https://www.businessinsider.com/war...gistics-self-driving-trucks-technology-2018-6

Broughton's analysis is not correct.

The technology is NOT available at this time and there are a lot of legal issues that must be resolved before self driving vehicles are allowed on public streets and highways. Insurance companies are going to have the last say and if vehicle manufacturers and operators can't get insurance, all bets are going to be off the table.

 

https://arstechnica.com/cars/2019/0...epeating-silicon-valleys-most-famous-blunder/

 

https://www.bloomberg.com/news/arti...-may-one-day-kill-car-insurance-as-we-know-it

On Jan. 30, Peate announced the creation of Avinew, with $5 million in seed funding led by Los Angeles’s Crosscut Ventures. Its insurance product will monitor drivers’ use of autonomous features on cars made by companies including Tesla, Nissan, Ford and Cadillac, determining discounts based on how the feature is used. Avinew has agreements with most manufacturers and is working to tie up the rest, Peate said, allowing it to access driving data once a customer gives it permission.

Deloitte, in its 2019 insurance outlook report, saw this coming. “The rise of connectivity … has generated a massive amount of real-time data and turned the insurer’s relationship with policyholders from static and transactional to dynamic and interactive.” Avinew said it expects to be writing policies later this year in select states.
...
The good news for the industry is that it has time. Stevens estimates that by 2035 there will be only 23 million autonomous vehicles on American roads—less than 10 percent of today’s total. And as of now the technology required for autonomous features is extremely expensive to repair, meaning premiums will initially rise as more cars featuring them roll off dealers’ lots.

 

I think it is inevitable at this point, what I find fascinating is the driver assist techs that are spinning off, such as automatic breaking (for pedestrians)and doors that lock if there is oncoming traffic from behind. Things that no one thought of when it all started.

 

https://www.theverge.com/2019/4/22/18511527/elon-musk-tesla-aggressive-autopilot-mode-fender-bender

Elon Musk said at an event on Monday that Tesla will someday allow drivers to select aggressive modes of its Autopilot driver assistance system that have a “slight chance of a fender bender.” Musk didn’t say when Tesla might roll out that option, only that the company would have to have “higher confidence” in Autopilot’s capabilities before allowing it to happen.

“Do you want to have a nonzero chance of a fender bender on freeway traffic?” Musk asked at the event, which was for investors in the company. He dubbed it “LA traffic mode,” because “unfortunately, [it’s] the only way to navigate LA traffic.”

Will he pay for the 'fender bender' when you switch to A$$hole Mode?

 

https://www.consumerreports.org/aut...nge-requires-significant-driver-intervention/

“This isn’t a convenience at all,” says CR’s Fisher. “Monitoring the system is much harder than just changing lanes yourself. Using the system is like monitoring a kid behind the wheel for the very first time. As any parent knows, it’s far more convenient and less stressful to simply drive yourself.”

 

Driverless cars now have a new use.
https://www.bbc.co.uk/news/uk-england-south-yorkshire-48924715

 

Driverless cars now have a new use.
https://www.bbc.co.uk/news/uk-england-south-yorkshire-48924715

Sounds very, very, improbable in reality. Lots of irrational charges and suppositions about intentions based mainly on fear instead of solid facts.

Miss Whyte told the jury that Salah was a supporter of so-called Islamic State, despite his being an Iraqi Kurd, a nationality usually associated with the fight against the terror group.

 

When self-driving (driver-assist emergency braking) works well.

 

Sorry if this has been said as I didn’t read all 18 pages.

Many comments are made regarding flagmen and/or police officers in the road. I suggest accommodating autonomous vehicles by including machine readable notices (QR Code) on warning signs. Similarly as the population of autonomous vehicles increased, MRS could be extended to traffic signs generally.

This solution creates opportunities for creation of an enterprise which provides a central clearinghouse and registration of all codes in use that map to an action data structure.

 

Sorry if this has been said as I didn’t read all 18 pages.

