Author Topic: EEVblog #1066 - Uber Autonomous Car Fatality - How?  (Read 59598 times)

0 Members and 2 Guests are viewing this topic.

Offline Brumby

  • Supporter
  • ****
  • Posts: 12288
  • Country: au
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #175 on: March 24, 2018, 03:27:20 am »
The whole way the 'interior' video is cut makes me think Uber is hiding something,
The cut may not be for that reason.  It may have been to prevent further stress on the safety driver.

a) Is somebody forcing them to watch it in a loop?

b) Having that face posted all over the Internet isn't stressful?

I suspect that the only reason they posted the interior shot at all was to show that the "safety driver" wasn't watching the road. What other reason is there for releasing it?

We could go around and around in circles over this - so I'm not going to discuss the motives in releasing of the interior video or when it was cut.


What I AM interested in addressing is the actual, defined, specific role of the "safety driver".

There have been FAR TOO MANY suggestions that this "safety driver" should have prevented this accident from happening and because they didn't, they are a contributor to fault.  Such opinions are, in my opinion (and the opinion of some others) absolutely flawed, completely unfair and totally wrong.  It is completely unrealistic to expect someone who is disengaged from the active role of driving to have been able to respond in time...
... would take a job babysitting an autonomous car? It has to be one of the most boring, soul destroying jobs around. It is implausible that whoever you put behind that wheel, with nothing much to do all day but look ahead, is going to maintain full attention hour after hour.

So ... if that be the case, then what IS the role of the "safety driver"?


In response to that question, I would like to ask the following - noting that NOBODY has raised this question so far:  As we have determined, the car DID STOP at some point - but why? 
 - Was it because an impact sensor triggered that response? 
 - Was it because a delay in the sensor processing had caught up and commanded the vehicle to stop
 -OR-
 - Was it because the vehicle continued driving as if nothing had happened and the safety driver took control and brought it to a stop?

Considering the vehicle did not take ANY identifiable action (from the video supplied) who's to say that it was ever going to stop because of the collision?

This, I believe, is a quite reasonable expectation for a safety driver and if that is the case, they did exactly what was expected of them, preventing any further danger.
« Last Edit: March 24, 2018, 03:35:01 am by Brumby »
 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 16560
  • Country: 00
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #176 on: March 24, 2018, 03:52:59 am »
So ... if that be the case, then what IS the role of the "safety driver"?

Scapegoat?  :popcorn:

(and that's why they employ felons)
 

Offline Rasz

  • Super Contributor
  • ***
  • Posts: 2616
  • Country: 00
    • My random blog.
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #177 on: March 24, 2018, 07:02:05 am »
It probably wouldn't have mattered due to that extra reaction time by the driver but I agree, the headlights look poorly adjusted. You can't even see the guys feet

we have been over this, nothing wrong with car lights, nothing wrong with road lighting, you are basing your opinion on a bad $30 dashcam mounted by company hiring criminals to drive people around.
what the road actually looks like to humanoids: https://twitter.com/thekaufaz/status/976686336877871104


So ... if that be the case, then what IS the role of the "safety driver"?

rubber stamped bureaucratic requirement
Who logs in to gdm? Not I, said the duck.
My fireplace is on fire, but in all the wrong places.
 

Online Marco

  • Super Contributor
  • ***
  • Posts: 6694
  • Country: nl
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #178 on: March 24, 2018, 08:44:53 am »
So ... if that be the case, then what IS the role of the "safety driver"?

AFAICS risk minimization (preventing run away due to bugs, ramming firetrucks, killing pedestrians) and handling edge cases the autopilot can't handle.

You could argue that they can't always handle the risks when low latency intervention is necessary, but that's no reason for Uber not to make them try their best. For that they need strict vigilance control systems. Not the pussy stuff Tesla&Co put in their cars, but ones which start being annoying immediately. Might give false positives when the driver scratches his nose or something, but false positives aren't really a problem.

