Ford, GM and Toyota Collaborate For Self-Driving Safety Rules (detroitnews.com)

Ford, General Motors, and Toyota have formed a new consortium called the Automated Vehicle Safety Consortium (AVSC) to develop safety standards for self-driving cars. "The newly formed Automated Vehicle Safety Consortium in conjunction with the auto engineering association SAE International says it will fill a critical need by providing a safety framework around which autonomous technology can responsibly evolve before self-driving vehicles are put into widespread use," reports The Detroit News. From the report: Being able to advance the safe deployment of fully self-driving cars represents a new step toward the benefits the technology will bring, said Edward Straub, director of automation for SAE and executive director of the new consortium. Straub said the automakers in the new consortium would turn information discovered through their self-driving testing over to SAE committees every three to six months, and the information would be discussed in public SAE sessions as a set of guidelines are being developed.

Straub said other automakers and technology companies would be welcome to join the consortium, provided they have experience testing fully autonomous cars. The announcement of the new partnership may be a reaction to the inability of Congress to pass legislation that would allow car manufacturers to sell thousands of self-driving vehicles in the near future, said Michelle Krebs, senior analyst for Autotrader. "GM, Ford and Toyota clearly saw a need to set standards that eventually may become regulations because the proposed regulations, which had been moving quickly, have now stalled," she said. Straub said the automakers in the new consortium are operating independently of the efforts to pass legislation in Congress.

Comments

Simplest rule

[SuperKendall]

Each car can electronically compare purchase price with each other, most expensive car gets right of way, least expensive car gets the ditch.

Re:Simplest rule

[K. S. Kyosuke]

Don't forget to add driver's net wealth into the equation.

Will they take criminal liability? or do an uber

[Joe_Dragon]

Will they take criminal liability? or do an uber where that safety driver may face criminal changes even when they need to look off the road at the cars systems.

What could go wrong?

[Anonymous Coward]

Toyota is handling the braking controls.
Ford is working on automatic rollover prevention.
GM is developing a new system to replace ignition controls.
What could go wrong?

Re:What could go wrong?

[dgatwood]

Toyota is handling the braking controls. Ford is working on automatic rollover prevention. GM is developing a new system to replace ignition controls. What could go wrong?

Also:

• Toyota is handling the accelerator pedal.
• GM is developing the steering.
• Ford is developing the fuel tank and the rear bumper.

Re:What could go wrong?

[krakelohm]

*fireworks not included.

Re:Tesla

[Anonymous Coward]

Will have full self-crashing by the end of the year.

Fixed that for you.

Put the customer in charge of the rules

[themusicgod1]

The right to repair guarantees that it's the *customer* who is the backstop of ethical behaviour of their vehicle, which is exactly where it should be. I do not want the decision of who my car is going to kill going to be made by some company that owes it's existence on sucking on the government bailout tit.

Not self-driving, so not invited

[thesjaakspoiler]

It will be a long way before the bicycle manufacturers come up with self-driving bicycles so they are not invited to the party. And concerning pedestrians, they are in the category 'humans' which are in general considered to be too incapable doing anything beynd posting messages on social media.

Re:Not self-driving, so not invited

[JaredOfEuropa]

A self driving motorcycle would be a boon, however. That way, the rider can give his undivided attention to wildly gesturing at drivers on their cell phones, and slapping wing mirrors of drivers he thinks are not giving him enough space. Meanwhile the AI can be set to accelerate fast if it detects someone pulling out of a driveway a couple hundred meters down the road. Maybe it can even handle the rev bombing...

Re: Not self-driving, so not invited

[Rick Schumann]

Thanks for making it easy to confirm that's just a trolling account, it'll make it easier for the site owners to delete it and block your IP address. Have a nice day! :-)

Re:Not self-driving, so not invited

[Viol8]

You'd need a special subroutine for 50cc mopeds - always filter to the front of the queue at the lights then when they change to green accelerate away at the pace of a kneecapped tortoise staying in the middle of the road at all times and making sure at least some drivers behind don't make it through before the next red. Also - see self driving bicycles.

That's not enough

[HangingChad]

It's a good start but we need communication protocols so cars can talk to one another and so traffic control devices can talk to them. We need uniform standards for road sensors, lane markers and broadcast obstruction warnings.

Maybe this is the incentive we need to finally fix our broken infrastructure.

Re:That's not enough

[dgatwood]

It's a good start but we need communication protocols so cars can talk to one another and so traffic control devices can talk to them.

Actually, no, we really don't, despite the old guard car companies' near-constant insistence that it is somehow critical. There's no plausible design for an inter-car communication protocol that can't be forged, and if you can't trust the data coming in, you can't really do much useful with it, so what's the point of even sending it? It's not as if the difference between the few milliseconds it takes for a computer to recognize what's happening visually and the few nanoseconds it takes to decode the signal electronically is going to make any real difference anyway, in practice.

Also, traffic control devices had better be visually obvious enough that humans can recognize them, or else they won't work, and if they are, then computers don't need any additional electronic communication. It just introduces more opportunities for bugs and hacking.

We need uniform standards for road sensors, lane markers and broadcast obstruction warnings.

This, I agree with. Of course, making that happen around the world is about as likely as Tesla reaching level 5 autonomy in 2019. :-)

Re:That's not enough

[Kjella]

I'm pretty sure that if traffic was all cars we'd have self-driving cars by now, for the simple reason that cars are bound by traffic rules. If it's an intersection and you're running a red light there's not much doubt who's at fault. If the rules are too ambiguous we can always make them clearer until say a lane merger is properly described and you can put a car's behavior into "at fault" or "not at fault". Right now we're trying to work backwards from "do not crash" into behavior that won't ever cause a crash and that's probably not happening, if you're a dick it's not hard to get other cars in trouble in ways that today would be considered a joint fault. And if you're trying to be a dick like speed matching a car overtaking you close to 100% at fault. We're going to have to describe correct behavior better and then deal with the rest in court.

Re:That's not enough

[dehachel12]

https://www.teslarati.com/tesl...

Tesla released its first Vehicle Safety Report after the third quarter of 2018. During that period, the electric car maker registered one accident or crash-like event for every 3.34 million miles driven with Autopilot active, and one accident or crash-like event for every 1.92 million miles driven with Autopilot disengaged

Re:That's not enough

[dgatwood]

And forging it will get you federal pound-me-in-the-ass prison time.

Only if they catch you.

I can take a radio out and start pretending to be an air traffic controller right now, but I don't and neither does anyone else, because to do so means going to prison. Are you going to propose that we remove radios from planes?

No, because if you tried that, they would trace the signal back to you and put you in jail. If you want to cause self-driving carmageddon, attach a radio to a Raspberry Pi with four D cells, program it to transmit a signal that says "Speed limit 90 MPH", and toss it out on the side of the road right before a hairpin turn. Good freaking luck catching the person who does something like that.

Also, UWB ranging can supposedly be made secure, so potentially the attacker would not be able to significantly forge his position.

You're still making the flawed assumption that the attacker is a "who" and not a "what". There's your first mistake. Start with the assumption that the cars are all connected to a cellular network, and can be hacked from anywhere on the globe. Now design a system in which cars are somehow supposed to make meaningful use of data sent by other similarly connected cars while ensuring that if one gets hacked and starts sending out a "Slow down here. There's a pothole" message every five seconds, it won't bring the entire traffic grid to its knees as the message gets passed on to car after car.

Any time you trust data from outside the vehicle, you run the risk of a malicious actor compromising the data provider and giving you garbage. That's why self-driving cars have to be able to handle nearly all driving without even so much as a map, or else they cannot ever be truly safe. This is not to say that they should be able to figure out where to go without a map, but they should not drive into buildings or drive off of cliffs or miss traffic lights or any of the other things that come from trusting outside data over what is immediately observed.

So if you have to rely on observation as your primary source of truth anyway, how much data can you realistically obtain from other cars that would be beneficial (I mean, of course, beyond what can already be determined from people's cell phones automatically checking in with Waze or whatever)? I can't think of anything, really.

Re: That's not enough

[Type44Q]

We don't need an incentive to fix infrastructure; those holding the wealth do.

Talk to them.

Re:That's not enough

[religionofpeas]

If the rules are too ambiguous we can always make them clearer until say a lane merger is properly described and you can put a car's behavior into "at fault" or "not at fault".

Properly describing a lane merger is already impossible. If you put all the responsibility for safety on the merging car, it becomes impossible to merge in busy traffic. The only solution is to become more aggressive in merging, and force traffic to brake for you.

Re:That's not enough

[religionofpeas]

If you want to cause self-driving carmageddon, attach a radio to a Raspberry Pi with four D cells, program it to transmit a signal that says "Speed limit 90 MPH", and toss it out on the side of the road right before a hairpin turn. Good freaking luck catching the person who does something like that.

Where do you get the key to sign the message ?

