Self driving buses, trains, trucks etc

bk

cnocbui wrote: »

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.

If you look on Google Maps and use the measuring tool, you will see that those street lights are actually 15 meters ahead of where she crossed. Your screen capture doesn't accurately represent the distances.

I can post detailed screen captures from Google Maps if you'd like.

magentis

That really is terrible.Even a vehicle equipped with AEBS should be able to stop in that situation.It looks like the vehicle didn't even slow.Im sure the cars camera would have been capable of picking that woman up 50 metres before we saw her in the video.Most likely a hardware or software fault caused this,we will never be told most likely.Such is the drive to get these lethal vehicles on the road.

bk

Another thing that is not quite obvious from the footage, but you can see on Google Street View, is that initially the road is 2 lanes wide, it widens out to four lanes where she crosses. From further down the road, the extra two left hand turn lanes are hidden behind bushes where she came from behind.

cnocbui

bk wrote: »

If you look on Google Maps and use the measuring tool, you will see that those street lights are actually 15 meters ahead of where she crossed. Your screen capture doesn't accurately represent the distances.

I can post detailed screen captures from Google Maps if you'd like.

Go back a page in this thread and look at the photo of the accident scene. Where is the bike, where is the car and where is the streetlight? I don't car about Google streetmaps.

bk

cnocbui wrote: »

Go back a page in this thread and look at the photo of the accident scene. Where is the bike, where is the car and where is the streetlight? I don't car about Google streetmaps.

The pictures of the crash scene are 40 meters further up the road (North) from where the video shows the crash happening, as best I can tell.

Actually the street lights you point out from the video are actually 60 meters south of where the crash scene pictures are, so clearly they don't match. I suspect the lady must have been propelled forward up the road and that crash scene picture shows where the lady and car came to rest, not where the actual accident happened. Or the police may have moved both up the road, so that they are out of the way and allow traffic to pass.

bk

Here are some pictures of the area from Google Maps:

The top arrow indicates where she crossed and where the impact seemed to happen. I point out the two streets lights which are a good 15 meters ahead of where she crossed.

You can also clearly see the terrible road design here. The big x in the medium is a nicely paved area that looks like a nice pedestrian crossing, but it actually has signs on it saying don't cross here!! However you can clearly see why someone might decide to cross there, it looks like a crossing.

Also note that the point she crossed out is hidden from a distance behind bushes.



This picture is from 150 meters south and shows the closest speed sign which is clealry 45mph and also shows that from a distance, the point she was crossing out would be blocked from view by the bushes:



The red car in the distance is roughly behind from where she crossed.

eeguy

magentis wrote: »

That really is terrible.Even a vehicle equipped with AEBS should be able to stop in that situation.It looks like the vehicle didn't even slow.Im sure the cars camera would have been capable of picking that woman up 50 metres before we saw her in the video.Most likely a hardware or software fault caused this,we will never be told most likely.Such is the drive to get these lethal vehicles on the road.

I agree it should have picked up the woman, but why wouldn't we be told about it?
Most of this stuff becomes public quick enough and many companies disclose it to authorities in the US.

Here's Waymo's disengagement reports for last year. kinda interesting: https://www.dmv.ca.gov/portal/wcm/connect/42aff875-7ab1-4115-a72a-97f6f24b23cc/Waymofull.pdf?MOD=AJPERES

Also, all cars have the potential to be lethal, regardless of how they're driven.

bk

And here is a more zoomed out view of the area:



Not justifying it, but it certainly don't look like a normal residential area, more like something like the Red Cow and it shows the madness of US road design that mixes pedestrians on roads like this.

magentis

magentis wrote: »

That really is terrible.Even a vehicle equipped with AEBS should be able to stop in that situation

Stopping distance at 38mph exceeds the distance to her so no human was ever going to be able to stop the impact from happening although I agree, the software should have detected her (based on the information available about Uber autonomous system, which isn't much to be honest)

cnocbui

Note the position of the poor unfortunate victim in relation to the storm drain. note the position of the storm drain in relation to the street light.

She was practically under it.

You were certainly right about the car and bike being moved down the road to where the blue sign and next street light is. The two street lights would have lit her enough for a human driver to have seen her and prevented the collision.

bk

cnocbui wrote: »

Note the position of the poor unfortunate victim in relation to the storm drain. note the position of the storm drain in relation to the street light.

She was practically under it.

You were certainly right about the car and bike being moved down the road to where the blue sign and next street light is. The two street lights would have lit her enough for a human driver to have seen her and prevented the collision.

Again, measuring it on Google Maps, that street light is 15 meters in front of the storm drain. Your screen captures are distorting the distances.

So a bright light, 15 meters ahead of her, blinding an oncoming car and the drivers night blindness and putting the area where the lady crossed in darkness.

magicbastarder

Is the exact position that important? Uber use LIDAR, so lighting is not really relevant.

bk

magicbastarder wrote: »

Is the exact position that important? Uber use LIDAR, so lighting is not really relevant.

I agree with you on that. The conversation was more around if it would have made any difference had a human been driving. I don't think it would have, she would still be just another one of the 16 pedestrians that die every day on US roads and which no one seems to care about.

I agree that Uber has questions to answer about why their Lidar didn't pick her up and at least try to slow down (I think it would have still have hit her and caused terrible injuries, but might at least be still alive).

cnocbui

bk wrote: »

I agree with you on that. The conversation was more around if it would have made any difference had a human been driving. I don't think it would have, she would still be just another one of the 16 pedestrians that die every day on US roads and which no one seems to care about.

I agree that Uber has questions to answer about why their Lidar didn't pick her up and at least try to slow down (I think it would have still have hit her and caused terrible injuries, but might at least be still alive).

You seem totally unwilling to accept that the human visual system is considerably better than a cheap dashcam. Your mentioning of night blindness I find confusing. There are no bright lights in the video that would have impacted a human's ability to benefit from their vastly superior visual system dynamic range.

On the dash-cam style video, we only see her 1.5 seconds before impact. However, the human eye and quality cameras have a much better dynamic range than this video, and should have also been able to see her even before 5 seconds.
...
From the dashcam video, it seems that there was very little time to react, and no fault for a human driver hitting somebody who "came out of nowhere." The police may not have a good way to evaluate the vastly superior dynamic range of human vision compared to the camera.
...
Cameras and HDR

The Uber reportedly has a large array of cameras. Good. That usually means that the system has been designed to do "high dynamic range" vision for driving at night. That's because what we see here is common -- uneven lighting due to the headlamps and streetlamps. This means either 2 or more cameras with different exposure levels, or one camera constantly switching exposure level to capture both lit and unlit objects.

A vision system based on HDR should also have easily seen her and triggered the stop.

Another option, not used on most cars today, is a thermal "night vision" camera. I have written about these a few times and I experimented with them while at Google back in 2011. Then (and even now) they are quite expensive, and must be mounted outside the glass and kept clean, so teams have not been eager to use them. Such a camera would have seen this pedestrian trivially, even if all the lights were off (headlights, streetlamps etc.) (LIDAR also works in complete darkness.) I have not heard of Uber using such night-vision cameras.

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.

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

It has nothing whatsoever to do with resolution, it has to with dynamic range (the range of which a camera can successfully capture the lightest and darkest areas of an image without losing detail) But even if resolution was an issue, the resolution of the crappy video being played on the internet is not necessarily the original resolution of the camera that look it.

Further, here is something else for you to think about. If the video on the internet has been converted to match the requirements of the website playing it (and it almost certainly has) then I can tell you for a fact that dynamic range will badly affected by the conversion. Taking a, say 1920 x 1080 MP4 file and converting to a 640 x 480 FLV loses a tremendous amount of dynamic range, as much as three or four stops in most cases.

Now, I can tell you that while a human eye might be better that the crappy video we see on the internet, I would be absolutely horrified if the autodrive system was actually using a video camera with such a poor dynamic range.

PS: I am professional image finisher with over 20 years experience in the trade. I deal with this stuff every day of the week.

http://www.internationalskeptics.com/forums/showthread.php?p=12229337

Even though the video appeared dark, King said there was likely more visibility than the footage suggested... (David King, an Arizona State University professor and transportation planning expert)

Uber is a malignant company that should be dissolved. Apparently they were testing in Arizona because they had a disagreement with California about their cars running red lights.

Uber has launched an aggressive battle with California over its controversial self-driving cars, with regulators and consumer advocates accusing the corporation of flagrantly violating the law, endangering public safety and mistreating drivers.

The intense fight with the state – which ignited hours after numerous self-driving cars were caught running red lights in Uber’s home town – has exposed what critics say are the unethical and illegal tactics that the company has repeatedly used to grow its business.
Uber blames humans for self-driving car traffic offenses as California orders halt
Read more

The ride-sharing company, which launched semi-autonomous vehicles in San Francisco without permits this week, was ordered by the California department of motor vehicles (DMV) to immediately remove the cars from the road or face legal action.

But Uber, which has not publicly responded to the state’s demands, blamed the traffic light violations on “human error” and suspended the drivers who were monitoring the cars. This bold deflection of blame further highlights the corporation’s refusal to take responsibility for potential faults in its technology and raises questions about the dangers of prematurely rolling out self-driving vehicles.

https://www.theguardian.com/technology/2016/dec/16/uber-self-driving-cars-california-illegal-unethical-tactics

magicbastarder

apparently the site of the collision appears at 33s in:

https://www.youtube.com/watch?v=CRW0q8i3u6E

cnocbui

Thanks. That's enough to make you wonder if Uber has doctored the video they released to make their obvious guilt seem like it innocence.

eeguy

cnocbui wrote: »

Thanks. That's enough to make you wonder if Uber has doctored the video they released to make their obvious guilt seem like it innocence.

I'd wait until there's an official statement before jumping to conclusions. There'll have to be an investigation and we'll find out all about the incident then.

plodder

On the point about autonomous systems having instant reactions, that’s not something I would just assume to be the case. Quite a lot of processing goes into interpreting the sensor data, assuming the sensors are delivering the relevant data. A feature of proving that these systems can ever be safe, will be demonstrating quick reactions when something unexpected happens. It’s surprising really that this (or at least full public disclosure of it) hasn’t been done already.

Slightly related, but my OH’s car has a camera that operates the headlights automatically, switching between full and dipped beam. It does an ok job, and works best when it sees cars first, from way in the distance. But, it does take a noticeable time to react on typical Irish twisty roads and it does happen that people get annoyed and flash you when it takes too long on occasion.

Carawaystick

Are the NTSB gonna have an investigation?

cnocbui

Tesla shares have just plummeted 6% because of the latest fatality which may well have been caused by the autopilot.

cnocbui

cnocbui wrote: »

Tesla shares have just plummeted 6% because of the latest fatality which may well have been caused by the autopilot.

That's not unusual. Associated firms in a similar business often take a hit when there's an issue in an industry.

It's possible the software was at fault, won't know until the full investigation is completed. The makers of the lidar equipment stated their equipment can see perfectly well in the dark and that the likely cause was due to the Uber software not being setup correctly to interpret the data being sent from the lidar

Anyway, it's a waiting game until a full report is issued

magicbastarder

Deleted User wrote: »

That's not unusual. Associated firms in a similar business often take a hit when there's an issue in an industry.

just in case of confusion - there's been another fatality involving a tesla:

https://www.mercurynews.com/2018/03/27/ntsb-investigating-fatal-crash-on-highway-101-involving-a-tesla/

magicbastarder

uber had disabled the volvo collision avoidance system in the vehicle involved in the fatality - that explains why that system did not act as a backup:

https://www.bloomberg.com/news/articles/2018-03-26/uber-disabled-volvo-suv-s-standard-safety-system-before-fatality?cmpid=socialflow-twitter-business&utm_medium=social&utm_source=twitter&utm_content=business&utm_campaign=socialflow-organic

cnocbui

All this misplaced hope that AI drivers will be better than humans seems to be ignoring that the systems are made by humans attempting to do something that is far more complex and error prone than humans just driving a car.

Look at the latest fiasco of the F-35 fighter program. They literally are unable to write the software to make that thing work as advertised, the complexity is obviously too great. The amount of money that has been thrown at this makes that spent on AI driving look a pittance - $1.4 T vs $8 B.

AlmightyCushion

cnocbui wrote: »

All this misplaced hope that AI drivers will be better than humans seems to be ignoring that the systems are made by humans attempting to do something that is far more complex and error prone than humans just driving a car.

Look at the latest fiasco of the F-35 fighter program. They literally are unable to write the software to make that thing work as advertised, the complexity is obviously too great. The amount of money that has been thrown at this makes that spent on AI driving look a pittance - $1.4 T vs $8 B.

You can say that about any task we apply AI to. Yet there are already tasks where AI can beat the best humans in the world at. Just look at AlphaGo and IBM's Watson. AlphaGo has beaten the best humans in the world at Go. Even a few years ago a lot of people thought that wouldn't happen for decades. IBM's Watson destroyed two of the best people to ever go on Jeopardy and that was 7 years ago, it would probably do a lot better now. Not to mention Watson has proven better at diagnosing things like cancers than humans are. AI might not beat the average human when it comes to something like driving right now but it will in time.

magicbastarder

cnocbui wrote: »

The amount of money that has been thrown at this makes that spent on AI driving look a pittance - $1.4 T vs $8 B.

your figures are *highly* misleading. one and a half trillion is the expected entire lifetime cost for the program - to 2070; of which $55bn is research and development.
so most of the money has not been spent, and when it is, it will have been spent on operations and support.

cnocbui

AlmightyCushion wrote: »

You can say that about any task we apply AI to. Yet there are already tasks where AI can beat the best humans in the world at. Just look at AlphaGo and IBM's Watson. AlphaGo has beaten the best humans in the world at Go. Even a few years ago a lot of people thought that wouldn't happen for decades. IBM's Watson destroyed two of the best people to ever go on Jeopardy and that was 7 years ago, it would probably do a lot better now. Not to mention Watson has proven better at diagnosing things like cancers than humans are. AI might not beat the average human when it comes to something like driving right now but it will in time.

Board games and driving cars are not in the same universe of problem nature, difficulty and variables. They are very neatly contained and constrained. It's like watching a baby take two doddering steps and proclaiming it will be a soloist ballet star next week. AlphaGo required 1202 processors and 176 GPUs



This is Watson:



Really practical and cheap to fit in a car.

The AI driving promoters are like stary-eyed fan boys. They have underestimated the difficulty of the problem and of human ability to codify the task and write software to perform it and they hugely underestimate the actual skill and performance of humans. It's like the F-35 manufacturers selling their expensive toy promising it will be able to do an amazing array of complex tasks and yet will be easier and cheaper to maintain than existing aircraft, whereas all that has turned out to be hype and lies.

One day car driving AI's may well match humans, but I think that is further away than estimated and I think it unethical to kill a lot of people to get there.

magicbastarder

comparing a computer beating a grand master chess champion is not the same as comparing a computer 'beating' a human at driving.
it'd be more instructive to compare a computer beating someone who has been playing chess for a year, and is reasonably competent at it. it's quite probable that a smartphone could do that. certainly, something which could easily fit in a car.

cnocbui

magicbastarder wrote: »

your figures are *highly* misleading. one and a half trillion is the expected entire lifetime cost for the program - to 2070; of which $55bn is research and development.
so most of the money has not been spent, and when it is, it will have been spent on operations and support.

It was the scale of the program and overall availability of funds I was trying to impart, the point being that even with unimaginable amounts of money at stake they still can't write the code they thought they could even with those financial resources..

magicbastarder

it was a clear false equivalence. you were (deliberately?) comparing apples with oranges.

why did you not quote how much is expected to be spent on self-driving cars over the next 50 years - R&D, purchase costs, and running costs? that is the figure you should be using if you're citing the comparable budget figures for the F35.

the actual R&D for the F35 is one thirtieth the amount you quoted. and not all of that is being spent on developing the computers which fly the planes.