Many comments are made regarding flagmen and/or police officers in the road. I suggest accommodating autonomous vehicles by including machine readable notices (QR Code) on warning signs. Similarly as the population of autonomous vehicles increased, MRS could be extended to traffic signs generally.

This solution creates opportunities for creation of an enterprise which provides a central clearinghouse and registration of all codes in use that map to an action data structure.

That sounds like a very juicy spot to hack the entire transportation infrastructure. I think we need systems to be able to handle asynchronous events (local changes in normal road traffic) like this independently or in a local real-time network of nearby autonomous vehicle sensors. A centralized system will have people stuck on dirt roads in Denver.

Truly autonomous vehicles need to operate at least as effectively as humans in areas without external guidance. The vast stretches of this country (many without any cell, data-links and a times questionable GPS reception) makes Dead reckoning a course of action absolutely necessary for a large percentage of the population.

 

Sorry if this has been said as I didn’t read all 18 pages.

Many comments are made regarding flagmen and/or police officers in the road. I suggest accommodating autonomous vehicles by including machine readable notices (QR Code) on warning signs. Similarly as the population of autonomous vehicles increased, MRS could be extended to traffic signs generally.

This solution creates opportunities for creation of an enterprise which provides a central clearinghouse and registration of all codes in use that map to an action data structure.

And, of course, every car will have the ability to distinguish between legitimate QR codes used as intended and malicious ones that are placed with the intent of causing the cars to do what they're not supposed to.

 

And, of course, every car will have the ability to distinguish between legitimate QR codes used as intended and malicious ones that are placed with the intent of causing the cars to do what they're not supposed to.

Why not? With a centralized source for the translation and intelligence in the vehicles, two-factor authentication and one time non-reversible encryption, this becomes a non-issue. The QR code is NOT the data. It’s a link to somewhere/something that requires authentication.

And not all actions require authentication. A QR code, that says Stop or Slow, has a low impact for malicious intent. Unless you're kidnapping a passenger. Even then, I believe that can be avoided.

What makes it attractive is there are no legacy systems that need to be accommodated. Plug in modules can change at will. And since multiple, physically separate systems are involved, hacking them is presented with the challenge of needing to hack multiple nodes, where at least one is random and unidentifiable.

Security is enhanced by the design. Let’s say you want to hack into a financial account. You can attack one node, your bank. But you have no idea where the other node is. To access the system, you need the physical presence of a relative. That relative has a key. But even you don’t know where that relative is. And each relatives key changes over time. So if you can find the relative and steal his key, by the time you use it - it’s invalid!

 

Why not? With a centralized source for the translation and intelligence in the vehicles, two-factor authentication and one time non-reversible encryption, this becomes a non-issue. The QR code is NOT the data. It’s a link to somewhere/something that requires authentication.

A non-issue? How does it solve the replay attack that I described (legitimate codes used in ways other than intended)? So if someone moves a legitimate QR code from one place to another place, how is the system going to recognize that it is now an illegitimate QR code? And the bad guy doesn't have to move anything -- take a picture of a legitimate QR code that tells the car X and then reproduce a sign with that code and place it in a location where doing X results in bad things happening.

So we need each QR code be geotagged in the database with sufficient resolution so that it can't be used anyplace other than the assigned location -- and that would have to be with enough resolution so that it can't be moved from the twenty feet from, say, a frontage road shoulder to the highway shoulder. Plus our police and firefighters have to generate the QR code, register it with the database, print it, and display it. And, of course, accidents always happen where there is sufficient connectivity to do all of this in real time.

But more to the point, it doesn't solve the problem. How does a QR code on a warning sign enable the car to follow the directions of the police and firefighters that are directing traffic at an accident scene? Or the person that is doing so on their own while waiting for the police/fire to respond? Or consider the teenage kid directing cars to parking spots in a dirt/grass field at an airshow or concert or just their kid's school's open house?

