Author Topic: EEVblog #1088 - Uber Autonomous Car Accident Report  (Read 5307 times)

0 Members and 1 Guest are viewing this topic.

Offline EEVblog

  • Administrator
  • *****
  • Posts: 27005
  • Country: au
    • EEVblog
EEVblog #1088 - Uber Autonomous Car Accident Report
« on: May 25, 2018, 02:29:32 pm »
The NTSB today released the report into the fatal Uber Autonomous car accident.
TLDR;
The RADAR, LIDAR, and cameras DID detect and classify pedestrian bicycle correctly.
The system DID determine that emergency braking was required.
But Uber disabled the systems emergency braking feature in autonomous mode.
Uber also disabled Volvo's inbuilt safety systems.
There is also no system to alert the driver that the system detected an emergency breaking scenario.



« Last Edit: May 25, 2018, 03:07:56 pm by EEVblog »
 

Online maginnovision

  • Frequent Contributor
  • **
  • Posts: 271
  • Country: us
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #1 on: May 25, 2018, 02:58:23 pm »
Just adding my comment from YouTube:

Quote
That math is bad, .77g is lateral acceleration. So you've got forward AND lateral acceleration working on the tires together. Just emergency braking straight ahead the vehicle should've stopped, or VERY nearly stopped. Braking from 60mph it'll stop in about 35 meters, just under. Not to mention the noise would've alerted the pedestrian and THEY could've taken action(In theory, drugs are bad.)

Somebody should be held responsible for this legally so the family can at least get closure(not just money). The family settled but AZ can still charge whoever they find deserves it.
 

Offline ez24

  • Super Contributor
  • ***
  • Posts: 3092
  • Country: us
  • L.D.A.
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #2 on: May 25, 2018, 03:07:45 pm »
FYI  Bicycle vs person

Toyota's current safety system, TSS-P, has pedestrian braking.  In the next version, TSS-P2.0, starting in 2019, has pedestrian AND bicycle braking.   Are bicycles harder to detect than people?  I have no idea.
« Last Edit: May 25, 2018, 03:27:38 pm by ez24 »
YouTube and Website Electronic Resources ------>  http://www.eevblog.com/forum/other-blog-specific/a/msg1341166/#msg1341166
 

Offline Decoman

  • Regular Contributor
  • *
  • Posts: 156
  • Country: no
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #3 on: May 25, 2018, 03:12:24 pm »
I wonder if any reflective material on the bike or person at night time might confuse the autonomous car software.
 

Offline Rasz

  • Super Contributor
  • ***
  • Posts: 2110
  • Country: 00
    • My random blog.
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #4 on: May 25, 2018, 03:40:12 pm »
I wonder if any reflective material on the bike or person at night time might confuse the autonomous car software.

Alien spaceship flying above might confuse autonomous car software, and I might get pregnant.
The fact is car DID detect pedestrian just fine, but emergency breaking was deemed too inconvenient at design stage.
Who logs in to gdm? Not I, said the duck.
My fireplace is on fire, but in all the wrong places.
 

Offline Brumby

  • Supporter
  • ****
  • Posts: 7699
  • Country: au
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #5 on: May 25, 2018, 04:26:21 pm »
After reading that report, I have no issue with the operator.  None at all.  She was doing what she had been instructed to do and the fact that she did react is a credit to her.

The fundamental issue with the "operator intervention" requirement for emergency braking is that the operator had been instructed to keep tabs on the autonomous systems monitor and make annotations for later review.  Having been directed away from watching the road, their situational awareness had been severely compromised.

Regaining that awareness was a process that would have required time.  From lifting one's head up from the monitor, the first stage is for their eyes to adjust to the lighting conditions and to focus.  Then the scene is "taken in" - the human process of understanding the image presented by the eyes.  Only then is it possible for the human to identify issues and be able to take action ... and that process also takes time.

All of this may only take a second or two, but that's all the time that was needed for this to end tragically.

Uber has much to answer for, IMO.
1. By allocating the system monitor task to the operator, they have compromised that person's ability to act in an emergency situation.
2. By not providing a warning to the operator of a potential problem, Uber denied the operator the opportunity to look up and gain situational awareness before the critical period.
3. By disabling the manufacturer's automatic braking system, Uber added a significant element of risk, especially considering their own software's ability to do so had been disabled.  This is, perhaps, the single most bizarre action since I would imagine it added NO value to the data collected.  If the Volvo's braking system engaged, then I would love to see that in the data and compare it against the Uber system's assessments.  This is just inexcusable.

The technology would appear to be capable ... it's the idiots running it that are making bad choices.
 

Offline G7PSK

  • Super Contributor
  • ***
  • Posts: 3590
  • Country: gb
  • It is hot until proved not.
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #6 on: May 25, 2018, 07:06:52 pm »
Why did uber not integrate the Volvo systems into its own in order to avoid collisions. As it is the car was set up to have a collision. It may well be worse than that the reason the driver was looking down was the diagnostic system bleeped due to detecting an unknown ahead, so she does what she is told to do, Takes her eyes off the road. Looks like Uber decided to cheap out, dis able the Volvos system rather than tie them in and not have a co driver to keep an eye on diagnostic systems etc. A fatality was inevitable at some point. There is a manager level person somewhere in Uber who should be doing time inside.
 

Offline orion242

  • Supporter
  • ****
  • Posts: 514
  • Country: us
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #7 on: May 25, 2018, 07:47:23 pm »
$10k question still lingers imo.

Who is liable and will they press charges?  If this was a human driver, someone would be in deep crap.  Eyes off the road and fatality....

Should be interesting to see the reports from the recent tesla autopilot crashes when they come out as well.
« Last Edit: May 25, 2018, 07:50:14 pm by orion242 »
 

Offline Tom George

  • Newbie
  • Posts: 1
  • Country: au
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #8 on: May 25, 2018, 07:55:55 pm »
Hi,
FROM YouTube reply..

If the "SYSTEM" relies on the operator to intervene, why doesn't  the operator just drive the car all the time and for safety sake forget about "Autonomous" systems.

The only way an Autonomous system can work is if there is an Autonomous only road, not lane but road. That is obstacle/pedestrian free?

The fact that the Volvo system was disabled is puzzling, the system must work otherwise Volvo would not have fitted it to their cars.
Why didn't the UBER system use it as a secondary/primary detection system.

Question, WHY didn't the pedestrian see the car(Headlights)? There didn't appear to be any other vehicles or obstacles in the way to obstruct her view.

Tom.. :)
« Last Edit: May 25, 2018, 08:08:33 pm by Tom George »
 

Online glarsson

  • Frequent Contributor
  • **
  • Posts: 682
  • Country: se
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #9 on: May 25, 2018, 08:02:48 pm »
Question, WHY didn't the pedestrian see the car(Headlights)?
What do you expect from someone high on methamphetamine and cannabis?
 

Online max_torque

  • Frequent Contributor
  • **
  • Posts: 879
  • Country: gb
    • bitdynamics
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #10 on: May 25, 2018, 09:39:38 pm »
The ultimate "fault" is clearly with the unfortunate victim of this crash (high on drugs, not paying attention, and crossing where specifically prohibited)

BUT,

The point of autonomous systems is to try to mitigate even such extreme circumstances (drugged up pedestrian wanders into the road) and so it's right to look at the failures in the automation.

Unfortunately, the leading edge of the bonnet is a very stiff structure, and despite "showing little damage" can easily result in fatal head injuries (especially for shorter victims) at a surprisingly low speed.

  I should point out that the EEVblog braking (not breaking!) analysis uses a figure of 0.77g, which is actually very low for a typical modern car on a dry road surface. The typical figure for fully developed threshold braking would be around 1.1g and can be up to 1.4g for transient events (before the mass transfer due to the deccel results in less optimum normal load distribution).

If the system had effectively "lifted off" at the 6.5 sec mark (typically around 0.05g) and then swapped to full retard at 1.3 secs, remembering it would have had plenty of time to already "pre arm" the brake system (ie applied a gentle pre-application pressure to push the brake pads out to the discs and stiffen the system so that full braking can be developed almost immediately (~75ms)

It's also worth pointing out that Davids analysis is imo, wrong, as he hasn't considered the fact that we have a time to impact measured at a constant speed, and hence in the distance domain, as you decelerate it takes "Longer" to travel a distance, and hence you then have more time to brake (and hence a 2.55 sec stopping "time" is available!

 We know there was 1.3 seconds advanced warning and that the car did not slow before hitting the victim, and so the critical distance was 1.3 sec x 43mph (19.2 m/s) = 24.98m prior to the impact point.  (we'll call it 25m). if we take a worst cased and delay the application of braking by 100ms (a long time for modern ABS system) we have    (25 - ( 0.1 * 19.2)) = 23m left to stop, under fully developed threshold braking.  And at 1 g, it takes just 18m (and 1.95 sec to stop completely from 43mph) ie, we stop 7 meters (more than one whole car length) away from the pedestrian!

In fact, with 23m to stop from 43mph, we can brake at your original -7.5m/s level and still juuuust stop before we hit them!

(this phenomena also explains why, when emergency braking a car, you need to stop hard as early as possible, to give you the most amount of time in which to stop.  Most human drivers crash into stuff that they could have in fact actually avoided if they had threshold braked immediately, rather than just got gently on the brakes to start with, and then only braked hard when they realised they needed too, but by then it's too late.....   Modern braking systems include a "brake assist" system for this very reason, that actually applies additional pedal force immediately on a rapid lift of the throttle and detection of brake pedal movement. Some drivers angrily complain this makes the car feel like it over reacted, however, what they miss is just how angry and annoyed they would be to spend the next few months in hospital (or on a charge of manslaughter) after actually crashing because they didn't brake hard enough.......)


 
The following users thanked this post: WN1X, Jacon

Online max_torque

  • Frequent Contributor
  • **
  • Posts: 879
  • Country: gb
    • bitdynamics
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #11 on: May 25, 2018, 09:47:43 pm »
XC-90, dry tarmac, 62mph to zero in 36m



Equivalent to 1.08g average decel. (not clear if that distance shown in that video includes the reaction and application time, clearly the reaction time should be very short as the driver is ready and armed to stop, but unless they are left foot braking, the application time should be valid.  We can't tell however if it is 36m from the point the vehicle starts to slow or 36m from where the driver reacts.

Still, shows how good at stopping modern brakes and tyres really are.  (most modern hi-silica compound tyres now generate around 0.85g even on wet roads)
 

Offline G7PSK

  • Super Contributor
  • ***
  • Posts: 3590
  • Country: gb
  • It is hot until proved not.
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #12 on: May 25, 2018, 10:11:21 pm »
Question, WHY didn't the pedestrian see the car(Headlights)?
What do you expect from someone high on methamphetamine and cannabis?

Where does it say that the cyclist was high on drugs. And did the car have its lih=ghts on at the time of the collision.
 

Online glarsson

  • Frequent Contributor
  • **
  • Posts: 682
  • Country: se
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #13 on: May 25, 2018, 10:19:55 pm »
Where does it say that the cyclist was high on drugs.
In the preliminary NTSB report. Second section from the end.

https://www.ntsb.gov/investigations/AccidentReports/Reports/HWY18MH010-prelim.pdf


 

Offline GerryBags

  • Frequent Contributor
  • **
  • Posts: 334
  • Country: gb
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #14 on: May 25, 2018, 10:31:01 pm »
It seems crazy to me that there is not much more regulation of companies taking safety risks on public roads while they are actively trying to reduce the number of jobs for people. In a world of full employment it makes sense to free up people from menial tasks, but we don't have full employment anywhere and driving is NOT a menial task. A study that came out last year found that the levels of concentration required for driving were far higher, and used much more of the brain's capacity, than previously thought.

 

Offline apis

  • Frequent Contributor
  • **
  • Posts: 722
  • Country: se
  • Hobbyist
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #15 on: May 25, 2018, 10:32:09 pm »
That underexposed dash cam video Uber (and the police!) released was pure misdirection. As many other youtubers have demonstrated; the lighting conditions were perfectly normal at the time of the accident. All sensors seem to have detected the pedestrian/bike. We were meant to believe the safety driver was looking down at her phone but instead it turns out she was looking at the diagnostics screen as she had been instructed to do, and whoever released the video must have known that. Clearly they were trying to put all the blame the safety driver! Unbelievable.

Apparently they used to have two safety drivers in every car; one driving and one looking at the diagnostics, tagging 'interesting' events. Recently they had switched to only one safety driver, presumably to get more testing done in time for a VIP demo.

Even if they say they completely relied on the safety drivers to brake (which is pretty jaw dropping, isn't this supposed to be self driving?), they could at least have the car give some sort of audible alert to warn the operators they might have to brake. That is just inexcusable.
 

Offline orion242

  • Supporter
  • ****
  • Posts: 514
  • Country: us
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #16 on: May 25, 2018, 10:37:03 pm »
The ultimate "fault" is clearly with the unfortunate victim of this crash (high on drugs, not paying attention, and crossing where specifically prohibited)

This could have easily been a child chasing after a ball and running out in to the road.

So a druggie getting mowed down by an distracted driver is ok?  Does this change if its a child then?
 
The following users thanked this post: Decoman

Offline G7PSK

  • Super Contributor
  • ***
  • Posts: 3590
  • Country: gb
  • It is hot until proved not.
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #17 on: May 25, 2018, 11:07:19 pm »
Where does it say that the cyclist was high on drugs.
In the preliminary NTSB report. Second section from the end.

https://www.ntsb.gov/investigations/AccidentReports/Reports/HWY18MH010-prelim.pdf
It only says the tests showed positive it does not state the pedestrian was off their head or even just a bit intoxicated. Tests are so sensitive now that just being exposed to fumes will show on the person.
 

