Good summary on the progress being made on Radio 4 this evening… personally I can’t wait 🙂

Radio 4

Volvo autonomous tech available today

Live trials in the UK

Self-shuttling? Yes please! 😉

How are they bike related?

safer innit.

safer innit.

exactly. ‘not safe’ being the #1 reason given by non-cyclists for being non-cyclists.

Interesting take on driverless cars from John Adams of “risk compensation” fame.

Basically saying that as these cars would need to be failsafe when interacting with cyclists and pedestrians they would be more or less halted in congested urban areas once other road users learned they no longer had to defer to cars.

The Driverless Car Revolution – Amazon Review

that sounds disastrous….
the tech’ll keep getting better.
once these things really get themselves sorted out I can see them talking to each other and forming little trains going to common destinations so you don’t end up with loads of individual units all the time. safer, more efficient.

I think it’s the ethical discussion that’s going to delay the arrival of autonomous cars.

Consider the following situation; you are riding along in your autonomous vehicle on a busy road with pedestrians on both pavements. There’s a truck coming in the opposite direction travelling at the speed limit.

Just as the truck begins to pass you a child runs out in front of your vehicle.

The ‘car’ has three options; swerve left and kill you (the occupant) in a collision with the truck or swerve right and kill a pedestrian on the other pavement or ‘it’ can carry straight on and kill the child.

Scenario’s like this are giving car manufacturers and their legal teams sleepless nights – I hope 😯

Scenario’s like this are giving car manufacturers and their legal teams sleepless nights – I hope

Why? There will be accidents. The advantage of the driverless car is that it will have cold hard data to show what happened.

Or are you Will Smith in iRobot?

The ‘car’ has three options; swerve left and kill you (the occupant) in a collision with the truck or swerve right and kill a pedestrian on the other pavement or ‘it’ can carry straight on and kill the child.

Scenario’s like this are giving car manufacturers and their legal teams sleepless nights – I hope

You’d hope it was, but if we’re talking about Google here, there would presumably be some quickfire cost-benefit analysis based on the web browsing history of the car passengers vs the child?

Also, I would hope that the driverless car’s collision avoidance algorithm is based around appropriate speed to avoid hazards and bringing the car to a stop in a straight line, rather than “swerve wildly while crossing your digits”.

I can’t wait for driverless cars. Imagine being in a queue at a traffic light and when it turns green, all the cars move off at the same time, instead of the dopey caterpillar we get now. Bliss.

They’ll be brilliant when the rules are standardised and we all know where we are but can you imagine the consequences of lots of differing software being used by different manufacturers not to mention the interaction with random cyclists,pedestrians and human drivers?

There’s a lot of work to be done before they can realistically be let loose on the road.

There’s a lot of work to be done before they can realistically be let loose on the road.

We let humans with different standards and conflicting wetware loose on the roads every day.

instead of the dopey caterpillar we get now. Bliss.

some caterpillars form trains where the one at the back walks along the top of the other ones so going a twice the speed until he gets to the front then the cycle repeats itself. Imagine that with cars!

Just as the truck begins to pass you a child runs out in front of your vehicle.

Aha, you need self driving chairs for pedestrians.

In the scenario above ^^^ , what does the human driver do that’s so much better than whatever option the autonomous car chooses?

Any mitigating factors, such as reducing speed in advance, having recognised the potential, could be done more reliably by the autonomous car, Shirley?

torsoinalake – Member

We let humans with different standards and conflicting wetware loose on the roads every day.

with, quite frankly, catastrophic results.

Indeed, whenever the ethical “ZOMG who would the machine choose to kill?” questions come up, normally with extremely convoluted and artificially constructed scenarios, the default response should be “what would the human do?”

Trying to work out if you could improve on that is the next step, whether that be by mitigating the risk in advance, or by an alternative course of action is all part of the fun, but you’re rarely end up with a situation where the machine would then choose a worse course of action, and if you do it’s because you’ve programmed it to arrive at that conclusion, which takes you right back to “what would the human do?”

In the example above the machine has the exact same options as the human, but with the added benefit of better sensory awareness, quicker reactions, and also possibly being able to calculate an alternative course of action, and at the very worst it will have better physical control and response when breaking and manoeuvring.

I’m not saying it can’t go wrong, but chances are it will go wrong less than the human would.

I also don’t think the tech is quite there yet, but it is improving very rapidly indeed!

You also need to think big picture on this, for every single unavoidable accidental death that would still occur with automated vehicles, you’d have avoided hundreds, if not thousands of minor bumps, fender-benders, whiplash injuries, minor collisions, and some major collisions that would have happened on a daily basis without automation. Seems like a worthy pursuit to me.

M Town

So a driverless car is travelling along a mountain pass, there’s a coach approaching too far over the white lines to avoid a collision without taking you and the passenger over the edge of the cliff. The collision will potentially end in a greater number of fatalities, how should it react?

how would a human react?

that’s the starting point at least.

but you’re missing key elements to this scenario, is it sufficiently far enough down the road (excuse the pun) that the coach is also driverless?

Are the two vehicles in communication and able to warn each other and stop both vehicles?

is there traffic behind the car? could it stop? could it reverse? how far away is the coach? how long has it been in visual/sensory range? could other steps have been taken earlier to mitigate this?

Normal concocted scenario problem of not enough information. And if you start being tighter with your definition to try and build a particular scenario you’re now into the realms of describing a scenario that is specifically trying to be fatal, which means you’re now designing a scenario to beat a system, not a system to deal with scenarios, and you will wind up with whatever answer you want as you’re building it.

You end up back at “what would the human do?” and then “can you do better” if you’ve artificially fixed the parameters so that you can’t then you’re at worst left with “do as the human would” and at best, opening up thoughts about developments and potential improvements you could make.

Indeed, whenever the ethical “ZOMG who would the machine choose to kill?” questions come up, normally with extremely convoluted and artificially constructed scenarios, the default response should be “what would the human do?”

Not really though.

In industry it’s common for numbers like 1:100,000 to be the acceptable chance of killing an individual worker on site, but the chances of killing someone off-site have to be 1:1,000,000 (i.e. 10x less). So you as a driver in the car have made your decision to be there and accept that risk, so the hypothetical situation where the choice is kill the driver or kill the pedestrian, the car would have to kill the driver unless it judged the chance of the pedestrian surviving to be 10x higher (which is unlikley given airbags etc).

So I’d expect them to always default to stuffing it in the hedge or infront of the HGV (killing the driver and injuring no one else).

For balance, that 1:100,000 isn’t random, it’s judged to be 10x safer than driving which is assumed to be the most dangerous day to day activity a normal person will do, hence stories like your 10x more likely to die in a car crash on the way to work than working on a oil refinery.

how should it react?

Just shout “Take over KITT”, and hit the turbo boost button.

Indeed, whenever the ethical “ZOMG who would the machine choose to kill?” questions come up, normally with extremely convoluted and artificially constructed scenarios, the default response should be “what would the human do?”

Trouble is, humans tend to ‘over self-preserve’ at the expense of others. The classic example is overtaking a cyclist at relatively low speed, car comes the other way, the instinct is to pull left even though that is much more likely to kill the cyclist than a head-on is to kill one of the drivers.

You can argue the driverless car wouldn’t have got in that pickle in the first place, of course, but there are some interesting equations about likelihood of death or serious injury to vulnerable road users vs car users where the ‘better’ option is accepting an injury to the car users.

@tinas and martinhutch

I get that, and those decisions and discussion will certainly be being made and had within these companies, but ultimately it’s still humans programming and deciding that. The scary thought is that someone somewhere WILL have to make the decision about what to do in one of those lose-lose situations, it might be that they default to prioritise the occupant and sell that as a safety feature to owners, I hope not, but I can imagine the adverts already….

What is clear is that we’re all making assumptions based on incomplete data (we have to sadly, as that’s the very nature of trying to build systems like this), and concocted scenarios like the above are never complete enough.

The argument about would a driverless car have got into such a situation in the first place is not one to be treated lightly, identifying scenarios like the above is as much about developing ways of avoiding getting into them as it is about deciding what to do once they happen.

you’re thinking very much with a human bias, you’re assuming that it would ‘stuff it into the lorry’ and kill the driver, when in fact under machine control it might be able to either avoid the situation entirely in the first place, or if not, come to a more controlled stop, or steer the car more precisely than the human could. The machine might be able to make a better decision than ‘stuff it into the lorry’

The obvious utopia is if all vehicles were driverless, or at least in communication with each other as a great many of these issues would disappear, the most dangerous and difficult phase will be the transition phase with both automated and manual vehicles togther.

amedias – Member

someone somewhere WILL have to make the decision about what to do in one of those lose-lose situations,

ok, here we go, the car industry can have this for free:

1) stick to the speed limit.
2) brake hard if there’s an unavoidable problem.

that should see most (if not all) car/person collisions brought significantly below a survivable 20mph.

that wasn’t too hard was it? but it’s still WAAAAY better than humans can manage.

The driverless car would always do the safest thing and more importantly being driving at the correct speed for the condition/circumstance.

If that means slowing to an almost stop because it is not clear what is coming round a bend than that is what it would do. Unlike a human that would just carry on at 30mph and then be shocked when they crash into a coach.

You’ve kinda missed my point guys – in my theoretical scenario whatever the car chooses to do is based upon the way it was designed and how it’s software is configured.

That means that the Vehicle Manufacturer is to blame for the outcome of the unavoidable collision. Whoever dies has died because in those circumstances the Vehicle Manufacturer decided they should die.

Vehicle Manufacturers and their Legal Teams deserve a few sleepless nights if they can figure that one out and protect themselves from the legal fall-out of that particular no-win scenario!

That means that the Vehicle Manufacturer is to blame for the outcome of the unavoidable collision

No they aren’t. The child is.

It seems far too easy to get bogged down in such hypothetical situations which will occur on an extremely rare basis. It’s ridiculous to let such issues delay the introduction of driverless cars given that on any reasonable cost benefit analysis even choosing the worst option in order to allow their use a day earlier would result in saving lives.

@jag

And you’ve kind of missed our point, that its not necessarily a no-win scenario as it’s not one that driverless cars should get themselves in to.

That means that the Vehicle Manufacturer is to blame for the outcome of the unavoidable collision

The liability aspect of it is another matter entirely! How exactly do you become liable for something that is unavoidable? Would the human be to blame in the above situation if it was truly unavoidable? Not to mention that as we already said, if designed properly it would hopefully become avoidable.

The system would never make the ‘choice’ to kill the driver, it would make the choice to take all possible action to avoid such a situation prior to it happening, and then take all possible action to mitigate the impact if it did find itself in such a situation, any deaths wouldn’t be a choice to kill, it would be a through a failure of all possible options to prevent it, and those options would be more numerous and more reliable than a human could manage. It’s the exact opposite of choosing to kill, it’s taking every possible step (more than a human could) to avoid a death, but failing to do so due to external influences that cannot be controlled.

+ 1 million for what aracer said.

These imagined scenarios are distractions really, they’re hypothetical, incomplete, extremely rare, and overwhelmed by the massive positives on offer.

It’s ridiculous to let such issues delay the introduction of driverless cars given that on any reasonable cost benefit analysis even choosing the worst option in order to allow their use a day earlier would result in saving lives.

What’s even better is that we will become the most dangerous road users. 🙂

Basically saying that as these cars would need to be failsafe when interacting with cyclists and pedestrians they would be more or less halted in congested urban areas once other road users learned they no longer had to defer to cars.

Is this really an issue though?

God forbid that people moving around under their own steam are given priority over people sat on comfy seats in air-conditioned boxes that they don’t even have to control.

There’s a lot of work to be done before they can realistically be let loose on the road.

Self-driving cars ARE loose on the road – in California. Collating data so there’s empirical data to be able to judge risk. IIRC there have been very few collisions and they have usually been human error, rather than an error in the self-driving tech.
You’ll also note the amount of autonomous tech available in Volvos in the UK… today.

The reason I put up the links is to show the number of live trials already underway in the UK ie. the ‘work that has to be done’ is already being done… autonomous tech is not a sci-fi theoretical scenario, it’s here, out on the roads…

And the whole debate about hypothetical decision scenarios misses a major point, which is that as self-driving cars will be driving within the speed limit and to the conditions, and not fiddling with their mobile phones/smoking/eating/falling asleep/not looking where they’re going etc, the biggest reduction will be in the base number of situations where there’s a risk of a collision in the first place.

Let’s assume in the UK that of all the times something could go wrong because of poor skill on behalf of the driver, that they do go wrong 1% of the time…
Let’s assume there’s 1,000,000 of these potential situations every day.
That means 10,000 collisions
With self-driving tech, let’s assume the lack of ability to stop the risky situation going wrong is no better than a human ie: 1% of all risky situations.
But because the car is tech-operated rather than operated by a stupid, emotional, angry, distracted human, the number of risky situations barely arise in the first place. ie: the car’s not speeding. Let’s assume only 1,000

1,000 risky situations x 1% = 10 collisions.

I know I’m using hypothetical figures here but the debate isn’t about how the autonomous tech performs in a collision-avoidance situation, it’s that it avoids the collision-avoidance situation in the first place, likely to lead to a dramatic fall in collsions and the hypothetical ‘how would a self-driving car cope with the situation better than a human’ question is null and void

Other aspects that often gets missed or not discussed is will there be any other health/wellbeing benefits to autonomous travel.

All that commuting and travelling time will to some extent be reclaimed. I wonder whether this will lead to any net reduction in stress and tiredness levels once people are freed from the task of having to concentrate on operating a vehicle, will it lead to more social interaction between passengers, more time to read, more time to think, or (sadly) more time to work. Will it reduce instances of road rage and driving induced stress?

Will there be any improvement in pollution and air quality if vehicles are able to operate at their most efficient? (whether powered by tiny explosions or tiny lightening bolts) And will the better traffic flows reduce congestion and travel times even further?

Will it reduce wear and tear and ultimately servicing and replacement costs and will this have any noticeable environmental impact?

It should reduce insurance bills, servicing costs, and fuel costs, will the financial burden of private* car ownership decrease?

Autonomous travel is a lot more complex than it seems at first glance and the benefits could be much broader reaching than we think.

*Private ownership I think will be here to stay for a long while, although we may see more of a shift towards shared and pooled ownership and usage as time goes on. Once the ‘driving’ is taken out of driving it largely becomes about whether a suitable vehicle is available and personal choice gets reduced to deciding what environment you want to sit in, as speed, handling and driving experience become irrelevancies.

This is all still very urban based, and whatever happens private (and manually controlled vehicles?) will still be in use on certain sites and in many rural locations.

Interesting and exciting times ahead….

You’ll also note the amount of autonomous tech available in Volvos in the UK… today.

Is that the Volvo that couldn’t have an accident and promptly ploughed stright into the demostration car in front of it?

The real worry to me is these car manufacturers do the ultimate cop-out – hand the car back to the ‘driver’ if it can’t decide what to do in these kind of scenarios. They’d probably have to put down their latte, magazine and ipad first mind…

Is that the Volvo that couldn’t have an accident and promptly ploughed stright into the demostration car in front of it?

no, read the link I put up in my OP – nine separate autonomous technologies available today.

Volvo autonomous tech

Do you think they’d be in-market if they didn’t work well enough?

How many news stories have you seen recently talking about the massive increase in Volvos crashing?

Yes and the possible benefits accumulate as you get more and more autonomous cars on the road.

Once they reach a criitcal mass you can network the cars together, traffic flow then greatly improves as all the cars know what the other cars want to do.

You could make cars about half the weight too, you don’t need 10 airbags, side impact protection and huge metal crumple zones if the chance of a collision is tiny

Do you think they’d be in-market if they didn’t work well enough?

Adaptive crusie control will happily keep you in a lorry’s blind spot for extended periods of time – it’s very easy to fall into a false sense of security and lose track of your positioning. 100% driver awareness is required at all times.

However, it’s fantastic in low-visibility and can see further than I can in very heavy rain. The tech has a long way to go.

They may well be in test but that’s still a long way from jumping into my autopod giving “SIRI” the destination (WORK/Boyfriend/Girlfriend/Mate/Pub/FOD/OTHER) strapping in and going.

I’m sure these things are going to revolutionize the transport landscape. We could end up with tiny personal transportation pods for short journeys and 4+ seat versions for the family, they may be pooled in communities and are very likely to be fully electric with only longer journeys being undertaken by IC engines.

For all that to happen, however, human beings will need to reset their love of cars and personal transport and the car as status symbol will have to be broken down. There is some thinking that that’s already happening in cities and with the kids but who knows.

it’s small steps and we’re at the beginning. All I know is its an opportunity for megacorp to sell us more stuff.

late to the thread but – Jag’s scenario is rubbish

automated car in a town – 28mph maximum
automated car spots pedestrians – starts to slow, now to 25ish max
kid runs out – car stops quicker than human because not distracted and not speeding… kid more likely to survive.

Automated cars are not going to mount a pavement, or drive on the wrong side of the road, those are poor choices for a human and machines don’t do those things

all this stuff about automated cars really feels like FUD from the motor manufacturers lobby to dissuade us from travelling in automated cars; they are the primary business threat to a company like BMW who make ‘the ultimate driving machine’