If Uber doesn't want to make partial low latency accident prevention part of their role then someone needs to make them. Making the safety driver more comfortable isn't worth lives, there will always be takers for that job regardless. There's worse jobs, far worse.
 

Offline TheDane

  • Regular Contributor
  • *
  • Posts: 209
  • Country: dk
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #179 on: March 24, 2018, 09:29:15 am »


If the driver can't (or won't) use feet to break the vehicle - it should be mandatory to being able to use 'only' hands, and the switch(ES) should be placed accordingly.
Naturally it costs money, and takes a brain and time to implement.
Perhaps the people in charge should be the ones doing the walk-in-front-of-car-when-testing to ensure proper care has been taken  :o

Maybe also a pressure switch under the drivers ass, so a person jumping in the seat will engage the safety system (or, at least arm it further). I know this is already mandatory on lawnmovers - those moving blades have cut people into small pieces because they tipped over a slope and the machine fell upon them.
 

Offline JoeN

  • Frequent Contributor
  • **
  • Posts: 991
  • Country: us
  • We Buy Trannies By The Truckload
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #180 on: March 24, 2018, 09:44:26 am »
Do these self-driving cars record and store all sensor data over reasonable timeframes so the raw data can be played back against the software to see where the algorithms went wrong and to prove or disprove the quality of the data collection capabilities of the sensors themselves?  This is a teachable moment, I hope they have all that data.
Have You Been Triggered Today?
 

Offline Brumby

  • Supporter
  • ****
  • Posts: 12288
  • Country: au
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #181 on: March 24, 2018, 10:32:43 am »
So do I.
 

Offline Towger

  • Super Contributor
  • ***
  • Posts: 1645
  • Country: ie
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #182 on: March 24, 2018, 10:49:11 am »
LIDAR is useless in the real world, for several years I have been wondering why Google etc rely on it so much.  Maybe useless is too strong a word, but it does not work in the rain.  Which is why all the testing is done in very dry areas.


There was recent interview on Embedded FM with a firmware developer (day job) of  LIDAR systems.  I'll try and add the link when I get to a proper computer.  It does not appear to be in any show notes.


I also have the say the released video looks doctored.  Way to dark.  Street lighting?  A basic dash cam is better.  An XC90 is a highend luxury car with HID lighting, probably self levelling.
 

Offline BrianHG

  • Super Contributor
  • ***
  • Posts: 7660
  • Country: ca
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #183 on: March 24, 2018, 11:59:08 am »
LIDAR is useless in the real world, for several years I
That's just crummy single beam laser scanning lidar.
Newer 3D lidar is coming which scans the 3D environment in front of you like a high speed 2D camera.  But, current costs and owners of the patents make it too prohibitive for today, it's cheaper just to insure up and gamble with existing tech.
« Last Edit: March 24, 2018, 12:00:55 pm by BrianHG »
 

Offline BrianHG

  • Super Contributor
  • ***
  • Posts: 7660
  • Country: ca
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #184 on: March 24, 2018, 12:10:48 pm »
As for all the comments and arguments here, forget the BS dashcam, which may have been deliberately crapped up, until the accident scene is replicated by forensics, IE an equally dressed pedestrian, in the same place walking at the same speed.  With an equivalent car & headlights, traveling at the same speed as well, same location, and a real human eye-quality camera + the same camera from the self driving car's AI are re-filmed properly without heavy compressing by a investigating third party with no ties to Uber or any self driving car firms, absolutely everything observed to date, and every other driver doing their own filming on the same road posting it on youtube only serves to spread BS and we will never be sure of anything about the true visibility from that accident.

If proper third party replication isn't done, this accident will get buried away with time, which is what I expect to happen.
« Last Edit: March 24, 2018, 12:20:45 pm by BrianHG »
 
The following users thanked this post: orin

Offline apis

  • Super Contributor
  • ***
  • Posts: 1667
  • Country: se
  • Hobbyist
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #185 on: March 24, 2018, 12:15:19 pm »
Do these self-driving cars record and store all sensor data over reasonable timeframes so the raw data can be played back against the software to see where the algorithms went wrong and to prove or disprove the quality of the data collection capabilities of the sensors themselves?  This is a teachable moment, I hope they have all that data.
Yes, that's the main task of the test cars: to collect data.

LIDAR is useless in the real world, for several years I have been wondering why Google etc rely on it so much.  Maybe useless is too strong a word, but it does not work in the rain.  Which is why all the testing is done in very dry areas.
Lidar is the best sensor at the moment without competition. It gives you accurate 3D data which is easier to analyse and interpret and works well regardless of lighting conditions. All types of sensors have problems. Vehicle radar have very poor resolution. Cameras (eyes) don't work in the dark, in fog and works poorly in heavy rain and snow. Lidar works poorly in heavy rain and snow and have limited range. But if you combine data from many different types of sensors they complement each other. So combining many types of sensors is the best option from a technical and safety perspective.

The problem with lidar is that it's expensive, which is why many companies want to get rid of it. After all, humans manage with only two cameras on a stick so it should be possible for a robot car to do the same... in theory. But computer vision algoritms aren't reliable enough yet, so you still really need lidar today. Anyone who's played around with computer vision should/would have realised that imho, and sadly it's evident from the tesla decapitation experiment and now possibly this uber experiment if nothing else.

Waymo is already driving cars completely autonomously (no human in the car) so clearly their lidar equipped cars are very useful in the real world. Just because you can't use it in every situation and weather condition doesn't mean it's useless in the real world. Like everything else the tech will improve gradually and be able to handle more and more situations and locations.

In the beginning we will probably see them used in dry climates with good road infrastructure working as taxis, busses or transports. I.e. where the vehicles operate on well defined routes or within a well defined area and you can have a command central that can quickly handle any unexpected situation and send out a tech if necessary. And as long as the safety statistics for the autonomous cars are better than the average human driver that is a win for everyone.

I also have the say the released video looks doctored.  Way to dark.  Street lighting?  A basic dash cam is better.  An XC90 is a highend luxury car with HID lighting, probably self levelling.
Yes, it is a well lit street according to people who live there.
« Last Edit: March 24, 2018, 12:21:33 pm by apis »
 

Offline Uwe Quast

  • Newbie
  • Posts: 2
  • Country: de
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #186 on: March 24, 2018, 12:25:05 pm »
Viewing the last moment before the crash, it looks to me as like this was recorded from a camera on roof level.
It's a bit strange  to see the white shoes (the lowest parts) suddenly appearing in the middle of the street, almost as if those yellowish lamp-lights drastically reduced the dynamic range of the camera as long as they where visible.
As a german car driver this dark field in the upper middle of the headlight illumination field looks to me like a bad headlight alignment.
 

Offline apis

  • Super Contributor
  • ***
  • Posts: 1667
  • Country: se
  • Hobbyist
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #187 on: March 24, 2018, 12:34:13 pm »
I'm just speculating now, but assuming they have several cameras with different exposure level (to get better dynamic range) this might be footage from the camera with the shortest exposure.
 

Offline SilverSolder

  • Super Contributor
  • ***
  • Posts: 6126
  • Country: 00
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #188 on: March 24, 2018, 01:09:29 pm »
FYI  Both the victim and Uber driver were female. 

Accepted, no disrespect to the women involved intended.

Who will be the lead in the investigation - I hope it is not the police department.

Apparently, the NTSB (National Transportation Safety Board), the organisation that normally investigates air accidents, will investigate this incident.
Those people conduct the highest level and most professional investigations...   but don't expect quick results.
 

Offline SilverSolder

  • Super Contributor
  • ***
  • Posts: 6126
  • Country: 00
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #189 on: March 24, 2018, 01:16:53 pm »
Who but the desperate would take a job babysitting an autonomous car? It has to be one of the most boring, soul destroying jobs around. It is implausible that whoever you put behind that wheel, with nothing much to do all day but look ahead, is going to maintain full attention hour after hour.

There are people that do boring, soul destroying jobs...   for example, an electronics engineer trying to eke the last bit of performance out of an analog circuit...  so you are right, most people are probably not suitable for babysitting an autonomous car, just like most people are not interested in electronics?

The problem here is might be that Uber did not take the babysitting job as seriously as they perhaps could have, so ended up hiring someone who was not suitably resistant to boredom.
 

Offline Decoman

  • Regular Contributor
  • *
  • Posts: 161
  • Country: no
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #190 on: March 24, 2018, 01:27:46 pm »
I wonder if *maybe* there is some unethical testing going on with autonomous cars.

I remember that there was this local story years ago in the news, that told about how a paramedic supposedly decided to go against the testing regime that was in progress with the use of a drug in ambulances, for being injected into people with cardiac arrest (as I remember it), and so, the paramedic when performing life saving tasks, was supposed to simply open an envelope that they carried with them, and there it would state, to issue the drug this once, or not. Obviously, the idea was to get some statistical data on the efficiency on the drug, instead of doing the best to save the patient.

So, this makes me wonder if maybe they test autonomous cars, with limited capabilities to see what works best.

I have also been wondering myself, if maybe manufacturers of vaccines are able to develop regional variants, that in a similar way, can be used for testing, which would be unethical if being experimental, if people taking such a drug (which might not exist) in good faith of it working as expected.
 

Online wraper

  • Supporter
  • ****
  • Posts: 16794
  • Country: lv
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #191 on: March 24, 2018, 01:29:34 pm »
The problem with lidar is that it's expensive, which is why many companies want to get rid of it. After all, humans manage with only two cameras on a stick so it should be possible for a robot car to do the same... in theory. But computer vision algoritms aren't reliable enough yet, so you still really need lidar today. Anyone who's played around with computer vision should/would have realised that imho, and sadly it's evident from the tesla decapitation experiment and now possibly this uber experiment if nothing else.
FWIW, there are around 250 decapitation deaths under the truck in the US annually. And more than 1000 people seriously injured. All because in US they don't use underride side guards unlike in Europe.
 

Offline Decoman

  • Regular Contributor
  • *
  • Posts: 161
  • Country: no
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #192 on: March 24, 2018, 01:37:13 pm »
Btw, I think that if the passenger on autonomous cars in testing is really supposed to react and at best try prevent accidents by interfere with the car's movement, then presumably it would make the best sense that such an individual would have the experience of that of those that teach other people to drive, which afaik is able to slow down the vehicle with their secondary set of pedals found in such cars at driving schools. I wonder if such autonomous cars are compatible with such type of interaction with a human being being able to slowing down the speed on demand.
 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 16560
  • Country: 00
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #193 on: March 24, 2018, 01:51:02 pm »
LIDAR is useless in the real world, for several years I have been wondering why Google etc rely on it so much.  Maybe useless is too strong a word, but it does not work in the rain.  Which is why all the testing is done in very dry areas.
Lidar is the best sensor at the moment without competition. It gives you accurate 3D data which is easier to analyse and interpret and works well regardless of lighting conditions. All types of sensors have problems.

I asked before but nobody replied so I'll ask again:

If there's several LIDAR scanners in the same place, do they interfere with each other?

 

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 16560
  • Country: 00
 

Offline nixxon

  • Frequent Contributor
  • **
  • Posts: 338
  • Country: no
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #195 on: March 24, 2018, 03:37:23 pm »
I have said many, many times that the technology is too premature to be viable. I live in a very high traffic area and as I drive down the freeway with 5 lanes full of traffic at road speeds (60-70 mph) and marvel at how all of the vehicles merge into other lanes and the varying speeds of different vehicles do not crash into each other because humans can anticipate and adjust. Even when there are traffic cones and construction, the number of fender benders, compared to how many vehicles on a given street is minimal. The statistics tell the story of vehicle fatalities, and, given the sheer number of vehicles on the road, the fatalities, IMHO are extremely low.

I dare say that many may not grasp this concept because in city centers like where I live, the traffic is horrifying for the unskilled driver in such conditions. The roads, streets and highways in the US are huge compared to most countries, so I can't even imagine how autonomous driving could even work in countries where the roads and streets are tiny.

Although this film has been around a long time, when an autonomous car can do this, then it might be ready for public release.

If you are a bit squimish about fast driving, do not watch this film.



Admittedly, the streets were fairly empty in this film.

HAHA! Is this some 1960's footage taken from a 32 HP Citroën 2CV?
 

Offline jonovid

  • Super Contributor
  • ***
  • Posts: 1425
  • Country: au
    • JONOVID
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #196 on: March 24, 2018, 03:49:43 pm »
smart roads will save autonomous vehicles in my opinion.
this is NOT about solar roadways to melt snow  :bullshit: .but a smart road is about technology that works!  :-+
magnetic peg tracking or magnetic lane markers are to help autonomous vehicles find the lanes in fog
or snow. bad weather. the magnetic pegs are set in to the road surface at a known spacing.
passive infrared sensors for the monitoring of pedestrians, animals , cyclists. useing street pole lights.
smart road street poles will not only support street lights, but infrared sensors and tracking cameras.
only when autonomous vehicles are married to smart roads will vehicles be truly autonomous, & truly safe!
at the present time many smart road proposals are to incorporate solar panels underneath the road surface,  |O
in my opinion electricity generation is not what smart roads are about! but its about supporting autonomous vehicles
secondly smart roads are about monitoring & conveying information to the users of it. wifi , dot matrix display signs
on street poles and or embedded in the road surface where it makes sense to do so.  lane markers.
« Last Edit: March 24, 2018, 03:59:12 pm by jonovid »
Hobbyist with a basic knowledge of electronics
 

Offline nixxon

  • Frequent Contributor
  • **
  • Posts: 338
  • Country: no
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #197 on: March 24, 2018, 04:06:24 pm »
How did the Uber autonomous car fatality in Tempe Arizona happen?

A relatively stationary object of significant size and mass was positioned in the path (lane) of the Uber car traveling at a modest 65 km/h. The Uber car failed to detect and brake before hitting the obstacle.

The LIDAR system did not seem to work as supposed to.
The radar system did not seem to work as supposed to.

Or maybe the brakes didn't work?

If the car was driven by a prudent human being, he or she would have applied the high beams of the headlights (night time with no oncoming traffic), and probably would have visually detected the obstacle at a distance of 100-150 meter. The driver would probably have had no problem stopping before hitting the obstacle, even with a 1 second reaction time (18 meters rolling at 65 km/h) before brakes were fully applied.

On top of that, a real person could easily turn the vehichle to the side of the obstacle, even when applying full braking power (if driving a car less than 15 years old with ABS).
« Last Edit: March 24, 2018, 05:51:02 pm by nixxon »
 

Online Marco

  • Super Contributor
  • ***
  • Posts: 6694
  • Country: nl
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #198 on: March 24, 2018, 05:37:58 pm »
If there's several LIDAR scanners in the same place, do they interfere with each other?
With the rotating LIDARs it's extremely unlikely. The receiver almost certainly has a lens and will only "look" at a narrow view, where the beam is currently being transmitted. Even when two LIDARs do match up, it will be for a small vertical section of the frame, the section would also move between frames. Easy to filter.
 

Offline SilverSolder

  • Super Contributor
  • ***
  • Posts: 6126
  • Country: 00
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #199 on: March 24, 2018, 05:51:42 pm »

It will hopefully be a long time before autonomous vehicles get the same level of situational awareness and precise vehicle control of a good human driver...

« Last Edit: March 24, 2018, 05:56:32 pm by SilverSolder »
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf