever reason or she was reckless (I regularity see people here in Dublin who just don’t care and walk in the middle of traffic expecting cars to stop for them).

But even though the pedestrian made the first mistake, the car should be designed to prevent collision if possible (although it gets me thinking, if cars become completely foolproof and can relabily coordinate and stop to prevent collisions, aren’t we gonna get even more idiots who walk in the middle of traffic and will we need some enforceable laws to punish that?)

_Brian

From what I’ve read online she was so close when she walked out that there was no time for anything to save her.

Apparently there’s dash cam footage kicking about but I’ve not seen it.

Bob24

_Brian wrote: »

From what I’ve read online she was so close when she walked out that there was no time for anything to save her.

Apparently there’s dash cam footage kicking about but I’ve not seen it.

The footage bertie4evr posted here does seem to indicate there would have been time if some kind of radar had detected the pedestrian. But not based on visual analysis of the situation as she was coming from a shadow.

So doubtful a human driver could have prevented the accident, but it was probably doable for a machine with better vision.

B-D-P--

Rumors are terrible thing:
"She was crossing 8 lanes of traffic"

Sad affair, you'd imagine the IR would have picked her up, the supervisor, however distracted he was, wouldn't have been able to react even if he was paying attention.

Hopefully they can learn from the crash as to what went wrong.

I'm not taking away from the tradigy, But if a person crashes into another person, you can only hope that single driver learns from the mistake.
If a Autonomous Vehicle does the same, the developers can check what happened and correct it for every autonomous vehicle, Its scary, but they are the future.

ted1

I think the footage is misleading, or rise Uber are just relying on cameras. Because the LIDAR, SONAR, Night cameras would surely have picked her up.

As for the safety driver. As I guessed they are just there to tuck a box

Bob24

ted1 wrote: »

I think the footage is misleading, or rise Uber are just relying on cameras. Because the LIDAR, SONAR, Night cameras would surely have picked her up.

Either this or the system simply didn’t do its job due to a hardware or software issue.

Impossible to tell at this point and we might never know.

kaymin

Looking at the video the driver should have seen some glints from the bicycle if they had been paying attention. An attentive driver would have at least applied the breaks before impact - there would have been time imo. It's not like she was walking out from a hedgerow- she was walking out from another lane on a long flat wide road.

KCross

I agree with all the other posts about, why didnt the car "see" her. If you pause at around 3 or 4 secs you can just see the person coming into human view... the car should have had "plenty" time to see her in advance of that and apply brakes to limit damage (I dont think full avoidance was possible). The issue is probably that the software is looking for cars up ahead and not yet clever enough to pickup a pedestrian crossing.



There must be a good 20-30 yards between the car and pedestrian there and they are already half way across the cars lane. The car should have seen her as soon as she crossed the white line!

I'd still have worries about the tech. For instance, driving through a town/village with lots of parked cars you (as a human) would be watching for the potential for someone to walk out between those cars... the autonomous car isnt.... its driving at the speed limit and waiting to react rather than pre-empting like a human would. It clearly needs to advance a lot more to be safe around pedestrians..... its great for preventing rear end crashes and staying in lane etc but its a bit away from replacing a human, I think.

bk

B-D-P-- wrote: »

Rumors are terrible thing:
"She was crossing 8 lanes of traffic"

She was! Here is the location that she was hit, she was hit on the right hand side, near where the four lines on the footpath are. Going from left to right, she would have already crossed the 3 lanes on the left before the median and would then have been crossing the 5 lanes on the right:



If you zoom out on Google Maps you see the whole area is on and off ramps of wide roads. Very poor pedestrian design around there.

Johnboy1951

IMO, the human 'driver' should have had more visibility.
If I were driving that road I would most certainly want better visibility ahead, and probably had put on main beams intermittently to ensure I had full information of what was ahead.
So if this was a car driven by a human only, that accident was (if not completely) avoidable, as brakes could have been applied, and avoidance action taken by crossing into the adjacent left lane.

So IMO, Uber most definitely have serious questions to answer.

Why did the autonomous system not recognise an obstacle in front?
Was it because of a malfunction of some sensors?
Was it because they do not have sensors suited to driving in such conditions (low light)? If this so, why did it not apply better lighting?

Yes of course the pedestrian made a stupid decision and paid the ultimate price.
The vehicle should not have struck her without, at minimum, braking and attempting avoidance.
Conditions were good on the road, and it was not busy.

I hope we eventually hear what really caused the collision .... failure of some system or maybe the complete lack of a system that can deal with this situation?

eeguy

Johnboy1951 wrote: »

So IMO, Uber most definitely have serious questions to answer.

Why did the autonomous system not recognise an obstacle in front?
Was it because of a malfunction of some sensors?
Was it because they do not have sensors suited to driving in such conditions (low light)? If this why did it not apply better lighting?

