Driverless Nissan nearly takes out cyclist

[growingvegetables]

... it is inevitable that over time they will network and create a "hive mind" to refine the AI further. ... after the first few big crashes there will be pressure to amend the programming "so that it can never happen again". Makes sense to build a networking "hive mind" ability in from day one rather than constantly needing to go to the garage for updates.

I admire your optimism. What suggests your "hive mind" will be positive for vulnerable road users? Given the Nissan incident, and the predilections of its management, the new "hive mind" could be .......... extremely negative?

 

[Drago]

The thing that really worries me about these Nissan autonomous cars.... the electrics will be supplied by Renault. Names such as Omega, Sabatier and Mag, are associated with products of the highest quality, precision and dependability. Renault does not which, in a product that regulates it's own behaviour with regards to the safety of those around it, is rather worrying.

I mean, if you had heart surgery would you want the surgeon to use a scalpel made by Renault?

 

[srw]

Actually, the Tesla "autopilot" and aircraft automation combine the worst of both worlds: most of the time the human operator does nothing, but still needs to maintain complete situational awareness, so that when the automation says "I give up, get me out of here" they can seamlessly take over. That's something humans aren't really good at. It takes time, even if you've been paying attention, to get up to speed so as to decide what the appropriate actions are. Of course, after hours of inactivity, attention is something that is likely to be at a premium.... (which takes us back to AF447).

One thing that occurred to me as I was driving back this lunchtime along the A3, M25 and M40 was quite how little active attention a normal driver pays most of the time. Driving is mostly done subconsciously rather than with alert and highly active conscious attention. I don't know enough about human psychology to know what that means for the times that the human driver needs to take active and conscious control.

Apart from anything else after the first few big crashes there will be pressure to amend the programming "so that it can never happen again".

But by then it will be too late. My observation is that the people who are pushing the technology (scientists, car companies, insurance companies) are alive to all the very knotty ethical problems they'll have to solve, and solutions will be built in at design stage. And if one of the constraints isn't make sure that the number of big crashes, especially ones involving children, is demonstrably considerably lower than with human drivers I'd be gobsmacked. If a car can cope with a completely unpredictable child it can certainly cope with a reasonably predictable adult on a bike or on foot.

 

[jarlrmai]

Some more possibilities with driverless cars

1. Governments will be forced to come up with some clear rules for passing cyclists because good luck writing a computer programme that tells a robot car to "give motorcyclists, cyclists and horse riders at least as much room as you would when overtaking a car." This option will be interesting in the US where laws vary per state.

2. Companies will be left to their own devices for how a car should react to a cyclist, now given the tech is new they will probably/hopefully opt for caution, god forbid they allow the settings to be user configurable.

3. Big companies decide cyclists are too difficult a problem for driverless cars and lobby for law changes to ban cyclists, using car accident reduction as a wedge.

I'm a programmer by trade and I see problems like the child on the road as possibly easier to solve, for one a child on the road is a rare unexpected event but a foreseeable one and second essentially the car performing an emergency stop is basically the best option this (you may get rear ended but this is infinity preferable to running over a child) , so "child in road = emergency stop." Now knowing its a child and not a dog is whole different kettle of logical and ethical problems.

A car cannot emergency stop every time it meets a cyclist, so it has to be able to adapt driving behaviour to cope with the situation and keep driving. So this might be more difficult for programmers.

 

[potsy]

. Big companies decide cyclists are too difficult a problem for driverless cars and lobby for law changes to ban cyclists, using car accident reduction as a wedge.

I still don't ever see these cars becoming reality on our roads, but if they do then I fear this could come true.

Much easier to get rid of the 'problem'.

 

[jarlrmai]

Does the phrase mean "leave as much room between the side your car and side of the bike as you would if you were going past a car" or does it mean "imagine there's a car and then give as much room as that."

It specifies nothing concrete so it becomes difficult to programme and actually have the programme match the law/advice, you need specifics to tell the computer what to do, perhaps the self driving car in the original video was told, when you go past a car leave 30 cm between you and the car then it it was told "give motorcyclists, cyclists and horse riders at least as much room as you would when overtaking a car." so it gave 30 cm

 

[KnackeredBike]

Does the phrase mean "leave as much room between the side your car and side of the bike as you would if you were going past a car" or does it mean "imagine there's a car and then give as much room as that."

It specifies nothing concrete so it becomes difficult to programme and actually have the programme match the law/advice, you need specifics to tell the computer what to do, perhaps the self driving car in the original video was told, when you go past a car leave 30 cm between you and the car then it it was told "give motorcyclists, cyclists and horse riders at least as much room as you would when overtaking a car." so it gave 30 cm

The rule around here is

If there is nothing oncoming overtake on the opposite side of the road.

If there is oncoming traffic leave at least a fag paper width when squeezing past.

If there is not a fag paper width drive as close to the back wheel of the cyclist as possible, do emergency stops whenever the cyclist slows so the driver behind knows how inconsiderate the cyclist is being. Ensure you floor it and swerve around the cyclist at the first opportunity. Then go home and write into the Daily Mail about how cyclists are clogging up your roads and not paying road tax.

This is why it's important self driving cars are networked, otherwise how will they contact the DM.

 

[jarlrmai]

The photo might just show how much space that specific driver gives a car when they over take.

 

[jarlrmai]

I am familiar with it are you saying that for sure that rule in the highway code means give a cyclist a cars width of space as if there were a car were the cyclist is.

Because you could argue that "as much room as you would when overtaking a car" could be interpreted differently depending on who reads it.

A computer needs unambiguous specific statements, not judgement calls or the judgement calls of the person programming it.

 

[jarlrmai]

They could of course be programmed to do that, but they will be likely programmed to match the law, but when the law has no specifics then how do you programme them?

So you need to amend the law if it's not specific, or make a specific rule for self driving cars which then conflicts with the rule for normal cars. Or rely on SDC companies to decide what's safe for cyclists.

 

[Smokin Joe]

Before driverless cars will be allowed for public use they will be tested to death and back to eliminate any risks as much as possible. The number of accidents they will be involved in will almost certainly be small enough to not even make a statistic.

 

[KnackeredBike]

Before driverless cars will be allowed for public use they will be tested to death and back to eliminate any risks as much as possible. The number of accidents they will be involved in will almost certainly be small enough to not even make a statistic.

Doesn't matter, any near miss or accident will be reported in a sensational style even if it is a tiny fraction of those caused by human drivers per km.

They are somewhat inevitable in the same way some pushbike accidents are inevitable: the roads are designed for cars and anyone else is using a resource for which they are as out place as a car driving down a pavement is.

 

[Smokin Joe]

Doesn't matter, any near miss or accident will be reported in a sensational style even if it is a tiny fraction of those caused by human drivers per km.

It will be very easy to publish comparative figures of before / after driverless to show a significant improvement in safety.

 

[KnackeredBike]

It will be very easy to publish comparative figures of before / after driverless to show a significant improvement in safety.

I'm not sure that numbers will convince the Daily "curse the EU and their ban on incandescent bulbs" Mail. Sensationalism sells papers.

My father in law still insists that incandescent bulbs are more efficient and that the reason why don't last as long is because the electric company send "too much power" to their house. It's difficult to argue with abject stupidity.

 

[srw]

I thought it was worth resuscitating the thread to post a link:


https://www.engadget.com/amp/2017/09/11/waymo-self-driving-car-simulator-intersection/

Real data is being tweaked by adding cyclists, pedestrians and so on, so that virtual models of driverless cars can learn how to cope.