I took a Boris bike through London once and the cyclists were definitely the scariest things on the road for me. (There was a very nice red bus driver though.)
The Dutch rule is basically, that the driver is liable if an accident happens, even if it's the cyclist's fault (unless you can prove that the cyclist deliberately caused the accident).
So there are not a lot of rules on what you should do. And this kind of passing is required often enough on narrow streets, that usually cars don't swerve if there is more space.
So that means that those self driving cars have to have a per country profile on how to behave. So if you have to do that for every country in the EU then that will make developing self-driving cars quite a bit more expensive than the current system where you have a common market.
Self-driving cars should try to avoid collisions regardless of fault and should not adjust that behavior based on who the local liability laws will hold responsible.
As far as traffic regulations specifying behavior go, yes, they differ from country to country and self-driving cars do need to be able to follow them. That doesn't necessarily mean the cars need to learn to follow a bunch of weird rules; it may be more of a lobbying problem. "Standardize your rules so we can offer self-driving cars that are X times safer than human drivers in your country" isn't a terribly hard sell.
>Self-driving cars should try to avoid collisions regardless of fault and should not adjust that behavior based on who the local liability laws will hold responsible.
I would love to watch a show called "Driverless cars in India".
That depends. The goal of local drivers is to go from A to B in reasonable time. If extreme risk deduction interferes that, people will revolt.
Some of the rules in some countries regarding overtaking cyclists would create a complete mess in the Netherlands. Of course, in theory you could have one set of rules self driving cars and another for human drivers.
I'm very curious if you can have a significant percentage of self driving cars in a city center without causing complete gridlock. My guess is that once cyclists and pedestrian realize how careful self driving cars are, they will just take priority and bring all car traffic to a halt.
> Self-driving cars should try to avoid collisions regardless of fault and should not adjust that behavior based on who the local liability laws will hold responsible.
In this specific case the issue was that the car didn't handle what seems like an unremarkable case where the bicycle was moving normally.
That said, driverless technology that can't handle special cases like bicycles doing something unexpected is not ready for prime time. After reading Ghosn's comments it makes me wonder how he would feel about a kid running out into the street after a ball.
well self driving cars will likely be mandated to carry the equivalent of dash cams if not black boxes which record everything over a hopefully short cycle (like last 15 minutes surrounding an incident with space to save a few)
Nissan CEO... (when introducing it he told the) arrival of the technology could be delayed by pesky cyclists who, he said, "don't respect any rules usually."
Which to be fair is true, but I also don't agree on picking on cyclists as a driver (as a pedestrian, it's a different issue)
No it isn't. It's confirmation bias of a stereotype. Motorists often display contempt for the rules, but everyone goes on about cyclists doing it. An asshole in a car will be an asshole on a bike given the chance. So afaict, we have an asshole problem, not a cyclist or a motorist one.
> don't agree on picking on cyclists as a driver
If you did, would you do it? Because "picking on cyclists" with the help of a two-tonne vehicle is what gets people killed.
However some cyclists think that cars (and trucks) have 360 degree field of vision and unlimited braking potential. That causes a lot of accidents as well.
Ah, OK, sorry. For many people, "picking on" means close passes and other illegal forms of intimidation. Like "picking on" someone at school means bullying.
Never heard it used that maliciously. "Picking on" always ranged from playful flirting to mild intimidation, or 'button-pushing'. Any actual violence, or grave threat of such, is never called by that term. That would be considered making light of the situation without doubt imho. Just some perspective, I have lived all over north america.
As as example of contempt for rules, I regularly cycle, and I am routinely overtaken illegally by motorists. In the UK, it is illegal to cross a solid center line in your lane to overtake a bicycle traveling faster than 10mph [1]. 10mph is very slow for a cyclist. The only time I'm traveling that slow is up the steepest hills, which generally don't have center lines. I estimate about 1 in 100 motorists actually obey the law. The only time I make any non-trivial journey without seeing illegal motorist behavior is when I'm riding exclusively in rural areas with no center lines.