What I really love about autonomous driving, is that when these questions have been answered, the cars won't make the same mistake again.

bk

Having watched the video, I don't think a human driver would have done any better. There is only about one second between her becoming visible and getting hit. Human reaction time is about 2.3 seconds and at that speed, there is little hope of stopping in time.

BTW this is one of the reasons why people want to reduce speed limits in our cities to 30km/h. It gives you more reaction time.

It also looks like the street lighting didn't help. There are street lights ahead of where she crossed, but not where she crossed, so the lights ahead blind you to the area where she crossed, putting it in very dark shadow. She only becomes visible when the cars head lamps hit her and at that speed it would already be too late for any human driver.

However it does look like Lidar should have been able to see her and react. So they are certainly questions to be asked.

But then that raises a question. Do we expect this tech to be just as good as humans, or better?

Johnboy1951

bk wrote: »

Having watched the video, I don't think a human driver would have done any better. There is only about one second between her becoming visible and getting hit. Human reaction time is about 2.3 seconds and at that speed, there is little hope of stopping in time.

BTW this is one of the reasons why people want to reduce speed limits in our cities to 30km/h. It gives you more reaction time.

It also looks like the street lighting didn't help. There are street lights ahead of where she crossed, but not where she crossed, so the lights ahead blind you to the area where she crossed, putting it in very dark shadow. She only becomes visible when the cars head lamps hit her and at that speed it would already be too late for any human driver.

However it does look like Lidar should have been able to see her and react. So they are certainly questions to be asked.

But then that raises a question. Do we expect this tech to be just as good as humans, or better?

Would you have driven that section of road without bringing your lights up so you could see ahead?
In effect you are comparing a human driving with inadequate lighting and the autonomous vehicle which is reported to not require the same level of lighting for correct operation.

If that car was controlled by a human driver then I would say the driver would have to burden a lot of the blame ...... essentially for driving blind!

I expect autonomous vehicles to be as good as a human in all situations, else they should not be approved.
In most situations requiring quick reactions, the autonomous vehicle should be better than the human in reaction time.

That alone requires the autonomous system to be capable of recognising potentially dangerous situations.
Otherwise they should not be allowed.

Bob24

bk wrote: »

But then that raises a question. Do we expect this tech to be just as good as humans, or better?

Isn’t it obvious we expect it to be better?

I mean being as good is the absolute minimum requirement when safety is involved. But if technology could fairly easily allows to be make better decisions (in this case by seeing an object a human eye couldn’t see and having a shorter reaction time), why shouldn’t we aim for that?

gk5000

As above - you should not drive faster than your eyes can see - be it dips or headlights

Otherwise - a human would not see a corner, a truck broke down etc....

The video may not show full human vision - but in general you should drive at a speed so that you can stop if something is blocking your path.

I believe many human drivers would have braked and if needed swerved to avoid her.

What the video clearly demonstrates is how important it is to make yourself visible. I'm a cyclist myself always wear full reflective gear, spoke reflectors and bright light at both ends. It's crazy how many cyclists are like the woman here dressed in black. Every day I meet cyclists who are dressed in all black and don't even have any reflectors let alone lights. In many cases I have also met joggers on a dark cycle lane and the only reason I have seen them (very late normally) is because I have a very bright headlamp and they have a tiny reflective strip of material on their shoes that eventually reflects some light. Then the rest of the black jogger appears form dark 5 meters from you.

Scary stuff.

Johnboy1951

samih wrote: »

What the video clearly demonstrates is how important it is to make yourself visible. I'm a cyclist myself always wear full reflective gear, spoke reflectors and bright light at both ends. It's crazy how many cyclists are like the woman here dressed in black. Every day I meet cyclists who are dressed in all black and don't even have any reflectors let alone lights. In many cases I have also met joggers on a dark cycle lane and the only reason I have seen them (very late normally) is because I have a very bright headlamp and they have a tiny reflective strip of material on their shoes that eventually reflects some light. Then the rest of the black jogger appears form dark 5 meters from you.

Scary stuff.

All true in my experience.
Also true is that the woman should not have been on the road at all.

But given the circumstances that existed, should we expect a better outcome?

I certainly do ..... regardless a human controlled or autonomous vehicle involved.

Bob24

It also shows a potential danger of semi-autonomous vehicles whereby the vehicle is effectively able to do everything but is not trusted to be reliable without human supervision to quickly take over in case something goes wrong.

I’m not finding excuses for the test driver or anything, but when you routinely spend hours in such car and are used to just being a spectator and letting the car make all the decisions, it is very hard to keep your full attention span - both due to boredom and to a false sense of security about a system which has been doing everything.

I am sure test drivers losing attention to the road happens constantly and companies employing them know it and have plenty of video footage showing it, but as mentioned in a previous post they might just see the human driver as a way to thick a regulatory box.

If humans are to be trusted to quickly take over from a machine as required, something has be done to keep them constantly engaged with the driving process so that if instant intervention is ever required they always have full attention on the road and are aware of what’s going on.