Self driving buses, trains, trucks etc

cnocbui · 20-03-2018 1:16pm

AndrewJRenko wrote: »

I wonder if the other 100 road deaths caused by motorists in the USA on that particular day will get anything like the same level of coverage and analysis?

Google have 8 Million Km logged in testing. If you were to be exceptionally generous and allow another 4 M KM for all the other autonomous driving developers, that would be 12 M. We now have 1 death per 12 M km driven.

How do humans do? Australia is about middle ranking for OECD countries. There, the fatality rate is 0.49 per 100 M km, so 1 per 200 M km. You can halve that rate for Norway who have the best rate.

I saw someone calculate the chances of this autonomous driving fatality having occurred at this point, if it actually was as safe as a human and it was about 10 to the -5, or one in 10,000, so autonomous driving has a long way to go before being anything close to being as good as humans.

All these ridiculous claims of autonomous driving being significantly safer and better than humans need to stop now and the authorities need stop listening to tech fanbois who think humans are bad but software and computers are better.

mickoneill31 · 20-03-2018 1:31pm

cnocbui wrote: »

Google have 8 Million Km logged in testing. If you were to be exceptionally generous and allow another 4 M KM for all the other autonomous driving developers, that would be 12 M. We now have 1 death per 12 M km driven.

How do humans do? Australia is about middle ranking for OECD countries. There, the fatality rate is 0.49 per 100 M km, so 1 per 200 M km. You can halve that rate for Norway who have the best rate.

I saw someone calculate the chances of this autonomous driving fatality having occurred at this point, if it actually was as safe as a human and it was about 10 to the -5, or one in 10,000, so autonomous driving has a long way to go before being anything close to being as good as humans.

All these ridiculous claims of autonomous driving being significantly safer and better than humans need to stop now and the authorities need stop listening to tech fanbois who think humans are bad but software and computers are better.

.
Edit.
Tesla has 1.3b miles logged on autopilot.

https://electrek.co/2016/11/13/tesla-autopilot-billion-miles-data-self-driving-program/amp/

cnocbui · 20-03-2018 1:52pm

mickoneill31 wrote: »

.
Edit.
Tesla has 1.3b miles logged on autopilot.

https://electrek.co/2016/11/13/tesla-autopilot-billion-miles-data-self-driving-program/amp/

Apples vs Oranges. The Tesla autopilot is category 1 and is only an assistant, the Volvo/Uber was category 5, fully autonomous. The autopiliot is managing the easiest, least demanding and lowest accident rate type of driving.

On April 28, 2017, Elon Musk predicted that in around two years drivers would be able to sleep in their Tesla until it finishes the trip.[25]

In the middle of 2017, Teslas plans were to demonstrate full self-driving by the end of 2017.[26][27][28] As of March 2018, the demonstration has not taken place.

Exactly the sort of over-optimistic foolishness that needs to be stopped by authorities.

plodder · 20-03-2018 1:56pm

cnocbui wrote: »

Google have 8 Million Km logged in testing. If you were to be exceptionally generous and allow another 4 M KM for all the other autonomous driving developers, that would be 12 M. We now have 1 death per 12 M km driven.

How do humans do? Australia is about middle ranking for OECD countries. There, the fatality rate is 0.49 per 100 M km, so 1 per 200 M km. You can halve that rate for Norway who have the best rate.

I saw someone calculate the chances of this autonomous driving fatality having occurred at this point, if it actually was as safe as a human and it was about 10 to the -5, or one in 10,000, so autonomous driving has a long way to go before being anything close to being as good as humans.

All these ridiculous claims of autonomous driving being significantly safer and better than humans need to stop now and the authorities need stop listening to tech fanbois who think humans are bad but software and computers are better.

It depends whose fault the accident was though. If it was entirely the pedestrian's fault then your figures are pretty meaningless. A data set with only one point doesn't exactly establish a trend either.