 

A non-issue? How does it solve the replay attack that I described (legitimate codes used in ways other than intended)? So if someone moves a legitimate QR code from one place to another place, how is the system going to recognize that it is now an illegitimate QR code? And the bad guy doesn't have to move anything -- take a picture of a legitimate QR code that tells the car X and then reproduce a sign with that code and place it in a location where doing X results in bad things happening.

So we need each QR code be geotagged in the database with sufficient resolution so that it can't be used anyplace other than the assigned location -- and that would have to be with enough resolution so that it can't be moved from the twenty feet from, say, a frontage road shoulder to the highway shoulder. Plus our police and firefighters have to generate the QR code, register it with the database, print it, and display it. And, of course, accidents always happen where there is sufficient connectivity to do all of this in real time.

But more to the point, it doesn't solve the problem. How does a QR code on a warning sign enable the car to follow the directions of the police and firefighters that are directing traffic at an accident scene? Or the person that is doing so on their own while waiting for the police/fire to respond? Or consider the teenage kid directing cars to parking spots in a dirt/grass field at an airshow or concert or just their kid's school's open house?

The replay attack would only be valid within a very narrow time frame. Let’s say, an authorization code is only good for 60 seconds. If you can’t hack the system in a minute, you are SOL.

I’m going to reply to your comments, this time, but I don’t want to be argumentative. You bring up excellent design considerations that can be addressed. As you have done by suggesting including GPS awareness into any solution.

It’s a case where “Don’t let perfect be the enemy of good enough”.

I don’t believe that any autonomous vehicle should strive for 100% autonomy. If direction by first responders is necessary, then shut the damn robot down.

First responders needn’t generate a new QR code nor anything else you suggest. They pull a sign out of the trunk of their cruiser and put it up. At a minimum it instructs the auto car to go to manual mode. Otherwise, a dispatcher can update specific instructions.

As far as that teenage kid at a concert? Damned if I’m going to park at a concert in auto mode. I want the autonomous vehicle to get me there; I’m parking my own car.

Assuming that autonomous vehicles can ever protect me 100% of the time is foolish.

I’m done! Thanks for your thought provoking comments.

 

A non-issue? How does it solve the replay attack that I described (legitimate codes used in ways other than intended)? So if someone moves a legitimate QR code from one place to another place, how is the system going to recognize that it is now an illegitimate QR code? And the bad guy doesn't have to move anything -- take a picture of a legitimate QR code that tells the car X and then reproduce a sign with that code and place it in a location where doing X results in bad things happening.

So we need each QR code be geotagged in the database with sufficient resolution so that it can't be used anyplace other than the assigned location -- and that would have to be with enough resolution so that it can't be moved from the twenty feet from, say, a frontage road shoulder to the highway shoulder. Plus our police and firefighters have to generate the QR code, register it with the database, print it, and display it. And, of course, accidents always happen where there is sufficient connectivity to do all of this in real time.

But more to the point, it doesn't solve the problem. How does a QR code on a warning sign enable the car to follow the directions of the police and firefighters that are directing traffic at an accident scene? Or the person that is doing so on their own while waiting for the police/fire to respond? Or consider the teenage kid directing cars to parking spots in a dirt/grass field at an airshow or concert or just their kid's school's open house?

Imagine replacing the QR code for "15 MPH" with one for "75 MPH". Carnage will result.

 

I just don't see any of the major autonomous vehicle pioneers relying on centralized systems because of the time-frames needed to build practical city/state/nation wide systems beyond a test track even if it was 100% safe, secure and reliable. Any system without the capability to locally handle common road hazards and ad-hock driving is not level-5 (true self-driving) automation, with all levels below this, the driver is expected handle those types of problems.
https://www.transportation.gov/site...uture-transportation-automated-vehicle-30.pdf

 

The replay attack would only be valid within a very narrow time frame. Let’s say, an authorization code is only good for 60 seconds. If you can’t hack the system in a minute, you are SOL.

I’m going to reply to your comments, this time, but I don’t want to be argumentative. You bring up excellent design considerations that can be addressed. As you have done by suggesting including GPS awareness into any solution.

It’s a case where “Don’t let perfect be the enemy of good enough”.

I don’t believe that any autonomous vehicle should strive for 100% autonomy. If direction by first responders is necessary, then shut the damn robot down.

First responders needn’t generate a new QR code nor anything else you suggest. They pull a sign out of the trunk of their cruiser and put it up. At a minimum it instructs the auto car to go to manual mode. Otherwise, a dispatcher can update specific instructions.

If they can pull a sign out of their trunk that has been sitting there for a week, then why can't I pull a copy of that same sign out of my trunk that I made after taking a picture of the cop's sign a week ago? I don't have to hack into anything -- I replay a legitimate sign. How does the car tell the difference?

As far as that teenage kid at a concert? Damned if I’m going to park at a concert in auto mode. I want the autonomous vehicle to get me there; I’m parking my own car.

And just how does manual mode work in a car that doesn't have a steering wheel or pedals? That's the car that everyone is pushing towards and that companies keep claiming they will have on the roads in the next model year (of course, like flying cars and nuclear fusion, they've been renewing that claim year after year).

Don't believe me?

https://www.wired.com/story/gm-cruise-self-driving-car-launch-2019/

https://www.businessinsider.com/fords-self-driving-cars-wont-have-steering-wheels-2016-8

https://money.cnn.com/2018/01/12/technology/general-motors-cruise-self-driving-car/index.html

https://www.reuters.com/article/us-...s-with-no-steering-wheel-brakes-idUSKCN1QW249

and countless more hits from a simple Google search.

The oft-repeated claim is that Level 5 autonomous cars that do not allow the passenger to control it at all is necessary before the safety and efficiency goals promised by the technology can be achieved -- with the often unstated assertion that it also requires removing all other vehicles that CAN be operated in manual mode OFF the roads.

 

If they can pull a sign out of their trunk that has been sitting there for a week, then why can't I pull a copy of that same sign out of my trunk that I made after taking a picture of the cop's sign a week ago? I don't have to hack into anything -- I replay a legitimate sign. How does the car tell the difference?
.

WBahn, you’ve baited me successfully. I said I was done, but once more into the fray.

First, I’m not going to address anything about “manual mode”. It’d be like arguing that a kayak is unsafe on an interstate highway. D’oh!

But with regard to police pulling a sign out of their trunk which has been sitting there for a week... they can’t. I’ve explicitly stated that any QR code means absolutely nothing by itself. So you have a copy of a meaningless sign? Tell me what the risk is?

I make up a sign. It says “ABADOUGJ TREMENOSTIK”! I put it on a highway. Is it going to cause a problem? Maybe if someone laughs so hard they run off the road.

The security is that any given sign has no meaning by itself. If the autonomous car authenticates itself with a rolling code and the sign mapping software also validates that it’s a legitimate sign with a similar rolling code and an authenticated user (who has been validated with s rolling code) has mapped an action to the occurrence of the sign - where is the replay hack?

A hacker would have to break at least three points of authentication, that are protected by a time dependent rolling code.

I don’t know about you, but three layers of defense authenticated by a continuous security change, is about as secure as I desire.

But maybe you are just paranoid

Only kidding, @WBahn, I know you realize how much I respect you.

Really, these similar issues always appear after the fact. But if the designers looked ahead, it’s a lot easier to design away a problem than try to retrofit a solution.

 

All of this complexity to solve a set of problems that we really don't know will be solved by self-driving cars unless we also totally redesign the trillions spent on manual driving infrastructure into a closed system of total control and surveillance of the populations private car movements. I just drove 3000 miles round trip at >75MPH legal speeds on most highways filled with human drivers. 99.9995 percent of the time we arrive without a major incident even with all of the human errors as autonomous tech filters back down to manual cars making them even safer.
I think it's going to be very hard to improve that with AI due to the physics of moving mass limiting corrective actions and random events that cause 'Trolley' problems.

https://en.wikipedia.org/wiki/Trolley_problem

Has anyone solved the self-driving car 'Kangaroo' problem?