Forum menu
But it’s notable that the police response was to blame the cyclist 100% and not even consider that the car/driver might have been a teensy bit at fault.
That's an American thing. Jaywalking is a crime. Therefore the driver/car is unlikely to get into major trouble. Will see if I am proven to be wrong on this.
Doesn’t this technology work in the dark?
Some of the very basic/early systems need visible light to function at 100%. As they rely on cameras. So at a very basic level, less light = less functionality. Anything driving around now, engineered by anyone who is paying attention, will have at least 3 systems using different media. (Visible light, lidar, radar, ultrasonic etc) and you'll need to physically block the sensor to cause any major issue.
I don't see how jaywalking being a crime could absolve the driver/operator from any guilt. Unless by the same logic I'm allowed to beat to death any driver I see using a phone or breaking the speed limit.
I don’t see how jaywalking being a crime could absolve the driver/operator from any guilt.
On a personal level, neither do i. But it happens with depressing regularity. Even in Europe.
It's possible that the US takes a different view than the UK, however we saw in the Charlie Alliston case last year that it's not acceptable under our laws to disregard the safety of pedestrians, even if the pedestrian had initially put themselves in danger by their own actions.If the driver could reasonably have been expected to have been able to take action to avoid or mitigate the collision then they can be held to account for their failure to do so. In my view, from the evidence I've seen, I think something could/should have been possible to mitigate the outcome but wasn't due to a combination of a completely ineffective safety system in the vehicle along with the almost complete inattention of the person that is supposed to supervise those systems. If it turns out there was any corporate awareness of the potential ineffectiveness of the system in those sorts of situation then we could be looking at manslaughter type charges (again, as was the case with the Charlie Allliston trial i.e. where a vehicle known to have its safety compromised was driven in a way that it would be reasonable to assume that it was risky).
On the other hand - if this was a normal car being driven by a person without (or possibly even with!) the video evidence being available then the police would probably just be chalking it up as a tragic accident due to the victims actions, even if they thought (but couldn't prove) that driver inattention was a factor.
Regardless of the technical reason, it's all about speed.
If you cannot stop after sighting an obstacle, you're going too fast for the conditions. That's regardless of whether you're an auto car or a driver.
The problem is most drivers in poor visibility conditions operate on the probability of the road being clear rather than the possibility there may be an soft squishy human there.
So it's that simple, if you can't stop, you're too fast.
uber auto was at fault IMO, as was the 'driver'. they've been playing catch-up WRT the tech. taking too many risks?
jeez edukator, i hope you're a nicer person in real life than you come across on this forum.
BTW, there is a decent, informed, review of the events here:
http://ideas.4brad.com/it-certainly-looks-bad-uber
by someone directly involved in autonomy.
they’ve been playing catch-up WRT the tech.
They do seem to have taken a few different innovative approaches to that though. Hence the recent court case.
I like how someone says, after watching a video where the events were known “i’d have managed to avoid her”. Sorry, but, no you wouldn’t
I'm guessing you haven't seen the video of a chap driving the same stretch of road at the same time of night?
It appears that visibility was no where near as bad as it seemed to be in the actual video of the incident...
BTW, there is a decent, informed, review of the events here:
http://ideas.4brad.com/it-certainly-looks-bad-uber
/a>by someone directly involved in autonomy.
Very interesting reading. If it turns out the LIDAR was turned off to test effectiveness of just using the other systems, there was only one inattentive operator, and the car was driving at its limit then that is unforgivable.
However, this being the US, the fact she was jaywalking will probably trump all that.
"Epicyclo
If you cannot stop after sighting an obstacle, you’re going too fast for the conditions. That’s regardless of whether you’re an auto car or a driver."
Which is true when there's an obstacle already in the road. When the obstacle walks in front of you it doesn't really apply.
That's not to say it's alright, just that this particular argument doesn't seem to hold up
When the obstacle walks in front of you it doesn’t really apply.
Depends on what you mean by "walks in front" really. Jumps out from hiding then maybe yes. Assuming what was hiding them wasnt something that should make you suspicious (for example a school bus).
If they walk across a multi lane road then despite cursing them should probably be able to avoid them. Speed depending.
In this case the sight lines (if you aint relying on a crap camera with no dynamic range) were reasonable. Not ideal but should have had at least some reaction.
