Author Topic: EEVblog #1066 - Uber Autonomous Car Fatality - How?  (Read 59616 times)

0 Members and 1 Guest are viewing this topic.

Offline Fungus

  • Super Contributor
  • ***
  • Posts: 16562
  • Country: 00
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #225 on: March 25, 2018, 08:44:21 am »
Think how many objects the computer has to track, and predict, in that scenario...  it almost amounts to a denial of service attack...

Most of them will move out of the way if you just drive slowly through the crowd.

Anybody here ever cross the street in Rome? If you wait for traffic you'll be there all day. After a while you spot that the residents don't wait, they just cross. The traffic responds.

It's scary the first few times you do it but it definitely works.
 

Offline G7PSK

  • Super Contributor
  • ***
  • Posts: 3859
  • Country: gb
  • It is hot until proved not.
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #226 on: March 25, 2018, 09:10:42 am »
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...
Most (99.9%) human drivers have ZERO situational awareness and ZERO predictability skills. Most modern cars already have ESP which does the precise vehicle control for the driver when necessary.

I live in the UK and drive a VW which has ESP but it certainly does not mean extra sensory perception, it is "electronic skid protection" ie. anti lock brakes with a fancy name, also has a button to turn it off which got used for the first time recently with all the snow we had. Road safety would be enhanced greatly by two things phone which turn off automatically when inside a car and the removal of radios and music systems.
 

Online nctnico

  • Super Contributor
  • ***
  • Posts: 26757
  • Country: nl
    • NCT Developments
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #227 on: March 25, 2018, 11:04:33 am »
Quote from: nctnico
Most (99.9%) human drivers have ZERO situational awareness and ZERO predictability skills.
What do you guys smoke  there in Netherlands?
I have no idea, but again, they kill almost half as many people on the roads, so they're doing something right.
I'm not writing about the Netherlands but more in general. I've driven in many countries and it seems that most people just tail the car in front of them and don't look ahead at all. There just isn't any planning involved. It is logical too: doing the driving yourself is a necessary evil and people don't like to pay attention.

@james_s: the roads in the NL are usually in very good condition and well laid out to offer save driving conditions. The NL government is very active to reduce the number of car accident related fatalities but it isn't easy in one of the most densely populated countries in the world. Getting a driver's license means passing a theoretical and practical exam. Passing these exams requires some serious training (usually it takes someone 6 months to a year). Cars are expensive to own so people usually have the least number of cars they need. There are around 17 million people in the NL and around 8 million cars. Together those cars drove 188 billion kilometers (2016) so that makes an average of around 15k km per car.

@Fungus: driving in Italy is quite different indeed  :scared:

Edit: I'm convinced self driving cars will be safer overall compared to people driving their cars themselves. Trying to think about single scenarios where a self driving car may not work is rather moot because I see some human drivers also struggle with some situations. The basic rule is not to hit anything. I'm also convinced more self driving car related lethal accidents (*) will happen but when you are going to look at the accident statistics (the big picture) the self driving car will beat the human driver hands down.

* People do stupid things like crossing the street without looking. This happened to me too when I was a teenager on a dark-ish morning during the winter. Somehow I overlooked a car and it hit me when I crossed the street. I still don't get why I didn't see it. Perhaps the driver didn't switch the lights on. My father (strongly) blamed the hood on the jacket I was wearing so ever since I wear a cap during the winter.
« Last Edit: March 25, 2018, 12:44:43 pm by nctnico »
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 

Offline Marco

  • Super Contributor
  • ***
  • Posts: 6694
  • Country: nl
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #228 on: March 25, 2018, 04:53:10 pm »
Trying to think about single scenarios where a self driving car may not work is rather moot
It's more of a question where will they work. I think till we have human level AI the best bet is to use them as slightly more flexible trams, but outside of Japan I don't really see the point ... labour costs not really much of an issue for trams.
 

Online nctnico

  • Super Contributor
  • ***
  • Posts: 26757
  • Country: nl
    • NCT Developments
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #229 on: March 25, 2018, 05:19:41 pm »
Trying to think about single scenarios where a self driving car may not work is rather moot
It's more of a question where will they work.
From the experiments so far it seems the self driving cars work in cities and on highways. That is close to 100% of what cars are used for. I do wonder how well an EV handles ice & snow.
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 

Offline Marco

  • Super Contributor
  • ***
  • Posts: 6694
  • Country: nl
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #230 on: March 25, 2018, 06:19:43 pm »
From the experiments using cars driving in perfectly mapped environment, with teams actively engaged in surveying roadworks to make them avoid those, with human backup for when their programming breaks. Also they aren't double parking in stupid situations yet, which they will be if they actually start doing a taxi service.
 

Offline gearshredder

  • Contributor
  • Posts: 42
  • Country: us
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #231 on: March 25, 2018, 06:50:17 pm »
I do wonder how well an EV handles ice & snow.

Well, depends on if it's a DC or AC motor and it's controller. Modern AC induction motor and controller is going to have amazing anti-slip capabilities.
 

Online wraper

  • Supporter
  • ****
  • Posts: 16799
  • Country: lv
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #232 on: March 25, 2018, 08:18:02 pm »
I do wonder how well an EV handles ice & snow.



But seriously:

« Last Edit: March 25, 2018, 08:24:34 pm by wraper »
 

Online nctnico

  • Super Contributor
  • ***
  • Posts: 26757
  • Country: nl
    • NCT Developments
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #233 on: March 25, 2018, 08:47:46 pm »
From the experiments using cars driving in perfectly mapped environment, with teams actively engaged in surveying roadworks to make them avoid those
I strongly doubt they do mapping of roadworks because that would invalidate the experimental results. It would also mean that the autonomous car can't handle obstacles like broken down vehicles, accidents and delivery trucks which can happen at any time. Perhaps mapping roadworks was done in very early (more controlled) tests with trained drivers but I can't imagine they still do it with 'generic' drivers while the car is supposed to do all the work.
« Last Edit: March 25, 2018, 08:50:55 pm by nctnico »
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 

Online nctnico

  • Super Contributor
  • ***
  • Posts: 26757
  • Country: nl
    • NCT Developments
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #234 on: March 25, 2018, 08:49:21 pm »
I do wonder how well an EV handles ice & snow.
Well, depends on if it's a DC or AC motor and it's controller. Modern AC induction motor and controller is going to have amazing anti-slip capabilities.
Crap  :palm: User I/O error. I meant an autonomous car ofcourse!
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 

Offline coppice

  • Super Contributor
  • ***
  • Posts: 8605
  • Country: gb
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #235 on: March 25, 2018, 08:59:11 pm »
From the experiments using cars driving in perfectly mapped environment, with teams actively engaged in surveying roadworks to make them avoid those
I strongly doubt they do mapping of roadworks because that would invalidate the experimental results. It would also mean that the autonomous car can't handle obstacles like broken down vehicles, accidents and delivery trucks which can happen at any time. Perhaps mapping roadworks was done in very early (more controlled) tests with trained drivers but I can't imagine they still do it with 'generic' drivers while the car is supposed to do all the work.
Google's presentations emphasise how much they rely on their 11cm resolution map of the area, and correlation between the lidar input and the map, to work out exactly where the car is. I haven't seen a presentation where they address the topic of what happens when the world doesn't match the map. It they really had good solutions for this, surely they would be bragging about them.
 

Online nctnico

  • Super Contributor
  • ***
  • Posts: 26757
  • Country: nl
    • NCT Developments
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #236 on: March 25, 2018, 09:09:53 pm »
From the experiments using cars driving in perfectly mapped environment, with teams actively engaged in surveying roadworks to make them avoid those
I strongly doubt they do mapping of roadworks because that would invalidate the experimental results. It would also mean that the autonomous car can't handle obstacles like broken down vehicles, accidents and delivery trucks which can happen at any time. Perhaps mapping roadworks was done in very early (more controlled) tests with trained drivers but I can't imagine they still do it with 'generic' drivers while the car is supposed to do all the work.
Google's presentations emphasise how much they rely on their 11cm resolution map of the area, and correlation between the lidar input and the map, to work out exactly where the car is. I haven't seen a presentation where they address the topic of what happens when the world doesn't match the map. It they really had good solutions for this, surely they would be bragging about them.
That may be but there is a difference between comparing stationary objects like buildings and needing to map temporary things like road blocks. I think they need the accurate maps to do accurate positioning so the car doesn't drive on a lane in the wrong direction. Unfortunately accurate positioning is very hard to do and GPS really isn't up to the task.
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 

Offline mikerj

  • Super Contributor
  • ***
  • Posts: 3233
  • Country: gb
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #237 on: March 26, 2018, 12:01:16 pm »
I live in the UK and drive a VW which has ESP but it certainly does not mean extra sensory perception, it is "electronic skid protection" ie. anti lock brakes with a fancy name, also has a button to turn it off which got used for the first time recently with all the snow we had. Road safety would be enhanced greatly by two things phone which turn off automatically when inside a car and the removal of radios and music systems.

ESP is "Electronic Stability Program", and is far more than just Anti-lock brakes.  These systems detect loss of steering control by comparing the demanded yaw rate (steering angle input and speed) vs measured yaw (using a gyro), and applies braking to individual wheels to try to correct the discrepancy.
 