One thing that would be interesting to find out would be incidences where the human operator has to take over when the automation does something stupid. There's likely to be a much larger dataset and they'd want to be making that public before they should be allowed to move onto the next stage of testing with no human operator.

LeinsterDub · 20-03-2018 2:00pm

cnocbui wrote: »

Exactly the sort of over-optimistic foolishness that needs to be stopped by authorities.

How or why would the authorities stop this ? He's entitled to his opinion and to express it just like you are.

eeguy · 20-03-2018 2:10pm

cnocbui wrote: »

Apples vs Oranges. The Tesla autopilot is category 1 and is only an assistant, the Volvo/Uber was category 5, fully autonomous. The autopiliot is managing the easiest, least demanding and lowest accident rate type of driving.

Exactly the sort of over-optimistic foolishness that needs to be stopped by authorities.

Tesla is Level 3. Level 1 is basically every car built before 2005.

The Volvo and Uber are category 4. Cat 5 are miles away.

Elon Musk is famous for overly optimistic deadlines. Nothing new there.

cnocbui · 20-03-2018 2:17pm

LeinsterDub wrote: »

How or why would the authorities stop this ? He's entitled to his opinion and to express it just like you are.

I wasn't meaning to refer to Musk making statements, of course he's entitled, i meant allowing Tesla to implement autonomous driving when it's being pushed and influenced by a ridiculously over-optimistic CEO who seems to advocate anything but caution.

You get people believing the hype and then defeating the autopilot restrictions deliberately because they think the hype is to be believed and the sober user manual instructions are to be ignored because they are obviously far too cautious. Boardsies talking about non gastronomic uses for oranges, as an example.

bk · 20-03-2018 6:34pm

Interestingly the latest from the Sheriff where this happened, is that after reviewing the footage from the car, they don't think a human or automated car could react to how quickly the crash occurred and don't expect to charge 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.”

More interestingly it seems it may be yet another case of terrible street/footpath design where people were tempted to cross where they weren't suppose to:

https://twitter.com/EricPaulDennis/status/975891554538852352/photo/1

Looking at the location myself, it looks like the poor lady was trying to get onto a terribly designed cycle lane that begins in the middle of a 4 lane road (one way, 8 total)! Or making a crossing to a bus stop, through an unofficial but clearly regularly used path across the 8 lane road.

AndrewJRenko · 20-03-2018 10:57pm

cnocbui wrote: »

All these ridiculous claims of autonomous driving being significantly safer and better than humans need to stop now and the authorities need stop listening to tech fanbois who think humans are bad but software and computers are better.

Let me know if you see anyone round here making those ridiculous claims, as that's certainly not what I said.

My point was simply a comment about the extensive degree to which society is apparently immune to road deaths caused by human drivers.

magicbastarder · 20-03-2018 11:17pm

cnocbui wrote: »

All these ridiculous claims of autonomous driving being significantly safer and better than humans need to stop now and the authorities need stop listening to tech fanbois who think humans are bad but software and computers are better.

we've gone from a point ten years ago from where self driving cars were a near fantasy, to where one causing this accident is literally remarkable.

the technology is far from mature; to judge the whole techology, and its potential, based on the current state of it is incredibly myopic.
(especially given that this accident involved the organisation most given to overhyping its capabilities, out of all the companies doing research in the area).

HonalD · 21-03-2018 9:22pm

cnocbui wrote: »

Apples vs Oranges. The Tesla autopilot is category 1 and is only an assistant, the Volvo/Uber was category 5, fully autonomous..... Exactly the sort of over-optimistic foolishness that needs to be stopped by authorities.

Perhaps the lack of factual accuracy in this post needs to be stopped by authorities first.

TEsla autopilot is not Level 1 and Level 5 doesn't exist yet.

mickoneill31 · 22-03-2018 9:46am

The dash cam footage of the Uber accident is on the Journal.

http://www.thejournal.ie/uber-footage-self-driving-3917329-Mar2018/

I can't speak for anybody else but I don't think I'd have been able to avoid hitting her either.

Its not a nice video to watch so don't click it if you don't want to see somebody being hit by a cat.

magicbastarder · 22-03-2018 10:16am

mickoneill31 wrote: »

I can't speak for anybody else but I don't think I'd have been able to avoid hitting her either.

but that's not the measure of what an autonomous vehicle should be able to do (or at least, the promises being made for them).
the car doesn't use human eyes to see - it has systems which operate on a broader spectrum than human sight is capable of, so should have been able to see her.

cnocbui · 22-03-2018 10:23am

Having seen the dashcam footage, I would immediately pull all of Uber/Volvo's licenses for autonomous driving.

The woman did not appear out of nowhere and jump out from behind bushes as the initial reports hinted at. She was crossing a two lane road and had already crossed the first lane.

The dashcam footage is totally misleading, which is why people and the tech-ignorant police say she appeared out of nowhere. Dashcams are built to a price and more than likely have 4 generation old image sensors in them the size of your little fingernail. The sensor wouldn't be close to what are in current flagship phones and even those aren't even close to being as good as what are in a DSLR in low-light capabilities.

Even full-frame top-end DSLR camera sensors aren't even close to having the dynamic range of the human visual system. The dashcam shows nothing but black before you see her feet, as if the headlights have no spill beam beyond the main beam, but of course for human vision that is not the case. A human would have picked her up way earlier when she was still on lane 1 just from spill beam and the streetlight on the left.

Even if the safety driver had eyes as bad as that dashcam, then the car was considerably outdriving it's lights as far as she was concerned, so again no excuses. Even if my eyes were as bad as the dashcams I would have at least been on the brakes before hitting her at a reduced speed.

Hopefully the image sensors actually used by the car for guidance and registering the environment would be considerably better than a cheap dashcam.

Leaving aside the inadequacies of the dashcam, the Volvo guidance system incorporates a LIDAR which scans the environment with an infrared laser to create a 3D model of it. That system can operate in total darkness and doesn't rely on headlights or streetlights. No doubt that system 'saw' her clearly well before the dashcam did but the software interpreting and doing decision making on the model failed totally - didn't even begin to apply the breaks or swerve.

Poor woman has probably done this before and relied on human drivers seeing her and making allowances.

magicbastarder · 22-03-2018 10:27am

also, it's worth noting that - based on the current information - the car did not attempt to brake; even at the point that she was clearly visible in the headlights. the car should have been able to apply the brakes with a reaction time that is a small fraction of human reaction time, so the failure to even attempt to brake sounds like a critical failure.

bk · 22-03-2018 10:44am

Having watched the video, I don't think any driver would have been able to avoid that crash.

There is less then 1 second between her becoming visible and getting hit. At 38mph no one would be able to stop.

If this had been a human driver and the above video was seen, we wouldn't even be discussing it, it would be another sad example of the 100 pedestrians a day that die on US roads.

Clearly the lighting on the road didn't help. Their are bright lights in front of her, but seemingly non where she crossed, so she is completely in shadow until the cars headlights come on her, by which time it is already too late at those speeds.

Also, I'd hate to victim blame, but she did cross a road as wide as the M50, with no lights or reflectors on and seemingly not even looking at the oncoming car herself.

What does surprise me, is that the Lidar should have detected her and at least tried to stop. So there are definitely questions to be answered.

But then that raises the question, should we expect self driving cars to be as good as a human or better?

But it also highlights the need for good road and pedestrian design (it was very poor here), but also highlights that even with self driving cars, people can't just walk across the equivalent of the M50 in pitch dark with no lights or reflectors.

cnocbui · 22-03-2018 10:52am

bk wrote: »

Having watched the video, I don't think any driver would have been able to avoid that crash.

There is less then 1 second between her becoming visible and getting hit. At 38mph no one would be able to stop.

If this had been a human driver and the above video was seen, we wouldn't even be discussing it, it would be another sad example of the 100 pedestrians a day that die on US roads.

Clearly the lighting on the road didn't help. Their are bright lights in front of her, but seemingly non where she crossed, so she is completely in shadow until the cars headlights come on her, by which time it is already too late at those speeds.

Also, I'd hate to victim blame, but she did cross a road as wide as the M50, with no lights or reflectors on and seemingly not even looking at the oncoming car herself.

What does surprise me, is that the Lidar should have detected her and at least tried to stop. So there are definitely questions to be answered.

But then that raises the question, should we expect self driving cars to be as good as a human or better?

But it also highlights the need for good road and pedestrian design (it was very poor here), but also highlights that even with self driving cars, people can't just walk across the equivalent of the M50 in pitch dark with no lights or reflectors.

If you have a phone with a camera, go out at night and take a video of a not well-lit street. Still standing there, play back the short video and compare what your own eyes register in the areas on the video that are just black. I am confident that you will see detail in the dark areas that you can not make out on the video.

mickoneill31 · 22-03-2018 11:06am

bk wrote: »

What does surprise me, is that the Lidar should have detected her and at least tried to stop. So there are definitely questions to be answered.

But then that raises the question, should we expect self driving cars to be as good as a human or better?

.

+1

If it was a human driver we wouldn't be talking about it and there have probably been 10 such incidents involving humans similar to this since the Uber incident.

Im not saying it shouldnt be examined with a fine tooth comb but if we applied the same standards to humans (and I think we should) then loads of people would be getting their licence revoked.

mickoneill31 · 22-03-2018 11:07am

cnocbui wrote: »

If you have a phone with a camera, go out at night and take a video of a not well-lit street. Still standing there, play back the short video and compare what your own eyes register in the areas on the video that are just black. I am confident that you will see detail in the dark areas that you can not make out on the video.

Are the cars just using the equivalent of phone cameras?
Serious question. I don't know how good the cameras are. And Id assumed they'd have other supplimentory sensors.

Edit: ah I was editing while cnocbui was replying.

cnocbui · 22-03-2018 11:10am

mickoneill31 wrote: »

Are the cars just using phone cameras?

No, that is my point. Judging this incident on crappy dash cam footage is like saying the car is guided by the vision supplied by the camera in a 3-4 year old mid-ranking phone.

magicbastarder · 22-03-2018 11:14am

mickoneill31 wrote: »

Serious question. I don't know how good the cameras are. And Id assumed they'd have other supplimentory sensors.

they use LIDAR as well as other camera systems; LIDAR should have no issues with low light or darkness.

https://techcrunch.com/2018/03/19/heres-how-ubers-self-driving-cars-are-supposed-to-detect-pedestrians/

bk · 22-03-2018 11:24am

cnocbui wrote: »

If you have a phone with a camera, go out at night and take a video of a not well-lit street. Still standing there, play back the short video and compare what your own eyes register in the areas on the video that are just black. I am confident that you will see detail in the dark areas that you can not make out on the video.

Now add night blindness caused by the bright street lights in front of her, but casting a shadow on where she is and a speed of 38mph and I honestly don't believe any of us would do any better.

magicbastarder · 22-03-2018 11:40am

the crucial point here is not that it failed because the incident was beyond its capabilities to deal with. it failed because there seems to have been a critical error in achieving what it is/should be capable of.

they will have a serious credibility issue now. how do they convince the authorities that they've found and fixed the bug? or that there are no other bugs?

cnocbui · 22-03-2018 11:46am

bk wrote: »

Now add night blindness caused by the bright street lights in front of her, but casting a shadow on where she is and a speed of 38mph and I honestly don't believe any of us would do any better.

I disagree totally. If your vision actually is as poor as that dashcam footage I hope you never drive at night.

I have marked The woman's shoe coming into visibility on the dashcam. It shows beyond question that she is crossing the road where it is lit on both sides by streetlights. She did not cross the road where it was unlit. The road only appears dim and unlit because the dashcam would be adjusting the exposure for the bright headlight-lit part of road, thus rendering darker areas far darker than they would appear to a human.

An iPhone 6 camera has a dynamic range of about 8 stops - a stop being a doubling. The human eye/brain visual system has a dynamic range in the order of 22 - 24 stops. The scale is logarithmic, not linear. You just can't say what you would have done or seen based on a lousy dashcam video. Where a dashcam camera renders pitch black, the human eye would see detail.

cnocbui · 22-03-2018 12:20pm

That is the same bit of road filmed with a TV broadacast camera - still not nearly as good as human eyes/brain are and shows how poor the dashcam is.

That was not a poorly lit bit of road.

magicbastarder · 22-03-2018 12:36pm

i don't think that's a useful photo though - it's almost beyond doubt that there would be extra lighting there in the wake of the collision.

bk · 22-03-2018 12:42pm

cnocbui, if you look on Google Maps, you can clearly see that the street light is ahead of the point where she crossed. I suspect from the footage that the street light points in the direction that the car is coming from and would cause a night blindness situation for a person driving towards the area.

A well designed crossing would put the street light behind where you cross, not in front of you. So that the light is cast across you.

I agree of course this shouldn't make a difference to LIDAR.

You should really take a look at the area on Google Maps. It really is a text book example of terrible and dangerous road design from a pedestrians perspective.

Looking at the area, you can easily see how pedestrian unfriendly US roads are. The whole area is covered in wide motorway style roads. Yet there are houses just next to it and they expect people to cross at crossings of 8 lanes wide!

Such roads like this mostly don't exist in Ireland, Motorways tend to be fully segregated from housing, etc. and there are no footpaths on them. Where people need to cross, we have overpasses.

THis shows how little the US tends to care for pedestrians and cyclists.

seamus · 22-03-2018 12:42pm

magicbastarder wrote: »

they will have a serious credibility issue now. how do they convince the authorities that they've found and fixed the bug? or that there are no other bugs?

They won't, basically. This will be the end of the road for Uber's foray into writing their own self-drive software.

This kind of error should never have made it off the test track, so forget about this incident - what other basic conditions has Uber failed to test in its software?

The manufacturers now need to set up a consortium to agree a framework of "readiness" phases - setting out the tests that any automation software must be able to pass, over and over again, before they can move to the next phase of readiness.

And then only at readiness phase 9/10/11 or whatever should they be permitted on public roads, even with a safety driver.

cnocbui · 22-03-2018 12:44pm

magicbastarder wrote: »

i don't think that's a useful photo though - it's almost beyond doubt that there would be extra lighting there in the wake of the collision.

That scene, particularly the road surface, is mostly yellow-lit from the high pressure sodium streetlamps. Check the video, most of the extra lighting is from the flashing blue lights of parked police cars. There is no portable flood lighting or such as they don't need it because of the two street lights.

cnocbui · 22-03-2018 1:05pm

bk wrote: »

cnocbui, if you look on Google Maps, you can clearly see that the street light is ahead of the point where she crossed. I suspect from the footage that the street light points in the direction that the car is coming from and would cause a night blindness situation for a person driving towards the area.

A well designed crossing would put the street light behind where you cross, not in front of you. So that the light is cast across you.

I agree of course this shouldn't make a difference to LIDAR.

You should really take a look at the area on Google Maps. It really is a text book example of terrible and dangerous road design from a pedestrians perspective.

Looking at the area, you can easily see how pedestrian unfriendly US roads are. The whole area is covered in wide motorway style roads. Yet there are houses just next to it and they expect people to cross at crossings of 8 lanes wide!

Such roads like this mostly don't exist in Ireland, Motorways tend to be fully segregated from housing, etc. and there are no footpaths on them. Where people need to cross, we have overpasses.

THis shows how little the US tends to care for pedestrians and cyclists.

The frame grab from the dashcam clearly shows she was directly between two streetlights. I have drawn a straight line between the centres of the two pools of light and it goes through her shoe.

Self driving buses, trains, trucks etc

Comments