Self-driving cars and the externalisation of danger (to other road users) - Thanks Mercedes Benz!

Thanks to @Velocio and this thread there is a discussion happening amongst some cycling and walking (yet to hear from the driving) orgs regarding a joint position on this issue.

So far there is a general consensus that these points need to be made:

1. A need to ensure these vehicles adequately allow for detection and avoidance of less predictable and more vulnerable road users like walkers and cyclists. Where avoidance is not possible the cars should be programmes to always protect people outside the cars over the users.

2. A need for clear and simple guidance to operate legal system that ensures the user is liable and responsible for any damage the car makes and that further legal action to negate that to any manufacturing issue would be the responsibility of the user or his insurance company and not an uninsured vulnerable road user or victim.

3. transition to such a system to be swift to minimise the potential for driven cars and driverless to use the same environment thereby adding an additional level of complexity. (Though this point may be beyond DFT control)

Let me know any other thoughts you may have

That was a Tesla with "steering assist" mode turned on which isn't exactly the same thing but I get your point. Tesla have already said they will immediately improve this feature. It was a camera only type deal and not with radar type systems that the full automated systems will hopefully have in place. Tragic story though

4. Cars must perform a sweet skid on request (forumengers are only getting older).

I think those are some great points. I can see now people trying to fight their way out of cases because it wasn't their fault, they weren't driving. There definitely needs to be some acceptance of responsibility for the car by the owner when it is bought/insured.
And yes as we said earlier, if you're sat in a six figure mercedes driving at urban speeds, you can be pretty sure that you're going to come out okay from most crashes. I agree that the priority in the rare cases of unavoidable collisions should be to protect external life over itself/its passenger.

I reckon Merc will arrange terms/conditions to lay blame on the registered owner operator of the vehicle, corporate manslaughter because there product will endanger/kill to protect the occupants of the vehicle, i don't see Merc taking that liability on.

Massive ethical debate and law to be decided on the driverless cars on the open road.

Random tale time.
About 10 years ago i rear ended a Merc S class estate.
By law, my fault, no contest.

The insurance company got back to the company i worked for (and carried on working for a few more years) they pointed out that i never stood a chance of not hitting the car, some other factors involved, but under ideal conditions i would still have hit the car.
The Merc had the latest active braking system.
I was taken for a test ride in one at a later date.

The Merc will out brake and out stop any other vehicle (bar another Merc with the same system)

So you have all these cars with fantastic computer controlled braking systems.
Then you have all the other vehicles that don't.
A lot of Mercs where getting rear ended.

So change that to driverless cars mixed in with other open road users and it will be messy.
Driverless cars in the far future when all other vehicles are driverless might work (barring software glitches) but for now expect more accidents.

Thats another good point.
It happens to me all the time cycling along roads where a pedestrian steps out without looking, is a few feet into the road, sees me coming, stops and steps back. Drivers can anticipate these kind of human response situations as where an automated car might step on the anchors as soon as someone steps foot off the kerb and I can see a lot of them getting rear ended in situations like this.

This is the article I was thinking of https://www.theguardian.com/technology/2016/aug/22/self-driving-cars-moral-dilemmas

“The main thing to keep in mind is that we have yet to encounter one of these problems,” he said. “In all of our journeys, we have never been in a situation where you have to pick between the baby stroller or the grandmother. Even if we did see a scenario like that, usually that would mean you made a mistake a couple of seconds earlier. And so as a moral software engineer coming into work in the office, if I want to save lives, my goal is to prevent us from getting in that situation, because that implies that we screwed up.

“It takes some of the intellectual intrigue out of the problem, but the answer is almost always ‘slam on the brakes’,” he added. “You’re much more confident about things directly in front of you, just because of how the system works, but also your control is much more precise by slamming on the brakes than trying to swerve into anything. So it would need to be a pretty extreme situation before that becomes anything other than the correct answer.”

Where avoidance is not possible the cars should be programmes to always protect people outside the cars over the users.

I'm not sure exactly what this means. If avoidance is not possible then how can the car protect people outside the car other than by (as @aggi pointed out) slamming on the brakes?

Given that self-driving cars are very modern and will obey speed-limits. The risk to the passengers will generally be minimal. So, to account for the rare occasions in which the choice is hit pedestrian vs. hit tree you could say "Where stopping distance exceeds free road space, the cars should be programmed to collide with inanimate objects rather than vulnerable road users."

I agree with this as a point of principle but, pragmatically, I think with this demand you may be creating a stumbling block where there isn't a need for one. As @Nick.Earthloop said, if self-driving cars/mode are safer for everyone else than manual driving then the main aim has to be the maximise uptake/use. If people feel that their own car will not prioritise their saftey then they may not use the self-driving mode. The stipulation could even therefore be reduced to "Self-driving mode should only be available if it is proven to be of less risk to surrounding people than manual driving."

If people feel that their own car will not prioritise their saftey then they may not use the self-driving mode.

That's their problem. Other more vulnerable road users shouldn't be made less of a priority when it wasn't their choice to drive a self-driving car.

He means avoidance of a crash, not avoidance of a person is not possible, you gave an example yourself, person or tree. Crash is unavoidable, car should go for tree, given that it has already in a split second worked out its stopping distance and recognised that it can't stop in time. As the article said, this is an extreme situation that will almost never happen but I think there needs to be some kind of programming there for eventualities like this.

Until people work out that you can crash an S class into a tree by jumping out in front of it and start doing it for fun.

Interesting point, but it would probably just stop. The troll in question would have to have some serious faith in the technology of the car to jump in front of it in a situation where theres no time to slam the brakes on

loop a loops. We need loop a loops...everywhere.

On a more relevant note. If I can find it, I'll post it, but there are still a long ways to go in terms of detection. Funnily enough, on a test run, one of google's cars was completely stumped by someone trackstanding and thus, wouldn't move. Apparently the logarithm that was used, was to assume anything that didn't fall within conventional safety parameters, as a possible vulnerable road user and so, stop completely.

Is there even such a car as an S Class Estate?

And even if there was (there wasn't) this is obviously untrue:

The Merc will out brake and out stop any other vehicle (bar another Merc with the same system)

Because a light car with grippy tyres will out stop a big lardy Merc.

https://m.youtube.com/watch?v=AsTxS6tg6xc

Looks promising!

Proof?
Lighter car will have budget rubber/brakes.
heavier car is heavier, but will have larger contact patch between tyre/ground. Computer controlled ABS and far superior quality 4/6 pot brakes.

I'll bet you fifty gbp that I can stop a Caterham from 70mph quicker than a computer can stop a big Benz.

Tho that's not really the point of this thread.

Hardly a budget car. Though fair point about weight, would pay to see that.

Relevant article.

That's their problem. Other more vulnerable road users shouldn't be made less of a priority when it wasn't their choice to drive a self-driving car.

No, it's everyone's problem. If a self-driving car has a vastly lower risk of accidents than manual driving even when it's in full protect the driver and fuck the pedestrian mode, then the greatest risk reduction is in getting the driver to engage self-driving mode in the first place. If they don't do that because we've stuck to our principles and made the car deprioritise driver safety then we've made life more dangerous for pedestrians. Who wins if we do that?

@Hefty Dunno, maybe not an S class estate, next one down perhaps.
Wrote the van off and "only" dented his tailgate.

I was taken for a demo in the little Kompressor coupe. If you had a gentle braking foot the car would brake like normal. if just lightly jabbed the pedal the cars brain decided it was emergency stop time and would take control and stop the car very quickly even if you jabbed and lifted, the car stopped.

It's a quirk thats been highlighted on Top Gear a few times. Some modern high performance cars can stop in fraction of the highway codes braking distance guidelines that Clarkson has argued they should be allowed to drive beyond 70mph.
Of course the issue is the car can stop allright, but it relies on the driver hitting the pedal and how capable is the driver at high speed thought processing and plain old good driving.

For many years Mercedes have had a maxim 'we like to think that no one ever dies in a Mercedes'. If they said their self-drive cars prioritise peds over passengers, frankly I wouldn't believe them... and you know what? Sometimes car manufacturers lie**

**VW

also, I was chatting to someone who works in transport tech a few weeks ago who reckoned self-driving cars will be safer if everyone is using one, but difficult to quantify whether they're safer or otherwise than a regular car. ie, it might need several years of data till we know. All a bit scary if you ask me.

Opportunity to input to a consultation here
https://www.parliament.uk/business/committees/committees-a-z/lords-select/science-and-technology-committee/news-parliament-2015/autonomous-vehicles-inquiry/

Some Cycling and walking orgs have done so.

Also thanks to David raising the issue there is a group of walking, cycling and driving orgs speaking about this to each other. David has suggested that this group form a loose body for the tech/manufacturing companies to speak to about safety issues and concerns in order to gain trust from these known organisations.

At CVPR'15 I asked Andreas Wendel about Google's take.. and the answer was "Accidents are avoidable".. :-)

I'm imagining code that looks like this going in to satisfy political pressure:

if in_the_shit:
   if have_reliable_information:
      trolley_problem()
   else:
      slam_on_brakes()

... and elsewhere:

if in_the_shit:
   # being in the shit is correlated with sensor failure
   sensor_error_likelihood += 0.1

...


if sensor_error_likelihood > 0.05:
   have_reliable_information = False

... so trolly_problem() is unreachable and can be coded in whatever way is most politically expedient.