Offline G7PSK

  • Super Contributor
  • ***
  • Posts: 3859
  • Country: gb
  • It is hot until proved not.
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #238 on: March 26, 2018, 01:05:35 pm »
I live in the UK and drive a VW which has ESP but it certainly does not mean extra sensory perception, it is "electronic skid protection" ie. anti lock brakes with a fancy name, also has a button to turn it off which got used for the first time recently with all the snow we had. Road safety would be enhanced greatly by two things phone which turn off automatically when inside a car and the removal of radios and music systems.

ESP is "Electronic Stability Program", and is far more than just Anti-lock brakes.  These systems detect loss of steering control by comparing the demanded yaw rate (steering angle input and speed) vs measured yaw (using a gyro), and applies braking to individual wheels to try to correct the discrepancy.
And totally useless as soon as conditions get tricky which is why there is a button to turn it off in icy or muddy conditions with recommendation to do so in the hand book under such conditions. IE. its just more electronic clutter on the car and cannot in reality offer much if any protection or help to the driver. 
 

Online nctnico

  • Super Contributor
  • ***
  • Posts: 26757
  • Country: nl
    • NCT Developments
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #239 on: March 26, 2018, 01:53:51 pm »
It probably depends on the brand of the car as well. People I know are very happy with ESP and it did help them to keep control over the car on icy roads.
There are small lies, big lies and then there is what is on the screen of your oscilloscope.
 
The following users thanked this post: thm_w

Offline james_s

  • Super Contributor
  • ***
  • Posts: 21611
  • Country: us
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #240 on: March 26, 2018, 04:57:27 pm »
For the average person these systems would likely be hard pressed to do worse. Several years ago there was a snow storm and the hill I live on became covered in ice. An older gentleman got his Prius stuck against the curb trying to go down the hill so I helped get it un-stuck. He then proceeded to lock up all 4 wheels and slide sideways then backwards down the hill, uselessly trying to steer. I repeatedly yelled to let off the brakes so the wheels could turn but he didn't and the car just kept sliding completely out of control, miraculously not hitting anything else on the way down. Had he let off the brakes the car would likely have remained straight and controllable, but as far as the ABS system was concerned the car was stationary. Quite a few drivers have virtually no understanding of physics or automotive systems.
 

Offline Koen

  • Frequent Contributor
  • **
  • Posts: 502
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #241 on: March 26, 2018, 05:44:38 pm »
and cannot in reality offer much if any protection or help to the driver.
I'm very thankful for my car's ESP. It's absolutely brilliant.
 
The following users thanked this post: thm_w

Offline thm_w

  • Super Contributor
  • ***
  • Posts: 6278
  • Country: ca
  • Non-expert
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #242 on: March 26, 2018, 08:39:42 pm »
And totally useless as soon as conditions get tricky which is why there is a button to turn it off in icy or muddy conditions with recommendation to do so in the hand book under such conditions. IE. its just more electronic clutter on the car and cannot in reality offer much if any protection or help to the driver.

I think you've got that wrong, according to my manual:
Quote
The vehicle should be driven with the Vehicle Dynamic Control (VDC) system on for most driving conditions. If the vehicle is stuck in mud or snow, the VDC system reduces the traction motor output to reduce wheel spin. The traction motor speed will be reduced even if the accelerator is depressed to the floor. If maximum traction motor power is needed to free a stuck vehicle, turn the VDC system off

They only recommend turning it off IF you are stuck. Turning it off when its muddy or icy is completely counter to the additional safety and traction it will provide you. There is a reason its mandated to be in all newer cars, its useful.
Profile -> Modify profile -> Look and Layout ->  Don't show users' signatures
 

Offline SilverSolder

  • Super Contributor
  • ***
  • Posts: 6126
  • Country: 00
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #243 on: March 26, 2018, 09:37:59 pm »

[ESP is] totally useless as soon as conditions get tricky which is why there is a button to turn it off in icy or muddy conditions with recommendation to do so in the hand book under such conditions. IE. its just more electronic clutter on the car and cannot in reality offer much if any protection or help to the driver.

Once, on the motorway, I got the passenger side of the car too far over to the side, into the soft snow.  As if by magic, ESP applied the brakes on the opposite side of the car which had the effect of straigthening things up quickly.   That's the only time I've seen ESP activate, but it worked in that situation...
 

Offline janoc

  • Super Contributor
  • ***
  • Posts: 3781
  • Country: de
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #244 on: March 26, 2018, 09:54:09 pm »
Not sure whether this has been posted already, but it is a pretty good analysis of how these technologies work and what can realistically be expected (e.g. LIDAR may not have saved the day even if it was installed and was working):

https://medium.com/@rebane/could-ai-have-saved-the-cyclist-had-i-programmed-the-uber-car-6e899067fefe

 

Offline Rasz

  • Super Contributor
  • ***
  • Posts: 2616
  • Country: 00
    • My random blog.
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #245 on: March 26, 2018, 10:10:39 pm »
Not sure whether this has been posted already, but it is a pretty good analysis of how these technologies work and what can realistically be expected (e.g. LIDAR may not have saved the day even if it was installed and was working):

https://medium.com/@rebane/could-ai-have-saved-the-cyclist-had-i-programmed-the-uber-car-6e899067fefe

Dont know what background author has, but I stopped reading after few paragraphs of nonsense.
Lidar doesnt "think" and doesnt do classification, it just provides additional data points to classification algorithm, plus this dude keeps using dashcap footage as a proof :/
Who logs in to gdm? Not I, said the duck.
My fireplace is on fire, but in all the wrong places.
 