Re:That's not enough

[fluffernutter]

Tesla wouldn't compare easy driving miles where Autopilot does work to harder miles where it doesn't would they? I'm assuming in your stats they only compared miles where people can use Autopilot? Also if a person didn't activate Autopilot that day in a certain place because there was a circumstance like construction they shouldn't compare that as well.

Re:That's not enough

[110010001000]

Autopilot isn't self driving.

Re:That's not enough

[fluffernutter]

I think if you compared most very expensive vehicles, you would find that the owners are careful with them. Also, keep in mind that there are fewer Teslas on the road so fewer chances of other people hitting them, which would pad the stats.

Re:That's not enough

[fluffernutter]

I was thinking incorrectly with that last line. Posting before coffee.

Re:That's not enough

[drinkypoo]

It's a good start but we need communication protocols so cars can talk to one another

You mean like these? Noob. You should at least give yourself a basic education in a subject before spouting.

and so traffic control devices can talk to them. We need uniform standards for road sensors, lane markers and broadcast obstruction warnings.

An AV which depends on V2X is untrustworthy. AVs have to work even when the communications network is completely down in order to be useful, because otherwise people will learn to depend on them in situations in which they are not dependable.

You have no idea what you are on about. Go away.

Re:That's not enough

[drinkypoo]

I'm pretty sure that if traffic was all cars we'd have self-driving cars by now, for the simple reason that cars are bound by traffic rules.

No. Those are laws and regulations, not rules. Rules are things like "don't crash" or "follow laws and regulations". The laws and regulations are complex in and of themselves, but actually following the rules of driving is orders of magnitude more complicated.

If traffic were all trains, we'd have self-driving transport by now. But since cars have to operate on tarmac (etc.) roads with pneumatic rubber tires, even if you had only passenger cars and they didn't have to mix with anything else, driving would still be very complicated.

if you're trying to be a dick like speed matching a car overtaking you close to 100% at fault. We're going to have to describe correct behavior better and then deal with the rest in court.

AVs won't do that shit at all. They will determine which vehicle can most efficiently get into a space, and then let that happen, because more road efficiency means more car efficiency. And they will be configured to be forgiving of other drivers, so a pushy human driver will be able to bully AVs out of the way... but they will also be automagically reported to the authorities by the AV coordination systems.

Re:That's not enough

[dgatwood]

If you want to cause self-driving carmageddon, attach a radio to a Raspberry Pi with four D cells, program it to transmit a signal that says "Speed limit 90 MPH", and toss it out on the side of the road right before a hairpin turn. Good freaking luck catching the person who does something like that.

Where do you get the key to sign the message ?

It's hard to answer that without knowing what the architecture would be. For any given architecture, there's a different answer. The most likely answer, though, is that someone borrows a device from an existing speed limit sign and copies the key out of it.

If we assume a nice architecture with each device having its own key with proper key signing from a root cert, that then becomes a trivial revocation problem, but only after they discover that it has been stolen, which doesn't happen until after the first car goes off the road and somebody analyzes the black box (or after the tenth car goes off the road and they do a really careful survey of the area near the road).

Unfortunately, the most realistic design for mass-manufactured devices like speed signs would be for the manufacturer to assign a single, standard key across all devices of a given model, which results in revoking all speed limit signs in the State of California and tens of thousands of man hours to physically remove all those transmitters (because they wouldn't be Internet-connected), rekey them, and replace them.

Or if we assume that the actual safety devices are connected to the Internet, then it becomes even more entertaining, because somebody could just crack one remotely and reprogram it, all while sitting safely in a non-extradition country.

after 3-4 year updates cost $250-$500/year dealer

[Joe_Dragon]

after 3-4 year updates cost $250-$500/year dealer only and if you need an bigger disk to say hold the super mapped roads 1TB $250 + $250 labor at the dealer to install.

Re:FUCK self-driving cars.

[Rick Schumann]

I normally don't agree with ACs but in this case you're right, nobody really wants this shit, and it's not ever going to work well enough to be practical, it's going to be a horrifying nightmare of unnecessary accidents and deaths.

Re:They have no standing

[dehachel12]

while tesla is already ACTIVELY preventing crashes : https://www.teslarati.com/tesl...

Not trusting it

[houghi]

Sounds like a way to make rules that will see that others are always at fault. The reason O think that us because it is two US and a japanese company.

It is telling if the company that gave up the patent for 3 ppint safety belts is not part of it.

Sounds like a health research group from the tobacco industry.

Re:Not trusting it

[mjwx]

Sounds like a way to make rules that will see that others are always at fault. The reason O think that us because it is two US and a japanese company.

It is telling if the company that gave up the patent for 3 ppint safety belts is not part of it.

Sounds like a health research group from the tobacco industry.

Actually it's more of a free ad for Toyota, GM and Ford.

The road rules (known as the Highway Code) here in the UK are very clear about what vehicles should be doing and when, when to go, when to stop, when to give way, what distance to maintain, when to indicate, how long to indicate for, so on and so forth up until who is at fault in a collision. Very little is ambiguous.

For self driving cars, the highway code simply needs to be turned into computer code. The problem is the highway code assumes the vehicle controller (A.K.A the driver, or at least the steering wheel attendant) can tell the difference between a leaf and a child or can determine which lane markings it's meant to remain within. Automatic systems can't do this yet... and likely not for a while.

Re:Not trusting it

[religionofpeas]

For self driving cars, the highway code simply needs to be turned into computer code.

That's far from simple, as you explain in your next sentence. A common set of safety standards would help to make sure they all do this correctly.

Re:after 3-4 year updates cost $250-$500/year deal

[Patent Lover]

I think you haven't been to a dealer lately. It's more like $1000 labor plus $500 for a 250GB drive.

This is useless

[Sqreater]

China will start selling dangerous self-driving cars and the whole thing will collapse.

AI models based on their adverts?

[shufflingb]

If so they're in for a bit of a surprise when they discover that their adverts ubiquitous portrayal of empty roads, driving wherever, at whatever speed and it's all just tickety-boo is one big lie that'd make even the Marlboro Man's death stick purveyors blush.

Seriously though, I think we're at a juncture were we need to understand very quickly what they're being lined up for with self-driving cars by the car/auto lobby. For while it's likely that the number of accidents will go down. It's also true, all else being equal, that the number of car journey will increase. Is there anyone outside of the auto lobby that thinks, yep, what we need is more car journeys. That'll help with the obesity crisis, pollution and global warming. Really?

So obviously, given the likely outcome of their endeavours what do we think these companies are going to do about it. Is it going to be a) Create policies that attempt to humanise their technology, reduce environmental damage and generally make life better for everyone including non-car owners/users. Or b), keep making vast amounts of money at whatever the cost to society and the planet? How well do we think our legislatures are prepared for this?

Re:AI models based on their adverts?

[drinkypoo]

while it's likely that the number of accidents will go down. It's also true, all else being equal, that the number of car journey will increase. Is there anyone outside of the auto lobby that thinks, yep, what we need is more car journeys. That'll help with the obesity crisis, pollution and global warming. Really?

This is why we need to promote rail. Here in the USA, Trump is attacking national rail, probably on behalf of the oil and car companies. It's happened before that auto companies destroyed profitable public transportation lines to increase demand for their product (with the aid of oil and tire companies, who wanted the same for their products.)

The future of ride hailing is the use of a single app to manage an entire journey. A trip could involve an autonomous car showing up to take you to a train station, where you'd transfer (on one pass) to a train that would take you to an airport, where you'd transfer again, etc. If we are vigilant, that journey will make sense.

Re:AI models based on their adverts?

[religionofpeas]

Rail is very expensive. An autonomous bus in a dedicated lane would be a much cheaper alternative.

Re:AI models based on their adverts?

[drinkypoo]

Rail costs more, but it also carries much more, and costs less to maintain. It's also much more efficient to use steel wheels on smooth rails than to use pneumatic rubber tires even on smooth pavement, let alone the roads we actually wind up with. Trains are also more compatible with mobile charging. They just make more sense for long-distance travel, and often for medium-range trips. Cars are still better for short ones. Buses on tarmac are a problem because heavy axles put high PSI loads on pavement. Trains spread out the load better, so the infrastructure is less self-damaging.

Re:They have no standing

[fluffernutter]

Again, article did not explain the methodology used for this study. Sorry but I don't trust Tesla or Musk to do a study properly.

Re:They have no standing

[fluffernutter]

What? They're saying the technology isn't ready yet and we should slow down with development because people may die? Say it ain't so!

I'll go with the companies that don't sell flamethrowers on the side, thanks.

Easy

[nospam007]

A car may not injure a human being or, through inaction, allow a human being to come to harm.

A car must obey orders given it by human beings except where such orders would conflict with the First Law.

A car must protect its own existence as long as such protection does not conflict with the First or Second Law.