Offline orion242

  • Supporter
  • ****
  • Posts: 514
  • Country: us
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #18 on: May 25, 2018, 11:46:40 pm »
I fail to see why it really matters.

This is a case that any reasonable alert driver would have had plenty of time to stop or maneuver preventing this accident.  Had this person walked out between parked cars feet from a moving car, ok.  That's not the case here.

It also doesn't compute IMO that the ped crossed in the middle of the road, outside of a legal cross walk therefore its the peds fault.  By that logic, its fine then for me to watch Dave's latest video going down the road and its open season on jay walkers?  Might as well get a cattle guard installed on my car then.  Would it then be distasteful to put stickers on the hood for each ped mowed down to serve as a warning for would be jay walkers?
« Last Edit: May 25, 2018, 11:50:35 pm by orion242 »
 

Offline Brumby

  • Supporter
  • ****
  • Posts: 7699
  • Country: au
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #19 on: May 25, 2018, 11:53:38 pm »
Lets not get distracted with these arguments.

The purpose of this discussion is related to the engineering and the decisions in relation to that, not the other stuff.
 

Offline orion242

  • Supporter
  • ****
  • Posts: 514
  • Country: us
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #20 on: May 25, 2018, 11:55:51 pm »
Would think the liability issue is very related to engineering decisions made with these products...
 

Online max_torque

  • Frequent Contributor
  • **
  • Posts: 879
  • Country: gb
    • bitdynamics
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #21 on: May 25, 2018, 11:58:35 pm »
The ultimate "fault" is clearly with the unfortunate victim of this crash (high on drugs, not paying attention, and crossing where specifically prohibited)

This could have easily been a child chasing after a ball and running out in to the road.

So a druggie getting mowed down by an distracted driver is ok?  Does this change if its a child then?

No. it would, by the letter of the law (assuming the driver wasn't speeding, or driving without due care and attention etc) still be the child's fault.

Our road system (At least in the uk where i live) gives right of way to the vehicle unless specifically contra-indicated (ie signed pedestrian crossings etc)

No one is saying it is "ok to be run down by a driver", just that the system is quite clear, in that the vehicle has right of way. if you run out in front of a car (for what ever reason) then that is your choice and by definition your "fault"

And whilst a "Good" driver, should drive in such a fashion as to be able to avoid, or at least mitigate the effects of an errant object in their path, the fact of the matter is that through distraction, boredom or poor skill (usually observational failures) human drivers often don't react until too late. As perfectly proven by this incident, where the "safety" driver, despite being there for one primary purpose (and being paid to be there, ie, as their job) failed to prevent the accident. (in this case, it seems that failure wasn't directly their fault, as the system was in-sufficient to allow them to react in a timely fashion. ie lack of warning etc)

So to identify a "primary" fault, we have to look at the one single action that would have  with 100% certainty prevented the eventual death, and that can clearly be seen to be the initial victims act of walking into the road (where specifically prohibited remember) without either looking, or after making a poor decision (ie failed to appreciate the speed of the oncoming car ).  Any other mitigating action, that could have possibly prevented or mitigated the eventual result, but that you can't say  would have 100% certainly prevented it  ie the car driving slower, the driver paying more attention, cannot be therefore be the primary fault/cause.  (it becomes a contributory factor)
 

Offline Brumby

  • Supporter
  • ****
  • Posts: 7699
  • Country: au
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #22 on: May 26, 2018, 12:01:16 am »
Would think the liability issue is very related to engineering decisions made with these products...

We could say that about Uber's decision to disable the Volvo braking system - but that is NOT why I mentioned it.

I mentioned it because it was a very bad decision that served no real engineering benefit and increased risk of undesirable impact.  The liability discussion happens in a different room - one with far less engineers.
 

Offline orion242

  • Supporter
  • ****
  • Posts: 514
  • Country: us
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #23 on: May 26, 2018, 12:12:50 am »
Our road system (At least in the uk where i live) gives right of way to the vehicle unless specifically contra-indicated (ie signed pedestrian crossings etc)

Quick check, that seems to be the case here as well.  If that is the case in NV, and it likely is, they prob won't file any charges.
 

Offline Brumby

  • Supporter
  • ****
  • Posts: 7699
  • Country: au
Re: EEVblog #1088 - Uber Autonomous Car Accident Report
« Reply #24 on: May 26, 2018, 12:17:01 am »
And there's the flip side ....

While Uber may not have trouble with the law ... they certainly have set themselves up for a bucketload of criticism from the engineering fraternity.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf