Forum menu
[i]scuzz wrote:[/i]
I like the bit where cars won’t be able to communicate because android and apple. Glad I’m using an apple device so can read this! Phew!
I have no idea how I can read what you're writing given I'm on Android.
This place sometimes, you haven't shut me down but have a read back, Aracer. Do some research on the sensors sbeing used on the Uber and Google cars, and Mercedes. It's all a lot less sophisticated than you make out. In many parts of the world the human hand is the same temperature as ambient even if the cars had super accurate IR cmaeras capablle of identifying a 200cm2 surface at 30 m and differentiating well enough from the noise to hit the brakes - which they don't.
People who know better than you (no not me) won't even post.
People who know better than you (no not me) won’t even post.
Um, there are two posting on this thread who are actually working on the technology now.
I used to be very much against automated decision making until I actually started working with it and realised that, at a minimum, it is better than the average human operator and more often than not is better than the best human operator.
I do wonder if we are raised to believe that a human will always be better than a machine because of Hollywood. In the movies the human always beats the machine because the human has 'gut instinct' or some kind of 'X-factor' that a machine can never have. I think the story of the company who made the air combat control system on a $35 Raspberry Pi board that could beat the best human pilots kind of disproves that.
I would be interested in what the various the posters on this thread who say they are software engineers and therefore know that driverless cars don't work and won't for a long time actually do. Personally I'm not a programmer. I come from the operations side of things (oil and gas) and I now work finding ways that digital methods can be used to replace humans or supplement human operators.
Edukator the ig it your point misses is what else is happening when your constantly looking for that hand? (assuming decent enough eye sight) can you spot hands on the left and right of the road at the same time, how do you evaluate the risk and assign your attention to monitor them?
Given that VW (and lets not kid ourselves that they're the only manufacturer doing this) can happily lie about their vehicles for years, and knowing that every vehicle is built as cheaply as possible, regardless of it's price; with components that are badly designed, and often badly looked after, I fully expect AV to have no impact on safety whatsoever. But that's not what they're for.
Like most things like this, they will be built initially for those that can afford them, and lanes will be created for AV use specifically to whisk them along, carefully separated from every-one else ( for y'know: safety wink wink), but my guess is that won't do anything for safety overall, or congestion, or pollution, that cost will still be borne by others.
The Uber car was speeding: it was doing 38mph in a 35mph zone. And it made no attempt to brake.
I think our utopian dream of safe, considerate automated drivers just died. They're going to be just as awful as meat-based drivers.
I don't really see anything in the article to support what you're saying.
If you don't want AVs on the road I think pretty much any incident will do if you want evidence as to why they'll never work.
Just to add, it doesn't say if it was 38mph on the speedometer or on the GPS. Most cars' speedometers I've had have been about 10% out which would make the actual speed at or just below 35 mph.
Google have already demonstrated a driverless car that can recognise cyclists and behave appropriately. When stopped at a junction, it waits for cyclists to clear the junction first before it moves. In my experience, human drivers can't even stop outside the ASL boxes...
Given that VW (and lets not kid ourselves that they’re the only manufacturer doing this) can happily lie about their vehicles for years,
Mercedes are also being investigated.
The autonomous car can know about that car BEFORE it comes round the corner
It could also know about a car which doesnt exist. Once you start networking the cars you start getting into interesting security issues. Given how badly the car firms are doing currently at handling security that doesnt make me feel confident.
The speeding factor is interesting, I read that the road was recently reclassified from 45? To 35, was the car using slightly out of date data I wonder?
"You really think a driverless will change lanes and drive on the wrong side of the road to get better visibility, Aracer?"
Googles car already maneuvers itself to get a better view like this.
It actually caused a crash once, because as it edged forward to get a better view whilst in the middle of a junction waiting to turn left (in the US so equivalent to right turn here), the driver behind thought the car was proceeding and drove into the back of it. There was a report on the internet somewhere but I can't find it.
Not all autonomous cars are equal though. The tricky part is for the government to determine who's developed the technology sufficiently.
We need a driving test for autonomous vehicles.
Given that VW (and lets not kid ourselves that they’re the only manufacturer doing this) can happily lie about their vehicles for years, and knowing that every vehicle is built as cheaply as possible, regardless of it’s price; with components that are badly designed
Having worked in the automotive business that is one of my main worries, if you look at most car manufacturers there are parts that commonly fail due to crap materials or design. Water pumps with plastic impellers is one good example.
Imagine what will happen when the French car companies start making these things, they can barely get the electrics reliable on a normal car let alone anything that relies on a load of sensors and electrical systems to work correctly!
Imagine what will happen when
Ah the finest imagination based reasoning....
[i]Edukator wrote:[/i]
This place sometimes, you haven’t shut me down but have a read back, Aracer. Do some research on the sensors sbeing used on the Uber and Google cars, and Mercedes. It’s all a lot less sophisticated than you make out. In many parts of the world the human hand is the same temperature as ambient even if the cars had super accurate IR cmaeras capablle of identifying a 200cm2 surface at 30 m and differentiating well enough from the noise to hit the brakes – which they don’t.
People who know better than you (no not me) won’t even post.
To some extent we're discussing what is possible, not just what is being done now, but I stand by my assertion that even the current cars with their current sensors are better than the average human driver (and bear in mind that inevitably most of the carnage on the roads is caused by the below average drivers). You still seem to be taking a very contrary position to your normal one on the standard of human drivers here. Meanwhile even with the sensors they are using the information collecting and interpretation is far better than you suggest.
If you're meaning who you seem to be meaning, then he went off in a huff because his arguments were robustly challenged, whilst he appeared to be unaware of the use of automated systems in trains, thinks that autonomous systems have to be perfect, has a strange lack of understanding for a software engineer of how it's possible for AI systems to generalise and abstract, and you should note his wording on what he's worked on carefully - I've actually done work on sensors and systems interpreting them, if not directly on autonomous cars. Yet because he agreed with you he was the one who knew best?
I think we'll see the equivalent of 'jay-cycling' laws not too long after the jay-walking ones take effect.
________ Red Bottecchia_________
My uncle has a country place, that no-one knows about
He says it used to be a farm, before the Motor Law
Sundays I elude the ‘Eyes’, and hop the Turbo Freight
To far outside the Wire, where my white-haired uncle waits
Jump to the ground
As the Turbo slows to cross the borderline
Run like the wind
As excitement shivers up and down my spine
Down in his barn
My uncle preserved for me an old machine –
For fifty-odd years
To keep it as new has been his dearest dream
I strip away the old debris, that hides a shining bike
A brilliant-red Bottecchia, from a better, vanished time
Silent wheels, the 'snick' of shifters responding with a touch
Tyres spitting gravel, I commit my weekly crime…
Wind in my hair –
Shifting and drifting –
Mechanical music
Adrenalin surge –
Well-weathered leather
Cool metal and oil
The scented country air
Sunlight on chrome
The blur of the landscape
Every nerve aware
Suddenly ahead of me, across the mountainside
A gleaming alloy air-car shoots towards me, two lanes wide
I spin around with skidding tires, to run the deadly race
Go sliding through the valley as another joins the chase
Ride like the wind
Straining the limits of machine and man
Laughing out loud
With fear and hope, I’ve got a desperate plan
At the old footbridge
I leave the giants stranded
At the riverside
Race back to the farm
To dream with my uncle
At the fireside…
(adapted from 'Red Barchetta' by Rush/Neil Peart)
I think we’ll see the equivalent of ‘jay-cycling’ laws not too long after the jay-walking ones take effect.
Tesla's autopilot already recognises cyclists, although it classifies them as cars. Because it's a Level 2 partial automaton it still requires the driver to initiate the overtaking maneuver. Not sure I'd want to be the one to guinea pig it, but with level 5 automation there's a pretty good chance the car will actually give you safe distance when overtaking. Punishment passes might actually be confined to the past.
Punishment passes
Holy shit. Let's hope close passes become a thing of the past. As for the desire to 'punish' cyclists (for what) with near/actual homicide? Hopefully that too will go the way of the white rhino.
Did someone say 'ASL box'? Funnily topical as I collected some data last week. My first junction travelling from home has traffic lights and an ASL/bike box. It's a dodgy junction so it affords me some level of optimism. I counted the times it was free for my use last week. Of the 12 instances that I had need for it (me on bike, lights on red) the ASL space was 11 times blocked by oblivious/careless/impatient/delete as applicable drivers. The 12th instance there was just me, no cars, so I managed to get away from the lights (without being forced to dismount, walk the bike to ped crossing and wait for those lights, cross ped crossing, then remount). At least self-driving cars would recognise ASL boxes.
The tech is coming. Its already safer than the average driver but still makes mistakes. this will improve.
Cyclists the tech does have real difficulty with but again this will improve. given how quickly the tech has got this far in 10 or 20 years how far will it get?
Personally I am all in favour of it simply for the reduction in deaths that will result
Exactly - the principle argument here is not how capable autonomous cars can be but how routinely crap most drivers are - it's so routine that it's normalised and we don't even notice how low the standard the computers have to beat is. All this ethical dilemma stuff - well humans already have to make those decisions and they're rubbish at it, you'd do just as using a PRNG in the software to make the decisions.
@tj - from what I can work out, some on this thread seem to think autonomous cars will still cause 1700 road deaths a year.
I think most people are on board with the fact that autonomous cars will be safer.
However, when a person causes an accident they can be held to account - financially or criminally or whatever.
I think the issue with this accident is it brings up the question of who is held to account when the autonomous car causes an accident (which may or may not be the case in this accident).
<div>Malvern Rider
<div>
<div>Member</div>
</div>
</div>
<div>Punishment passes
Holy shit. Let’s hope close passes become a thing of the past. As for the desire to ‘punish’ cyclists (for what) with near/actual homicide? Hopefully that too will go the way of the white rhino.
</div>
The "punishment" would be for holding the driver up for a few seconds, or for passing them in slow moving traffic, or perhaps even filtering past stationary cars. I thought it was a fairly common term among cyclists.
I think the issue with this accident is it brings up the question of who is held to account when the autonomous car causes an accident (which may or may not be the case in this accident).
Firstly there will be significant logs of the event to review, facts can be found and evidence presented. As for the liability then there needs to be a mechanism for that
It is an odd situation.,. we currently report incidents as if they are already non-human controlled eg "Car kills cyclist in London Road Accident". The reality of the incident above is that a person on a bike died after being hit by a person driving a car and there are likely identifiable causes, if we can access accurate data. And I would imagine usually human error on one side or both is the major factor. Data reporting is poor, hence we are not able to accurately asses the causes, hence we cannot make effective changes to limit the risk in future, Hence why it gets attributed as an "accident".
It is difficult for me to see how that system can be _significantly_ improved. Driverless cars seem to me offer a hope of a continually improving system (that's the point of machine learning, no?), in that data should be more accurate in the main (it won't always be of course) and hence we can learn how to mitigate risk as we move forwards. "Blame" and "negligence" become less important than learning. Perhaps no fault compensation helps those injured. Just insurance by another name really.
There will always be incidents where driverless cars are to blame. And individually it is right to dissect critically, and they will look terrible. But FFS our roads are not safe. They don't even feel safe. Yes they are safer than X year or X country. But something like 5 deaths a day and 60-70 life changing injuries a day on UK roads alone. I'd be interested to know the data of human vs driverless inuries/deaths per million miles etc as things currently stand.
I don't think we should accept a sudden wholesale shift. But I can see a time soon when cars (or maybe lorries) offer self driving for motorways only (initially) and hand back to the human for the rest. And the process of data collection continues.......Is this not similar to the airline model?
The video is amazing, I'm not sure a human could have spotted her and stopped in time. The question now has to be why didn't the lidar system spot her.
I'm off to buy more reflective gear, she was invisible 🙁 . Modern lights are so good at being bright on the road but cutting out all spill that could hit other driver's eyes that there was nothing visible to indicate she was there until too late
Yeah, that is a really tough one to avoid. I don't think any of us could have avoided that.
This in an interesting article about the road/pavement design in that area and the "didn't use the crosswalk" angle.
It was an accident waiting to happen when you see the (bureaucratic) lack of thought. Looks like a clash between two departments/systems' priorities.
It's one of those the car sensors should have picked up but the driver was also notdoing anything to help themselves.
The real point there is how many drivers would have seen that?
The camera doesn’t pick up anything outside a 5m light bubble though. Human eyes are better than dash cams, but if that’s all you could see you would go slower, right? But this is a robot car with LIDAR night vision, she is clearly coming from the other side of the road not stepping out from behind an obstacle. I’m disappointed with the cars lack of reaction, not good enough.
Yeah, it's strange. It seems like it's exactly the type of incident an AV should be vastly superior to a human driver. Especially strange if the car didn't try to brake at all.
I think we can rule out a human driver having avoided this death.
The victim is effectively invisible in the shadows against the lights of the city ahead. Not visible enough to slow, swerve, or to have been going slower in the first place
So, that just leaves why the autonomy failed to spot them. For which all the data is available, and so and objective assessment can be made.
What it does highlight is that a human riding along as backup is irrelevant. Even if they had been paying attention (and they weren't) there is no way they could have reacted fast enough to take over and avoid the incident
That video is horrible. It's clear that not much is visible on screen until the last moments. We've all struggled taking photos at night of things we can see well enough, so maybe the human eye may have seen more. What I would want to know is whether Volvo City Safety automatic braking system was active or bypassed, this seems a possible situation where it's infra red radar may to have been able to get at least some brake input and so reduce the speed at impact. It would be very bad if one safety system has been bypassed and overridden with something not as good.
I don't think we can rule out the question of whether a human driver would have made a difference. I read somewhere that the car made no attempt to brake. The pedestrian was visible for at least a second before impact so if the driver was paying close attention then they possibly would have been able to scrub off a few mph which might have been enough to make a difference.
The key questions are why did the sensors not pick her up and why was there no attempt to brake whatsoever.
Also, is there any point in having a human supervisor?
So no worse than a human but as you say the reasons will be analysed and technology improved. The same doesn't happen each time a human driver kills someone. Most of the time that same driver can drive again with no changes made to them and there is no call to stop all humans driving.
Be good to know where the nay sayers line is - taking UK we have ~1700 deaths a year. What would be an acceptable number if we had 100% AV ?
It would only take one or two anti-cyclist activists/pranksters to procure a bike then ride at 3mph in front of growing train of self-driving cars. This would spark a full-out social-media campaign against cyclists on the road.
Paranoid? Yes and no.
The answer may be to reduce car speeds in urban areas to a level that will not kill.
If an industrial environment had heavy machines whizzing past unprotected humans at close range the way they do on the road, it would be shut down.
However I suspect industry lobbying will result in excluding pedestrians and cyclists from the roads.
Yes paranoid.
@Malvern Rider - They're still cars. You know, with a steering wheel and pedals.
It does seem to have been an ideal situation that would have shown how good the AI system was. Instead, it totally failed and was no better then a human driver.
It was a clear road, dry, not much traffic (if any) around. If the system can't cope with that, it's not unreasonable for people to question how it will work in the rain, with traffic etc.
The video is amazing, I’m not sure a human could have spotted her and stopped in time.
The quality of that video is somewhat dubious though. This is a video from a local going down the same road. Its dark but nowhere close to what that video shows. Looks a crap dynamic range on the camera.
The real point there is how many drivers would have seen that?
Not sure that is the point.
From the footage the cyclist was already well across the road, and directly in front of the car. They didn't just leap out.
Surely the car isn't just relying on cameras that only detect light!? There should be systems that detect objects in all weather and light conditions. What about fog, heavy rain, driving snow?
If that is the case I can't see how they can safely work - I blame the technology in this case. If the tech is no better than the human what is the point of it?
The tech is coming. Its already safer than the average driver but still makes mistakes. this will improve.
Given that every autonomous car to date also has a driver then all we've proved so far is that the combination of a driver and automated car is maybe safer. We have no proof whatsoever that automated cars on their own do better than drivers. Google, Uber and co aren't clear about how often their drivers intervene - and I don't believe them anyhow because they've lied and decieved about so many other things.
What I would want to know is whether Volvo City Safety automatic braking system was active or bypassed
Afaik that only operates at low speeds so wouldn't have been active at this point anyway
and I don’t believe them anyhow because they’ve lied and decieved about so many other things.
You are adopting full conspiracy theory on this one, they could publish everything and you would still be unhappy and claim it was false.
However I suspect industry lobbying will result in excluding pedestrians and cyclists from the roads.
Go to any major city and you will have arterial roads that are fenced and walled in, do not have crossings and footpaths, at 8 lanes wide that one probably should have been as the time to cross without a refuge is inherently dangerous. It makes complete sense to have sections of road that are there to flow and routes for pedestrians and cyclists that also allow them to flow uninterrupted.
The quality of that video is somewhat dubious though. This is a video from a local going down the same road. Its dark but nowhere close to what that video shows. Looks a crap dynamic range on the camera.
I agree. Given the speed with which news organisations were told about the victim's marijuana convictions, there almost seems to be some kind of effort to move the debate to 'she was jaywalking, nothing could be done about it, carry on' as quickly as possible.
Even if it was as the official video shows, the LIDAR should have detected her and her direction of travel. There is zero reaction at all from the vehicle, not even at the last moment. It's a massive failure.
If conditions were like the comparison video shows, a human driver could reasonably be expected to at least start braking much earlier, perhaps changing the outcome to injury rather than death.
If conditions were like the comparison video shows, a human driver could reasonably be expected to at least start braking much earlier, perhaps changing the outcome to injury rather than death.
And yet there was a human driver in that car. I guess his degree of culpability depends on what grade of autonomy the car was supposed to be operating at and the relevant laws in that state.