Of course, not all of those illegal overtakes are actually dangerous, but motorists often insist on a zero tolerance attitude to cyclist law enforcement, so I think they should be held to the same standard.
I'm not sure why you're quoting the bit about cyclists not respecting rules, when in this example the cyclist is riding close to the left hand side of the road, and it's the car and the driver who break English law by passing too close.
If you're a car driver and you wonder why cyclists assert themselves by taking the lane: this is why. It's correct lane control, and it's something cyclists are supposed to do to prevent drivers from being dangerous.
The clue is in the name, it's a mag about cycling. It's run by Carlton Reid, a well-known expert on cycling - his most famous book is "Roads were not made for cars", a history of the intertwined histories of cars and bicycles and how motor transport came to dominate our roads almost to the exclusion of all others. So I'm not sure it's a "bias" so much as "what he thinks about cars and bikes".
> More interestingly
No, that's a minor point. The important thing is that cyclists get killed in close-pass situations, and those that don't are often put off cycling permanently. Along with the Uber car jumping a red light in SF, and Uber basically saying "it doesn't matter", this shows a worrying tendency for self-driving at any cost, including safety, which many of us were hoping would be a benefit of such technology.
I expect the incidence of these close passes will be drastically lower with self-driving cars than with manual ones. And it would take just one update to the software to fix ALL of them, as opposed to years of lobbying and re-educating stubborn drivers.
You're right cyclists get hurt or killed in close-pass situations (by humans).
It seems here the navigation identified the cyclist and avoided it. It incorrectly didn't follow the legal clearance but nobody got hurt (because the avoidance happened correctly it seems). It's an unsafe situation but the car did not ignore the cyclist.
The chance for someone to get injured or an accident happening seemed to be much higher in the Uber case.
I assume raverbashing's point was that a LHD system might have certain programming in place to handle passing a cyclist on the right side that behaves differently when the cyclist is on the other side of the road.
The car clearly identified the bicyclist, so this just seems like poor training/rules, which is easy enough to fix.
I'm more interested in the second video from the BBC where it couldn't identify the maintenance vehicle. If the driver hadn't intervened, would it really have plowed straight into it? I seems like even a current, non-self driving car, would have intervened with automatic braking (or slowing down with auto-cruise control) in that situation.
Heh. Yea, passing too closely is what many humans do.
Speaking of potentially murdering cyclists, since they already got all those sensors, they could prevent drivers and passengers from "dooring" cyclists.
The "dutch reach" (opening the door with your other hand which automatically makes you look over your shoulder) is unfortunately not commonly taught in driving schools.
"The "dutch reach" (opening the door with your other hand which automatically makes you look over your shoulder) is unfortunately not commonly taught in driving schools."
Are you sure? I read bycicles are really hard to detect. And I thought a bigger radar signature, pumping up some confidence value, might influence the behaviour too.
Well remove cyclists from streets. Its as simple as that. Remove humans from streets. Also remove all cars from the streets that need a driver. It will happen sooner or later. Humans will eventually be the single point of failure.
Swerving is good way to get yourself killed. If you find yourself swerving to the center of the road without making sure there is no traffic trying to overtake at the moment, then you have to figure out what to do to avoid that.
Relying on each an every driver to leave enough space just in case you need to swerve is a losing proposition.
a cyclist should be treated the same as any other vehicle on the road. they might need to swerve to avoid an obstacle while the car is overtaking, just as a car may need to swerve while another car is overtaking. when a safety margin is possible, as it was here, it costs nothing to apply one.
I don't know where you live that you have that much space. My personal rule is don't swerve. If you don't have time to signal that you are going to move, or look to make sure other traffic is not in the way, then either brake hard or continue straight (with the obvious exception if not swerving causes you to hit another person).
There are just too many roads where you can't swerve without causing a head on collision. If you make it a reflex to swerve, then you will also do it when there is no space for other traffic to avoid you.
You still have to account for a cyclist moving due to obstructions in the road or even the wind. I know I've had some close calls in gusty winds with people passing too close.
Heck, gusts once caused me to completely switch the side of the road in a fairly short time. Luckily no cars anywhere in that moment, but windy weather can be scary as a cyclist.
Where cyclists ride on the road is often filled with debris kicked up by car tires. So it is not a rare occurance that they will need to swerve to avoid a rock or stick, especially given the very narrow (23-25mm) tires commonly found on road bikes.
This is perhaps the real risk of AI. In the first wave, the industry is dominated by a wish to make a great product. In the second wave, the bean counters walk in and start to optimize everything. And then we start getting the behavior from Fight Club.
I don't see any way out of this loop except for penalizing bad business behavior (on an individual level) heavily while protecting engineers who fight them.
I fully agree. It would solve a lot of problems we currently face in technology, but there are two issues that I see:
1) Doctors, surgeons, engineers (actual professional engineers, not people with comp sci degrees who call themselves engineers), lawyers, etc. are called professionals because they have professional colleges of their peers that oversee things like ethics and can revoke licenses, etc. They're self-regulating.
This works because for things like medicine and engineering and law, there is a large barrier to entry. You have to go to a recognized school and pass an ethics/comprehensive exam. This isn't so with programming. There's almost no barrier to entry.
Any kid can open up a tutorial on MySQL and call themselves a back-end engineer. Unless the software industry is willing to require formal qualifications on all their hires, this will remain a problem.
2) One of the great things about software is the democratic aspect. Software is great precisely because any random kid can open up a tutorial and notepad.exe and start writing code. To rectify problem #1, you'd to a certain extent kill the democratic nature of learning which is a large part of why many people are in the industry today. I also think that a large portion of people who work in software would be highly resistant to this sort of thing, to say nothing of the difficulty of enforcing ethical behaviour if you can't see what's going on due to proprietary licensing and obfuscation.
58 comments
[ 6.6 ms ] story [ 195 ms ] threadThe real real big problem is that AI will learn to exploit all our flaws and biases.
AI will evolve to be supernaturally intelligent and supernaturally psychopathic at the same time.
Red light isn't a suggestion, the side walk isn't s bike lane and the cross walk isn't there for you to make a u turn.
Do you have any references for this?
So there are not a lot of rules on what you should do. And this kind of passing is required often enough on narrow streets, that usually cars don't swerve if there is more space.
So that means that those self driving cars have to have a per country profile on how to behave. So if you have to do that for every country in the EU then that will make developing self-driving cars quite a bit more expensive than the current system where you have a common market.
As far as traffic regulations specifying behavior go, yes, they differ from country to country and self-driving cars do need to be able to follow them. That doesn't necessarily mean the cars need to learn to follow a bunch of weird rules; it may be more of a lobbying problem. "Standardize your rules so we can offer self-driving cars that are X times safer than human drivers in your country" isn't a terribly hard sell.
I would love to watch a show called "Driverless cars in India".
Some of the rules in some countries regarding overtaking cyclists would create a complete mess in the Netherlands. Of course, in theory you could have one set of rules self driving cars and another for human drivers.
I'm very curious if you can have a significant percentage of self driving cars in a city center without causing complete gridlock. My guess is that once cyclists and pedestrian realize how careful self driving cars are, they will just take priority and bring all car traffic to a halt.
In this specific case the issue was that the car didn't handle what seems like an unremarkable case where the bicycle was moving normally.
That said, driverless technology that can't handle special cases like bicycles doing something unexpected is not ready for prime time. After reading Ghosn's comments it makes me wonder how he would feel about a kid running out into the street after a ball.
More interestingly, this was a LHD car in an UK road, might have something to do with it (though the navigation system seemed to be aware of it).
Which to be fair is true, but I also don't agree on picking on cyclists as a driver (as a pedestrian, it's a different issue)
No it isn't. It's confirmation bias of a stereotype. Motorists often display contempt for the rules, but everyone goes on about cyclists doing it. An asshole in a car will be an asshole on a bike given the chance. So afaict, we have an asshole problem, not a cyclist or a motorist one.
> don't agree on picking on cyclists as a driver
If you did, would you do it? Because "picking on cyclists" with the help of a two-tonne vehicle is what gets people killed.
However some cyclists think that cars (and trucks) have 360 degree field of vision and unlimited braking potential. That causes a lot of accidents as well.
> If you did, would you do it?
I think you're misreading my comment. "Picking on" in this case means blaming. (def. 32 http://www.dictionary.com/browse/picking)
Of course, not all of those illegal overtakes are actually dangerous, but motorists often insist on a zero tolerance attitude to cyclist law enforcement, so I think they should be held to the same standard.
[1] http://www.legislation.gov.uk/uksi/2002/3113/regulation/26/m... paragraph 6 section f
If you're a car driver and you wonder why cyclists assert themselves by taking the lane: this is why. It's correct lane control, and it's something cyclists are supposed to do to prevent drivers from being dangerous.
The clue is in the name, it's a mag about cycling. It's run by Carlton Reid, a well-known expert on cycling - his most famous book is "Roads were not made for cars", a history of the intertwined histories of cars and bicycles and how motor transport came to dominate our roads almost to the exclusion of all others. So I'm not sure it's a "bias" so much as "what he thinks about cars and bikes".
> More interestingly
No, that's a minor point. The important thing is that cyclists get killed in close-pass situations, and those that don't are often put off cycling permanently. Along with the Uber car jumping a red light in SF, and Uber basically saying "it doesn't matter", this shows a worrying tendency for self-driving at any cost, including safety, which many of us were hoping would be a benefit of such technology.
You're right cyclists get hurt or killed in close-pass situations (by humans).
It seems here the navigation identified the cyclist and avoided it. It incorrectly didn't follow the legal clearance but nobody got hurt (because the avoidance happened correctly it seems). It's an unsafe situation but the car did not ignore the cyclist.
The chance for someone to get injured or an accident happening seemed to be much higher in the Uber case.
I'm more interested in the second video from the BBC where it couldn't identify the maintenance vehicle. If the driver hadn't intervened, would it really have plowed straight into it? I seems like even a current, non-self driving car, would have intervened with automatic braking (or slowing down with auto-cruise control) in that situation.
Heh. Yea, passing too closely is what many humans do.
Speaking of potentially murdering cyclists, since they already got all those sensors, they could prevent drivers and passengers from "dooring" cyclists.
The "dutch reach" (opening the door with your other hand which automatically makes you look over your shoulder) is unfortunately not commonly taught in driving schools.
Excellent idea, going to use that one.
>Remove humans
Relying on each an every driver to leave enough space just in case you need to swerve is a losing proposition.
There are just too many roads where you can't swerve without causing a head on collision. If you make it a reflex to swerve, then you will also do it when there is no space for other traffic to avoid you.
Maybe a self driving car could react quicker in a situation like this, but it's still good to leave cyclists some space.
Your self driving car shouldn't be getting points on your licence for you.
I don't see any way out of this loop except for penalizing bad business behavior (on an individual level) heavily while protecting engineers who fight them.
Programmers should have a similar guild/organization that medical doctors do, and people who abuse the ethics should be barred from practicing.
1) Doctors, surgeons, engineers (actual professional engineers, not people with comp sci degrees who call themselves engineers), lawyers, etc. are called professionals because they have professional colleges of their peers that oversee things like ethics and can revoke licenses, etc. They're self-regulating.
This works because for things like medicine and engineering and law, there is a large barrier to entry. You have to go to a recognized school and pass an ethics/comprehensive exam. This isn't so with programming. There's almost no barrier to entry.
Any kid can open up a tutorial on MySQL and call themselves a back-end engineer. Unless the software industry is willing to require formal qualifications on all their hires, this will remain a problem.
2) One of the great things about software is the democratic aspect. Software is great precisely because any random kid can open up a tutorial and notepad.exe and start writing code. To rectify problem #1, you'd to a certain extent kill the democratic nature of learning which is a large part of why many people are in the industry today. I also think that a large portion of people who work in software would be highly resistant to this sort of thing, to say nothing of the difficulty of enforcing ethical behaviour if you can't see what's going on due to proprietary licensing and obfuscation.
Anyone could have called themselves a doctor-analogue 100 years after the invention of "medicine", too.