Offline G7PSK

  • Super Contributor
  • ***
  • Posts: 3859
  • Country: gb
  • It is hot until proved not.
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #246 on: March 27, 2018, 07:45:13 am »
Not sure whether this has been posted already, but it is a pretty good analysis of how these technologies work and what can realistically be expected (e.g. LIDAR may not have saved the day even if it was installed and was working):

https://medium.com/@rebane/could-ai-have-saved-the-cyclist-had-i-programmed-the-uber-car-6e899067fefe
Having read that article it looks like all pedestrians will need to be fitted with a radar transponder and a strobe light.
 

Offline janoc

  • Super Contributor
  • ***
  • Posts: 3781
  • Country: de
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #247 on: March 27, 2018, 10:18:18 am »
Not sure whether this has been posted already, but it is a pretty good analysis of how these technologies work and what can realistically be expected (e.g. LIDAR may not have saved the day even if it was installed and was working):

https://medium.com/@rebane/could-ai-have-saved-the-cyclist-had-i-programmed-the-uber-car-6e899067fefe

Dont know what background author has, but I stopped reading after few paragraphs of nonsense.
Lidar doesnt "think" and doesnt do classification, it just provides additional data points to classification algorithm, plus this dude keeps using dashcap footage as a proof :/


Next time read with comprehension, mate. He explicitly talks about the AI algorithms doing the "thinking" and that LIDAR only feeds the data in! Most of the text is about how the system needs to balance data from multiple sensors.

And re his background - if I am not mistaken, it is this dude:
https://www.researchgate.net/profile/Martin_Rebane

I would say he has more clue about what he is talking about than most, given that he is working directly in the field.  What is your background to judge the article as "nonsense"?
 

Offline janoc

  • Super Contributor
  • ***
  • Posts: 3781
  • Country: de
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #248 on: March 27, 2018, 10:45:08 am »
Having read that article it looks like all pedestrians will need to be fitted with a radar transponder and a strobe light.

Possibly.

But the fact is that the current tech is massively overrated and often ascribed almost magical capabilities. I don't work in autonomous vehicles but I can certainly relate to what he is saying about the LIDAR data processing - e.g. that LIDAR scans 10x per second doesn't automatically mean you have the data 10x per second!

I have been working on SLAM and point cloud fusion (mostly for 3D scanning and virtual reality navigation - the algorithms used are very much the same) and this stuff requires enormous computing power, especially when you want dense and accurate point clouds, otherwise you can literally hide an elephant (or a cyclist in this case!) in the holes in the data. E.g. merging a measurement campaign worth of high-res scans from a laser scanner (which is basically slower but much more accurate LIDAR) can take a few hours on a PC, even with a GPU acceleration.

Now in a car, even at 10Hz, that's 0.1s * 60km/h (16.7m/s) the car was doing = ~1.7m traveled between every scan where it needs to rely on other sensors only. If the car is fusing the data slower (which is likely due to the complexity of the process), you do the math. E.g. at 1s effective data rate you have almost 17m traveled between the map updates. Then you typically need two or more "frames" for the AI system to evaluate the new data as an obstacle (as opposed to e.g. spurious reflection or some transient noise - e.g. a car moving nearby). So, in the worst case, during those 2-3s the car is effectively driving "blind" (or rather - with outdated information) as far as LIDAR is concerned - that's some 50m traveled! You can literally have an elephant step in front of the car during that time and the LIDAR-based system wouldn't see it.

Of course, a real car would likely do this faster and also have other systems covering that "blind" time that could prevent the collision as well (and the author actually says that explicitly) but the point is that LIDAR is far from the miracle it is often "sold" as.

« Last Edit: March 27, 2018, 10:53:39 am by janoc »
 

Offline SilverSolder

  • Super Contributor
  • ***
  • Posts: 6126
  • Country: 00
Re: EEVblog #1066 - Uber Autonomous Car Fatality - How?
« Reply #249 on: March 27, 2018, 11:34:07 am »
But the fact is that the current tech is massively overrated and often ascribed almost magical capabilities.

Wouldn't some of the "lower tech" collision avoidance technology (the type you can already get on production cars) have noticed the bicyclist at all, and at least have started to brake?
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf