Hacker News Reader

Comment by femto | Hacker News Reader

What’s your view on whether self-driving cars should automatically be 100% liable for any accidents?

I ask this in the context of machines being governed by classical deterministic physics, so there is an argument that there is no such thing as an accident involving a self-driving car: only a design flaw.

This is a genuine question, as I can see that companies with self-driving systems that work, and who do serious fault analysis and rectification, might be in favour of 100% liability. 100% liability would stop cowboys from entering/surviving in the industry and sullying the reputation of self-driving. A company’s system would have to perform well enough that any residual risk of injury could be covered by an affordable insurance policy.

View on HN

If you listen closely, you can hear Cruise's legal team shouting, "No you can't make a public comment on what level of liability you think we should accept" no matter where you are in the world!

Of course, knowing that self-driving cars are 100% liable would incentivize some people to attempt to be hit by one of these vehicles for a payout. A more realistic level of liability would be for 100% liability for accidents resulting from an "unforced error".

I still think the cars should anticipate risks and behave accordingly. Out and out fraud aside, they should basically never injure anyone.

I agree

For one thing electric cars are too quiet.

There should be a law that someone with a bell has to walk in front of the car to let people know it is coming

Well, a noise making device has already been mandated because the startup car company refused to voluntarily install one. It hadn't been formalized because every other brands basically had it forever.

There's always some kind of risk. The bridge you're on could fall down. Somewhere, someone will have to judge whether a bad outcome was a failure.

Well as a pedestrian or cyclist, I'd like to make that risk judgment, not the vendor of the car that the person passing my bought from.

100% liable as drivers.

The most common cause of motorcycle injuries that make it to the hospital (and statistics) is someone turning left in front of them in an intersection where they have right of way.

Not quite this, but you get the idea:

https://nypost.com/2021/05/20/motorcyclist-rider-survive-hor...

I couldn't find the clip of a motorcyclist patiently stopped at an intersection get taken out by an out of control left turner. Lots of fully legally stopped vehicles get hit.

How does one take into account lack of maintenance by the end user in a strict 100% liability situation?

This problem should be solvable in software. The car can simply refuse to operate in a situation where maintenance is required.

I wonder about that. The top maintenance issue that comes to my mind is sufficient tread on tires. Bald tires will still work great on dry streets but as soon as it starts raining, you start skidding. I honestly don’t know if software could intervene quickly and reliably enough there.

The software could require a trip to the dealer for a visual inspection of the tires at set intervals. Hopefully free of charge for something so simple. A quick hookup to the computer and the interval is reset.

That seems like bad value when you can look at your own tires. (And should in case they get a sidewall bubble.)

In a world where most dealers and manufactures what you to pay a subscription for everything it isn't likely to be free of charge.

Tesla vehicles can detect and notify the driver of tires with low tread remaining. It's detected by a delta in rotation speed between other tires and the tire needing replacement. Seems like a software implementation is straightforward.

https://driveteslacanada.ca/software-updates/your-tesla-can-...

That doesn't help with end of service life, it helps with uneven wear.

If all the tires wear evenly this detection won't help.

Was editing my comment while you were commenting (removed service life, as software won’t detect dry rot or other defects undetectable from wheel speed measurements). Assuming tires wear evenly, you could still detect the change in rotation speed over time due to tread wear.

You can detect 3mm radius decrease? I would say no.

Remember that wheel slip depends on surface properties.

I think you can. Remember you have a lot of time to work with. The car already knows when wheels are slipping to discard that data. You just pick a time when you are going straight on a nice somewhat flat dry surface going a consistent speed and measure then, cars do that all the time even on curvy mountain roads you will find plenty of such stretches and you only need to measure every few hours.

3mm is about a third of total tread depth, and 1% of the tire's radius. Why wouldn't this be detectable? ABS sensors tend to have 48-tooth tone rings and there's no reason why you couldn't vastly increase this number if you wanted.

Longitudinal tire slip is caused by thrust in excess of the tire's grip, which is a function of slip speed among many other things. Grip peaks with a mild amount of slip, but slip isn't the norm outside of racing. Mild acceleration produces zero slip.

Lateral slip angle is a different story.

The slip is only 0 when free rolling by definition of the rolling radius.

You also have the rolling radius depend on load (car mass) and tire pressure.

I don't say I know it is impossible, but it feels like there is way to much noise.

It does if you have a GPS speed input. In the race car daq I use I can see speed differentials due to tire wear.

A typical 17" tire has a radius of 318 mm and 8 mm of tread. So a bald tire is 2.5% smaller than a new one.

This is what service intervals are for - your car likely requires an service every 12 months or XXXX kilometers whichever comes first. The service doesn't just include actual work on the car it includes an inspection of the lights, tyres, etc and a report to the owner saying "tyres need replacing in the next couple of thousand Ks they're almost at the wear indicator".

In the general case it's impractical for electronic sensors to accurately measure the mechanical state of the vehicle. How do they tell us the suspension is rusted out and about to break? (In theory you can play some clever tricks with eddy currents or something but that's not going to be feasible for real world sensing.)

That is not much of a problem in practice. A 'rusted out' suspension doesn't happen overnight. There could be regulatory requirements for self-driving cars to be considered 'streetworthy'. Out of compliance, robotaxi disabled.

The tricks you mentioned are already used for some aircraft inspections.

What the software needs to worry about would be other types of failures. Software is much more likely to detect issues before the driver. Say, brake performance is outside the expected range, or appears to be degrading too quickly.

How do you know maintenance is required in a completely automated fashion?

My fear is that car manufacturers will turn cars into a totally dealer serviceable only thing (even more than they are now), like the car version of the glued shut Microsoft surface that gets a 1/10 on the ifixit repairability score.

In the case of Cruise this wouldn't be a problem because you wouldn't own the vehicle. Its a robotaxi service. Your point is still valid, though I'd ask, how do you even solve certain classes of issues? Like lets say you had to replace a camera. You can't just plop one in and have it work. There is a ton of complex calibration work that needs to happen, both intrinsic & extrinsic.

Service intervals based on time and usage combined with certified repair. From a passengers perspective airlines are strictly liable but presumably airlines could then sue the relevant third parties in such a case. I suspect a similar model could work fine for self driving cars.

Put a rfid tag in the tire, store how many rotations that tire takes over time. Once it reaches a threshold, refuse to spin that tire further.

Tire wear is a complex interaction of various factors such as compound, slippage, road surfaces, torque, weather etc and not just wheel revolutions

I've given it some thought, and I think the SDC manufacturer must be liable for any accidents the SDC causes. Who else is there? The passengers certainly can't be responsible for any programming or manufacturing errors.

There are corner cases and exceptions, but that has to be the rule.

Which should mean that as and SDC owner, you don't have to pay car insurance.

>the SDC manufacturer must be liable for any accidents the SDC causes.

There's the rub... How do you seperate those from the rest?

We already have a system that relies on assigning fault among the multiple parties involved in an accident. This same approach applies just as well to SDC accidents. It would be even easier than the status quo, given the much richer data SDCs could be regulatorily mandated to provide.

The whole event will be recorded by at least the SDC vehicle.

If there is disagreement between the relevant parties, through the court system (or through insurance agreements).

When you say accidents you mean where the robo car erred? Asking because ofc it's possible to get into an accident where you are not at fault and I think this would also be true of robocar

Reply on news.ycombinator.com