Driver reaction Time is not a black and white science (like most things) but generally, most studies split "having to stop suddenly" events into three broad categories:
"Expected Sudden Braking Requirement"
Sounds a bit contrary, but covers for example driving towards a green traffic light, which you have observed in plenty of time, but know can suddenly change to STOP at any given moment, you just don't know when. In this case, an observant driver is prepared, and reaction times (lights change to start of braking) can be as little as 0.7 seconds latency as the driver may be already coasting up to the lights covering the brake pedal
"Unexpected Sudden Braking Requirement"
Where you have spotted a potential hazard in good time, but it is unusual to actually have to brake. Example is a pedestrian standing on a pavement waiting to cross. Most times, they have seen you, and wait till you have passed, but occasionally they haven't and might step out in front of you. Again, a good driver is watching them and is prepared to have to suddenly stop. In this case however, they are still on the accelerator, and so an extra delay occurs in moving their foot to the brake, and often in deciding if they should not stop, stop, or swerve. Sometimes it's not clear that there will be a collision until rather late (often the "will they won't they" initial step off the pavement happens followed by them wavering having seen you, but then continuing regardless). Average reaction time to these events is considered to be in the order of 1.2 seconds
"Surprise Sudden Braking Requirement"
Here the driver has NO prior warning. You are just driving along and suddenly something is in your way. You did not, nor reasonably could have expected them to be there (like the pedestrian in this current case) They probably have made a poor judgement decision eg choosing a poor crossing place for example. Here reaction time is typically, on average (as shown by the majority of studies) to be around 1.5 seconds.
.
.
In all cases, the time to get to fully developed threshold braking is longer, studies show that for most drivers that takes an additional 300ms, assuming they do eventually press the brakes hard enough to get to max retardation (many tests and studies have shown that most drivers not only don't know what "full braking" actually is, are unaware of how hard they need to press the brakes to achieve it, and can be 'spooked' into lifting off the brakes when the ABS kicks in)
You need to post some links, Maxtorque.
How about the ruler test someone mentioned above
20cm is about 200ms. Have ago
How about learning to drive sites. There's one below somewhere
How about the debates about jump starts in formula one (drivers generally take about a quarter of a second to react so less is suspect).
What about in other disciplines such as music. Most rock is between 140 and 170 bpm so three down picks a second for Chuck Berry and he still corrects left hand fluffs between beats.
I've been watching old motorsport vids today with tenth of second clocks running. Drivers react to errors or the unexpected in about 0.3 seconds.
Rather than talking about studies, link us some because your times fly in the face of traditional wisdom taught to learners the world over. It's conservative so as not to be misleading and cause road users to be overly optimistic.
https://www.permisdeconduire-online.be/snelheidwet6.htm
And getting back to the Uber car then if the woman was pushing the bike at walking pace then she was in the road for 4-5 seconds before the car hit her so even using your slow reaction times a calculation says a normally vigilant driver would have been stopped or travelling very slowly. The Uber car didn't react at all.
You're very happy to marvel at cars that do 0-100kmh in 5 seconds but won't believe drivers can stop a car from half that. A Tesla will reach 35mph in under two seconds and almost any modern car will stop from 35mph in under 2 seconds.
Have you founf the post where peoplesaid the car was not at fault yet?
Honestly I'd have to agree with maxtorque there, it correlates highly with what I see everyday,
F1, you ar are anticipating the green light it WILL happen, nobody drives around with that level of anticipation
Ruler test again you KNOW it WILL happen
If a random stranger across the street threw a ruler at you would you catch it? Are you prepared for anyone to throw a ruler at you at any point?
Every day thereis evidence that people do not react in time.
Oh my God. Circular argument is circular.
Edukator; you can’t compare reaction times of highly alert racing drivers to everyday joe drivers in mundane situations. It would be useful if you actually explained what exactly you disagree with of Maxtorques assertions. You actually seem to agree on a lot of points.
Maxtorque; I think it’s fair to say (and I think you agree) that the Uber technology failed badly. I think it’s also fair to say that a competent driver, paying a normal amount of attention could have at least started to take evasive action; which would have undoubtedly increased the survivability of the incident. If the driver could see much further, due to vastly superior dynamic range of the human eye, than the poor dash cam footage suggests, then it would make it even more likely that an attention paying human could have avoided or mitigated the collision.
So could a computer controlled car. The question is why didn’t it?
and that’s for Uber to answer, probably in court.
[i]v8ninety wrote:[/i]
Maxtorque; I think it’s fair to say (and I think you agree) that the Uber technology failed badly. I think it’s also fair to say that a competent driver, paying a normal amount of attention could have at least started to take evasive action; which would have undoubtedly increased the survivability of the incident. If the driver could see much further, due to vastly superior dynamic range of the human eye, than the poor dash cam footage suggests, then it would make it even more likely that an attention paying human could have avoided or mitigated the collision.
So could a computer controlled car. The question is why didn’t it?
As one of the more vociferous supporters of autonomous cars earlier in this thread, I have to agree with all that - given the evidence we now have it seems likely a human paying attention would have done better (though of course we come back to how much attention the average human pays - the actions of the safety driver tends to be evidence in favour of fully automated cars, ones which work properly!) The automated car did a rubbish job - TBH any decent implementation should have spotted the woman and stopped. I'm still not sure whether an average driver would have avoided the collision, but an autonomous car should do much better than an average driver.
This certainly isn't evidence against autonomous cars, but it is evidence that Uber need to do a much better job, and likely that they are criminally negligent (as is the "safety" driver).
This certainly isn’t evidence against autonomous cars, but it is evidence that Uber need to do a much better job, and likely that they are criminally negligent
That's very much my thoughts on the situation. The mobile phone video posted early in the thread by an independent person showed that visibility was good.
The pedestrian was already half way if not more across the multi lane road with no other traffic complications.
And darkness /shadow wouldn't have affected the radar /vidar or whatever it's called, so something was clearly very wrong with the scenario, and it wasn't the bored human driver.
I can't help but think the rush to market, and the relaxed acceptance criteria for live road tests in that region are the real reasons for that poor woman getting mowed down.
I guess you don't like me then Edukator? Is it because you feel threaten by me that you feel you need to attack and insult me in every post you make?
Anyway, back on topic:
It's not about me (how i drive) or about you (how you drive) it's about how an average person drives, and there are plenty of Reaction time studies (google them) to back up the figures i have suggested above.
Of course an F1 driver can react to the "Lights" in less than 1 second. Drag racers can even react BEFORE the lights change by pre-empting that change. However, that is completely irrelevant to how the average person drives, especially when the thing to which they are reacting is completely unexpected.
I have made an attempt to describe these facts, with some objective figures where ever possible, and you just call me names and insult me. Well thanks, but no thanks.
Please don't insult me again.
TBH any decent implementation should have spotted the woman and stopped
All we know is that car didn't stop. We don't know why. It's entirely possible the car did see her and for whatever reason, decided not to stop.
We don’t know why. It’s entirely possible the car did see her and for whatever reason,
Exactly. I would take a wild guess that the reason was that that something failed (hardware or software) as can't believe such an oversight was known (or didn't come up in testing) yet the car was still released for use on the road.
Even after another 10-20 years of development where they will clearly be better than any driver there will still be accidents (things go wrong, software or hardware, and not every single situation can be accounted for)
Taking the UK, if the deaths dropped from 1700 to 100 per year that would be a success. Not sure why anyone would be against that progress and highlight the 100 deaths as being the reason we shouldn't have Autonomous vehicles
^thats very much my view. The other thing is, each serious failure will be studied and learnt from and improvements can be rolled out fairly quickly across the whole fleet, whereas when a human driver cocks up, people just shrug and ignore it.
From the link above:
http://ideas.4brad.com/it-certainly-looks-bad-uber
/a>
Uber needs to say why this did not happen. I have seen one report -- just a rumour from somebody who spoke to an un-named insider, that the LIDAR was off in order to test operations using just camera and radar. While that might explain partly what happened, it is hard to excuse. Even if you want to do such tests -- many teams are trying to build vehicles with no LIDAR -- the LIDAR should remain on as a backup, triggering braking in exactly this sort of situation when the other systems have failed for some reason, or at least triggering a warning to the safety driver. It would be highly unwise to just turn it off.
And from this BBC link:
http://www.bbc.com/news/technology-43523286
Our Lidar can see perfectly well in the dark, as well as it sees in daylight, producing millions of points of information.
However, it is up to the rest of the system to interpret and use the data to make decisions. We do not know how the Uber system of decision-making works.
The other youtube video's showing how well lit that road is are pretty damning for both Uber and for anyone who's attempting to argue there wouldn't have been time for a human driver to react.
Based on those video's and how good the visibility actually was, then I'm 100% sure I wouldn't have killed that woman if I'd been driving.
Northwind
...Which is true when there’s an obstacle already in the road. When the obstacle walks in front of you it doesn’t really apply...
Which gets back to my comment on probability versus possibility.
Gambling on the probability of there being nothing on the road in poor visibility conditions (usually a safe bet), as opposed to driving based on the possibility of something appearing.
If you can't react safely in the range of your visibility (and that includes peripheral), you're going too fast.
who is actually mad enough to get into a coffin on wheels with a bit of software in charge?
About 3.5 billion a year as far as i can tell. ;o) And most of them are several thousand metres up. 😉
</div>
you are kidding rigght?
Comparing a highly regulated system of trained pilots and air traffic controllers with multiple safeguards with a bunch of drivers trying to deliver parcels, do the school run, tweeting and god knows what else that have completed a driving test which is a minimum standard that they probably passed many years ago...
...has got to be one of dumbest things I've heard in a while 😉
carry on .
As ever "theory of errors" there are multiple mistakes and errors here no one of which is wholly responsible.
the woman walked in front of a car
the autonomous systems failed / were turned off
The "driver" was not paying attention
I am interested in how errors happen in my (nuring) world. anytime you do analysis there is always a string of errors that when they ocur individually cause no issues but when they occur together you get catastrophe
"A human would have been able to react"
Lets say it again shall we, repeat after me everyone:
THERE. WAS. A. HUMAN. IN. THE. CAR
And even worse, they were there with the explicit understanding that they were the FAILSAFE. They had the overall responsibility to prevent the autonomy from causing an accident! And worse, they were PAID to be there!
It wasn't someone just driving to pick their kids up from school, someone nipping down the shops for a newspaper. no, it was someone who was driving to achieve a task where the driving was "secondary", their entire raison d'être was to be there as back up.
And they failed. For exactly the same reasons that un-assisted human drivers regularly fail, they also failed.
Arguing that an average human driver would have easily avoided this accident is silly, because we have already proven without doubt that in this case, the human driver did not avoid this accident!
However, for the avoidance of any doubt, I am not saying that under idealised conditions an alert, skilled,observant driver, couldn't have at the very least mitigated the severity of the impact. An F1 driver driving that car with a much higher experience and skill level may well have spotted the pedestrain earlier, braked quicker, braked harder, precisely severed around them, but, inevitably, it wasn't an F1 driver in the driving seat. In fact his case proves the very reason we need Autonomy in the first place, which is "average human drivers" are not actually very good at driving.
Like most accidents, there is unlikely to be a single cause, and there will be multiple small affectors (the swiss cheese model) where any of which being removed would have prevented the accident occurring:
Until all the facts are known, there is no way to know what was the most important factor. But as i have said before, the difference this time is that, thanks to this being an autonomous vehicle we do have objective data from which to make a valid conclusion, and that conclusion can then be used to prevent the same accident occurring again!
Arguing that an average human driver would have easily avoided this accident is silly, because we have already proven without doubt that in this case, the human driver did not avoid this accident!
No we havent. What was proved was someone couldnt switch from being an observer to being a driver.
Note that the observer role has several aspects not just monitoring the road for missed obstacles but also monitoring the vehicle readouts. They also have to deal with an even more high probability of getting distracted than the average driver since they have more excuses.
But as i have said before, the difference this time is that, thanks to this being an autonomous vehicle we do have objective data from which to make a valid conclusion
Maybe, maybe not. Depends on whether the issue can be identified. It is also quite possible even if it is solved then it will be applicable only for Uber and, of course, could introduce a dozen new bugs.
The dude sitting in the car shouldn't have been pissing around with his phone. If the video is accurate then it looks like the cars lights were turned off, no car has a low beam that is that poor.
Surely uber should vet who they are hiring for something as important as this project.
While it seems that the observer in the car should have been behaving as a driver it's very clear that they were not - so it's just plain stupid to use them being there as validation therefore that a driver couldn't have avoided the collision.
The dashcam footage shows the observer, in the time leading up to the impact, as looking away from the road for over 5 seconds. Before that they'd have a quick look having previously been looking away from the road for over 7 seconds. Now it is possible a very reckless driver looking at their phone might have done the same, however if they did then they'd very likely (in the UK at least) to be charged with causing death by reckless.- particularly if they were also speeding at the time. I'm pretty sure that, if they were in the UK, that observer would be going back to jail. Arizona does seem very keen to have the driverless car testing done there though...
The dude sitting in the car shouldn’t have been pissing around with his phone. If the video is accurate then it looks like the cars lights were turned off, no car has a low beam that is that poor.
The video gives a pretty bad idea of the lights - apparently as standard the dipped beam on the Volvo that was involved should light even a pitch black road clearly for about 4 seconds worth of driving at 38mph - not the 1.5 seconds or so that the video implies.
Lets say it again shall we, repeat after me everyone:
THERE. WAS. A. HUMAN. IN. THE. CAR
But the point is: THE. HUMAN. WASN'T. DRIVING
The computer driving the car must have been distracted.. checking it's Facebook or twitter feed rather than using its processing power for actually driving..
They were supposed to be though. In the same way that the human is driving when in a car with lane assist and active cruise control. Ultimately it's the human who is responsible and they failed. Not the only thing to fail, though you do wonder if the car was in a standard mode and this accident could have happened any time, it's just that as with bad human drivers they'd been lucky to avoid such situations.
I'm not sure I agree with all of maxtorque's post - given other videos available I'm now thinking I'd have avoided that collision if I'd been driving - but he's right that this is also evidence of the need for fully autonomous cars, given there are already vehicles on the road with enough assists to encourage drivers to act in the same way as here.
Ultimately it’s the human who is responsible and they failed.
I am not so sure. Its going to be natural for someone to lose concentration in that sort of scenario.
Its poor design that the car was operating at level 3 but doesnt seem to have anything in place to ensure the operator is prepared to take over. Especially since Uber had cut back to a single operator rather than a pair.
given there are already vehicles on the road with enough assists to encourage drivers to act in the same way as here.
Or you make sure those assists are restricted in usage and the driver is forced to remain fully active.
I do think level 3 should be banned. Go from limited 2 to 4.
[i]dissonance wrote:[/i]
Ultimately it’s the human who is responsible and they failed.
I am not so sure. Its going to be natural for someone to lose concentration in that sort of scenario.
Well exactly, they failed. That it's natural doesn't change that they failed, because that was their job.
Or you make sure those assists are restricted in usage and the driver is forced to remain fully active.
Horse has already bolted.
Well exactly, they failed. That it’s natural doesn’t change that they failed, because that was their job.
But they were always going to fail. How the MMI functions is every bit as important as the algorithms used and the performance of the sensors. It is an integral part of the driverless car issue. Uber have effectively ignored it.
That it’s natural doesn’t change that they failed, because that was their job.
It shows that the job description was wrong for all but a few. So a failure of recruitment and the safety processes used,
Epicsteve
The video gives a pretty bad idea of the lights – apparently as standard the dipped beam on the Volvo that was involved should light even a pitch black road clearly for about 4 seconds worth of driving at 38mph – not the 1.5 seconds or so that the video implies.
Which makes me wonder if the video hasn't been doctored to make it look darker so that the bike appeared more suddenly.
Was the video released by Uber or by the police from an original unedited recording?
What's going to happen when autonomous cars start getting a few years old? You know, dirty sensors, electrical failures of sensors etc? It happens with other sensors in a car, so why not with those?
(I'm not against self driving cars, I welcome them because in a few years time I'll probably be of an age where I shouldn't be behind the wheel.)
dissonance
Its poor design that the car was operating at level 3 but doesnt seem to have anything in place to ensure the operator is prepared to take over. Especially since Uber had cut back to a single operator rather than a pair.
Where did you find out that it was level 3?
I do think level 3 should be banned. Go from limited 2 to 4.
As I understand it (might be wrong) level 3 autonomy is limited by GPS which monitors speed and location to determine when it's safe for the car to take over. So below 40mph, outside the city, it'll take over, at least those are the parameters Audi are using.
What’s going to happen when autonomous cars start getting a few years old? You know, dirty sensors, electrical failures of sensors etc? It happens with other sensors in a car, so why not with those?
They'll probably have to develop some kind of annual road worthiness testing procedure for cars over a certain age.
[i]epicyclo wrote:[/i]
Which makes me wonder if the video hasn’t been doctored to make it look darker so that the bike appeared more suddenly.
Probably just a cheap sensor in the dashcam which isn't good in low light.
What’s going to happen when autonomous cars start getting a few years old? You know, dirty sensors, electrical failures of sensors etc? It happens with other sensors in a car, so why not with those?
I guess they'll have to make them so they self test and detect failures (and poor performance because they need cleaning) and don't let the car operate if there's a problem.
jimjam
...They’ll probably have to develop some kind of annual road worthiness testing procedure for cars over a certain age...
Aye, I thought of that. The MoT is good at spotting physical deterioration in mechanical devices, but electronic stuff usually doesn't show signs of degradation, just goes phhht.
There's going to need to be some continuous self- testing procedure - or failsafe redundancy.
Who do Uber and Google (and the rest) need to convince that driverless cars are safe? Is there an automotive version of the FAA and CAA?