Jan and Klodi's Party Bus - part II **off topic discussion**

Hurrache

Baron de Charlus wrote: »

It's rare to hear of a copper email someone after an encounter, effectively saying "I was wrong and I'm sorry".

That's great, but I wonder would the apology be forthcoming had they been a bunch of ordinary riders without cameras.

tomasrojo

I suspect that if there were a way of stopping narcissists from driving, it would be more effective and cheaper than developing autonomous vehicles and then reconfiguring existing cities to compensate for the AV's shortfalls. I dunno, take the licence away when people drive in such a way to make it unambiguously clear that they regard any delay or slight as requiring the unleashing of enormous power?

AVs seem to have made amazing progress, but they still have severe trouble with city travel, poor weather, or even unusual parking, and the current fix seems to be to have remote human operators on standby. This is not an elegant solution as things stand. "Thank you for calling Google Car. All our operators are busy right now. Requests for remote operation will be dealt with in strict rotation."

CramCycle

tomasrojo wrote: »

I suspect that if there were a way of stopping narcissists from driving, it would be more effective and cheaper than developing autonomous vehicles and then reconfiguring existing cities to compensate for the AV's shortfalls. I dunno, take the licence away when people drive in such a way to make it unambiguously clear that they regard any delay or slight as requiring the unleashing of enormous power?

I used to agree with you but more and more, until we have a garda presence capable of enforcing such bans and increased punishments on top of just simply removing the licenses, I no longer think it will.

The number of people who get picked up by AGS for no insurance, no tax, no license, already banned, on a daily basis, considering these are a fluke that they even got caught in the first place, is shocking. If enforcement was at the right levels then they would be all of them but we all know that is not true. My mother tells me of people losing their licences down home and I see them out driving the next time I am down, without a care in the world.

I say this as someone who has handed in the license plate number and description of a family member who is medically unfit to drive and therefore has no insurance. I know they still drive when I am not there. I took their keys and they ordered new ones under the pretense that they may sell the vehicle. My brother declared it off the road to revenue, but they still drive it. So the gardai have the name, plate number, know there is no tax or insurance and still this person gets away with driving. My brother started letting the air out of the tyres, they bought a pump and leave it stored in the vehicle.

Some people don't care and never will, they cannot see the danger or even that they are the danger.

ED E

All we need is ANPR for those "Disco Drivers". Gardai get an alert every time such a person comes anywhere near a town. After the third arrest they'll give over.

CramCycle

ED E wrote: »

All we need is ANPR for those "Disco Drivers". Gardai get an alert every time such a person comes anywhere near a town. After the third arrest they'll give over.

100% agree, but so far the only people putting ANPR up are motor insurance companies on the M50, whereas it should be built into speed cameras, the tolls and every red light camera in Ireland over time.

ford2600

https://www.theverge.com/2018/3/21/17149958/tempe-police-fatal-crash-self-driving-uber-video-released

The Deceased lady was pushing a bike across a road, had crossed another lane before entering uber car's lane; given there should be no darkness for AV and perception, reaction and brake actuation times should be in milliseconds this is precisely type of accident AV should not be having.

Long way to go with this technology on basis of this accident; it's not as if she stepped out from behind a parked van, depending on road width and a walking speed of 5km/h(1.38m/s) their paths were a collision course for at least 4 seconds; probably more. That's a lot of emergency braking time; assuming a friction factor of 0.8 (on a dry road and a modern Volvo) 4 seconds emergency braking has you stopped from 113km/h

buffalo

ford2600 wrote: »

Long way to go with this technology on basis of this accident; it's not as if she stepped out from behind a parked van, depending on road width and a walking speed of 5km/h(1.38m/s) their paths were a collision course for at least 4 seconds; probably more. That's a lot of emergency braking time; assuming a friction factor of 0.8 (on a dry road and a modern Volvo) 4 seconds emergency braking has you stopped from 113km/h

Aye, if the tech is supposed to be better than humans, this example doesn't fill me with confidence.

Video isn't graphic, but might be distressing to some:

seamus

To be fair to the safety driver, while he wasn't paying attention, from the video alone it seems unlikely that a human driver would have fared any better avoiding that collision.

Maybe it is just the video, but the street lighting alone seems very poor, I was watching it thinking, "why doesn't the car have its full beams on?". But video often has difficulty showing more subtle differences in lighting, so to the human eye it may have been a good deal brighter.

Nevertheless, as said this is exactly the kind of incident that AVs should be proving they are miles better at than humans. They should (theoretically) be capable of driving in pitch blackness without lights.

Lumen

That video is strongly misleading. It would be more instructive to see videos from the lidar system, not the optical cameras.

Since Uber cars can see in the dark, the question is simply "why did it drive into a pedestrian it could see?".

This has to be the easiest accident to avoid (easy for me to say as the non-programmer).

If a robot car cannot avoid a pedestrian walking out slowly across an empty, straight road whilst wheeling a bicycle fully side-on, what hope for the hard cases?

magicbastarder

vaguely related topic to the above - do any manufacturers yet include any sort of flight data recorder system in cars?
probably issues related to data protection (and car manufacturers not wanting to implement something without being forced to); but the technology to include a GPS unit and the last hours worth of info of driving data, would be very useful to have. you could isolate it from all the other car systems, so you'd need physical access to the car to get the data.

doozerie

ford2600 wrote:

The Deceased lady was pushing a bike across a road, had crossed another lane before entering uber car's lane; given there should be no darkness for AV and perception, reaction and brake actuation times should be in milliseconds this is precisely type of accident AV should not be having.

Long way to go with this technology on basis of this accident;

Lumen wrote:

This has to be the easiest accident to avoid (easy for me to say as the non-programmer).

Absolutely. When I've read about the challenges posed to autonomous vehicles by cyclists, the emphasis has been put on the relative unpredictability and variability of cyclists - a cyclist may appear on the left or right of a car, the cyclist's speed could be anywhere in a wide range, etc. By comparison, pedestrian behaviour falls into a much narrower range of speed and position which, despite still being challenging to code for (because ...humans and their behaviour), was portrayed as "manageable".

But more so than ever before, I wonder how much of what has been reported (positively) about autonomous cars has been little more than marketing and propaganda. As above, the video of this collision suggests that it simply should not have happened, this doesn't appear to be the kind of edge case that any and all systems struggle with, this seems to be a relatively "simple" scenario for a computerised car to deal with.

The behaviour of the driver is a real concern too, he seemed to be no more than an entirely passive passenger in the car. Presumably he is there as an extra safety measure should the car not perform "properly", but given how little time he spent with his eyes on the road his behaviour suggests that he expects the car to do everything and to do it well. I read somewhere that some states in the US require that autonomous cars be manned by a safety driver before they be allowed on the roads in their current form, this video demonstrates that even this basic requirement is not always enough and I wonder whether such cars will be prohibited from using public roads for the next while. That's not to say that the driver could actually have done anything had his eyes been on the road throughout, but his apparent complacency behind the wheel is still shocking.

magicbastarder

i wonder if the fact that it was a 'pedestrianised cyclist' is pertinent - would the car have detected a bike and rider behaving like a pedestrian, and confused the system?

i know it's only one incident, but if you were to pick a pedestrian at random for the car to hit, the chances they are pushing a bike would be very low.

tomasrojo

Lumen wrote: »

If a robot car cannot avoid a pedestrian walking out slowly across an empty, straight road whilst wheeling a bicycle fully side-on, what hope for the hard cases?

There will be a lot of uses for the aspects of AVs that actually do work; I imagine the cheaper features will be standard safety features for human-driven cars. I can imagine freeways/motorways having stretches where no human intervention is required for freight. But the robot car coming to collect you and bring you to a destination in the city seems way beyond them right now.

And, as you say, this scenario where an AV failed isn't as hard as the city journey scenario.

(And then there's the problem with the cars also being remotely controllable. I can't imagine that being left unhacked for very long.)

buffalo

magicbastarder wrote: »

vaguely related topic to the above - do any manufacturers yet include any sort of flight data recorder system in cars?
probably issues related to data protection (and car manufacturers not wanting to implement something without being forced to); but the technology to include a GPS unit and the last hours worth of info of driving data, would be very useful to have. you could isolate it from all the other car systems, so you'd need physical access to the car to get the data.

I have vague memories of some insurer offering to put a black box device in your car in return for reduced premiums.

magicbastarder wrote: »

i wonder if the fact that it was a 'pedestrianised cyclist' is pertinent - would the car have detected a bike and rider behaving like a pedestrian, and confused the system?

i know it's only one incident, but if you were to pick a pedestrian at random for the car to hit, the chances they are pushing a bike would be very low.

The car thinks, "that thing moving across the road slowly isn't at all pedestrian shaped, so I'll ignore it"? Seems like a recipe for disaster.

magicbastarder

buffalo wrote: »

I have vague memories of some insurer offering to put a black box device in your car in return for reduced premiums.

yep, but i think that's more for tracking than actual collision investigation.

e.g. there was a case reported on (and discussed here) where a motorist struck a pedestrian at night on a road with a 100km/h limit, and the motorist at the scene estimated he'd been doing about 90. but the garda team who assessed the site revised that down to 45-50km/h i think (slightly hard to believe). having this information from the car would clear up a lot of the ambiguity about collisions.

i don't know if data collected by insurance companies would be useful (or admissable) in such a scenario.

tomasrojo

buffalo wrote: »

I have vague memories of some insurer offering to put a black box device in your car in return for reduced premiums.

Yes, you get the reduced premiums if the black box shows a pattern of consistent, calm driving. I read about in the Money section of The Guardian:
https://www.theguardian.com/money/2016/mar/26/black-box-car-insurance-cuts-young-drivers-premiums

Mostly aimed at younger drivers, allows them to demonstrate they don't fit the statistical profile of their age bracket. Superficially (as in, I haven't given it any in-depth thought, and it's not something I have any expertise in), it sounds as if it should be a standard approach, not just an opt-in for the young.

New cars rolling off the production lines are already installed with the next wave of on-board diagnostic tools, negating the need to install an actual device.

tomasrojo

magicbastarder wrote: »

yep, but i think that's more for tracking than actual collision investigation.

e.g. there was a case reported on (and discussed here) where a motorist struck a pedestrian at night on a road with a 100km/h limit, and the motorist at the scene estimated he'd been doing about 90. but the garda team who assessed the site revised that down to 45-50km/h i think (slightly hard to believe). having this information from the car would clear up a lot of the ambiguity about collisions.

i don't know if data collected by insurance companies would be useful (or admissable) in such a scenario.

Don't know if this means it's admissible in court. Presume so?

The data can also be used to reconstruct crashes to determine fault.

Not quite the same, but also from that article:

This technology can also protect policyholders from false accusations being made against them. Richard King says: “One of our young drivers was recently arrested and accused of being involved in a serious crime. We were able to prove within moments that he wasn’t near the scene of the crime and that his number plates had been cloned.”

Diarmuid

Lumen wrote: »

If a robot car cannot avoid a pedestrian walking out slowly across an empty, straight road whilst wheeling a bicycle fully side-on, what hope for the hard cases?

As heartless as it sounds, there will be the easy deaths and the hard deaths with AVs before we get to a point of "zero"deaths. The big difference is once you teach (program) the car to handle a scenario, it, and ALL of those cars won't make the same mistake again. Humans will keep making the same mistakes as long as they are driving.

tomasrojo

I'd bear in mind that the deaths per billion km figure for AVs is potentially misleading when compared with human drivers, since
an awful lot of the journey time has been in places like Arizona.

In the long term, I assume all the problems will be ironed out. But right now a lot of people have invested a lot of money in a technology that isn't ready to do what the creators said it would be able to do now. So already you're getting calls for cities to be simplified, cyclists and pedestrians to be electronically tagged. Because AVs right now don't work as advertised.

EDIT: looking around, the deaths per billion miles in the USA seems to be about 12.5 for driving overall. The death total for AVs seems to be two on a total of a few million miles. The non-fatal collision totals are more favourable for AVs, but, with very few data points, it's hard to tell what the underlying death rate might be for AVs. If you calculate it right now, they're faring worse than human-driven cars, but the usual caveats about very small numbers apply.

(And, as I said, a lot of those miles are on roads with few users.)

ford2600

magicbastarder wrote: »

vaguely related topic to the above - do any manufacturers yet include any sort of flight data recorder system in cars?
probably issues related to data protection (and car manufacturers not wanting to implement something without being forced to); but the technology to include a GPS unit and the last hours worth of info of driving data, would be very useful to have. you could isolate it from all the other car systems, so you'd need physical access to the car to get the data.

Yes.

Best of luck getting your hands on it.

doozerie

Diarmuid wrote: »

As heartless as it sounds, there will be the easy deaths and the hard deaths with AVs before we get to a point of "zero"deaths. The big difference is once you teach (program) the car to handle a scenario, it, and ALL of those cars won't make the same mistake again. Humans will keep making the same mistakes as long as they are driving.

Zero deaths is certainly what should be aimed for but I'm not convinced that can ever be achieved in reality. Software bugs are almost inevitable, and computers sometimes fail, etc., but perhaps more than anything else they will continue to share space with arguably the most unpredictable and unreliable variable of all, human beings in the form of motorbikers, cyclists, and pedestrians.

For now, at least. Perhaps there will be a big enough push behind autonomous vehicles at some point that the case will be made that unpredictable elements shouldn't be allowed on the roads at all. "For their own good", etc. Feck that.

rubadub

Lad in work told me of this death with a big shocked head, I was expecting to see the car careering off the road into pedestrians. After seeing it I wondered how many of these are on the road, and what the death rate is compared to normal drivers.

I was thinking of the stats you see about ecstasy tablets, papers headlining news of a single death once a year, when in fact they are really advertising the fact it ridiculously relatively safe compared to other recreational drugs. You simply do not see headline news about alcohol related deaths, or the numerous daily road deaths in the US due to drivers.

The youtube clip had comments like

Even with the poor quality of the video, there's a solid 3 seconds between the first hint of an obstacle and the moment of impact. I'm pretty sure the average driver could've seen the pedestrian much earlier. According to the authorities, the vehicle was travelling at 38 mph (17 m/s). A typical car can stop from that speed in about 20 m. Even with poor braking, the car could've stopped in less than 2 seconds, leaving a whole second for reaction. Considering that the average human being can react to visual stimuli in about 0.25 s, it's not only more than enough time for a human driver to react, it's essentially ages for a computer, especially one that has access to LIDAR data, that can see way farther than the headlight can illuminate.

This Uber car either had a malfunction, or their setup is unsafe. There's just no way around it.

There are several going on about how much better they expect humans to be. But they seem to be ignoring the woman who was effectively a driver seemingly did nothing -was she very much different to the "average driver"?

tomasrojo

rubadub wrote: »

There are several going on about how much better they expect humans to be. But they seem to be ignoring the woman who was effectively a driver seemingly did nothing -was she very much different to the "average driver"?

Probably, in that operators of autonomous vehicles, including planes, pay an awful lot less attention than operators who are fully in control.

On top of that, the video shows that Uber's "safety driver" was looking down at her lap for nearly five seconds just before the crash.

Quite a lot of stuff here, including about Uber's corner cutting in this field of AV.
https://arstechnica.com/cars/2018/03/video-suggests-huge-problems-with-ubers-driverless-car-program/

doozerie

tomasrojo wrote:

Quite a lot of stuff here, including about Uber's corner cutting in this field of AV.
https://arstechnica.com/cars/2018/03...s-car-program/

That makes for interesting (and concerning) reading. The following is pretty damning in terms of an approach to anything where safety is a real concern:

And, of course, safety drivers are the ultimate form of redundancy in driverless cars. If they do their jobs well, safety drivers should be able to ensure that self-driving cars are at least as safe as a human driver during testing—since the driver will take over if the car makes a mistake.

By contrast, court records unearthed by the recent Uber/Waymo lawsuit showed that a key architect of Uber’s driverless car technology was contemptuous of this approach to driverless car safety.

"We don't need redundant brakes & steering or a fancy new car; we need better software," wrote engineer Anthony Levandowski to Alphabet CEO Larry Page in January 2016.

...when you have a large hammer (a team of software developers in this case), everything can start to look like a nail (something that can best, and perhaps only, be solved by software alone). As proves to be the case in many areas of life, things are rarely that simple.

LollipopJimmy

Being a petrolhead I am against AV and EV... I'm a Neanderthal and I like it that way, a car should be powered by a lead foot and gallons of fire and fury.

In saying that I don't believe in cars for commuting, I do it the odd time but am aware I am being lazy when doing this. For me going for a drive in something nice is a release.

Won't get a chance to get out on the bike this weekend, I've to head to Tipp for the night tomorrow so went hard in the gym the last few days. Light sparring last night and nearly had my head taken off my shoulders by a pro mma fighter :pac:

RobFowl

Must resist ....

https://www.adverts.ie/road-bikes/vitus-992-ovoid/15041490

magicbastarder

first rule of selling a bike - forget to mention the size.

magicbastarder

52cm - was on adverts 9 months ago, sold by someone else. bars have been changed since:

https://www.adverts.ie/road-bikes/vitus-992-ovoid-52-cm-road-bike/12960483

tomasrojo

doozerie wrote: »

...when you have a large hammer (a team of software developers in this case), everything can start to look like a nail (something that can best, and perhaps only, be solved by software alone). As proves to be the case in many areas of life, things are rarely that simple.

I think the informatic disciplines are going to have to come round to considering ethical and social ramifications. On the other hand, it was hard to make biologists, chemists and physicists take these on board, and they're not as Ayn Rand-ish on average as the tech bros.

Lumen

Diarmuid wrote: »

As heartless as it sounds, there will be the easy deaths and the hard deaths with AVs before we get to a point of "zero"deaths. The big difference is once you teach (program) the car to handle a scenario, it, and ALL of those cars won't make the same mistake again

Regression bugs!