Limitations of "radar cruise" augmented driving

There are six levels of autonomous driving, and I've always said that the difference between utopia and dystopia will lie in whether or not humans are allowed to do any of the stick and rudder ops for themselves.
As long as there are even a few cars at or below L3 cruising on the roads then we are still a quite a few tech doublings away from letting Robbie the Robot take over the helm completely.
Heck...
We can't even manage oil pipelines..... and as a telecommunications worker I have to hear about all sorts of conspiracy theories about JUST GETTING CELL PHONES to 5G JUST in big cities!....

The IOT security for interconnected "smart car" swarms will require a level of security that will make the TSA look like a retired mall cop.
(Note to millennials: A "Mall" used to be something like an enclosed retail or outlet center with some eateries and perhaps a bar and theater INSIDE one ginormous building. They often had their own security personnel which used to be something like campus police only older and with less authority and a LOT less mobility...
They were popular back in the long ago before the tech bros started dismantling brick and mortar establishments....)

Anyway....
The interconnectivity between cars will require an infrastructure R/D, buildout and maintenance budget that is still more than a few years off technologically and will nearly certainly take trillions to build and billions to operate.

There are approximately zero L5 semis on the road, and technologically speaking they would be the EASIEST to operate autonomously just from point to point on the interstate, and the real-world ROI would be an economic disruptor unseen since Edward "call me Ned" Ludlam beat the crap out of some knitting frames in the early 1800s.
AND.....presuming that this technology will reach maturation begore AGW destroys the planet......who gets to do the coin-flip in a Tesla versus School Bus interaction?
Would Robbie violate the first law of Robotics and destroy itself (and the hapless occupant!) to save a fellow vehicle that happened to have a more societally valuable cargo?
Who lays out the ethics matrix for THIS?
The same government that we haven't even been emaskapated from MONTHS after the flattened curve?

It's for darned sure I won't be around when we're finished laying down the beam ends for THAT eco system.....

THANK GOD!

 

Paraphrased Dilbert Cartoon, Dilbert talking to friend,
Friend: The government is forcing me to program a remote "kill switch" into our new self-driving car.
Dilbert: By kill switch I assume you mean the car will safely coast to a stop after exiting the road?
Friend: No, you're thinking of a "stop switch".

 

For liability reasons, engineers must choose safety as first priority.

But a very substantial fraction drivers are suffering from inattention, distraction, or fatigue, or intentionally choosing impatience, intoxication impairment, machismo, revenge, and/or to blow off the traffic safety laws, rules, and best practices. These human foibles cause a substantial majority of traffic fatalities, but are easily kept out of computer systems.

Strawman. The hospital already prefers you to dial 911 and wait for first responders, who are usually closer and can transport the patient with emergency lights and warning sirens while simultaneously providing treatment or monitoring.

 

Self driving systems don't need to be perfect before getting rolled out to the market, they just need to be better than typical humans in many conditions. And to know when to stop and tell a human to take over.

Most people forget that their own human foibles are the causes of most crashes and fatalities.

Primes are nowhere close to the current state of the art, so are a bad example. Look instead at the real self-driving efforts that are not (yet) on the consumer market, and to the more advanced driver assists that are on today's consumer market.

 

The only way to keep those human foibles out is to make it impossible for the human owner of the car to control the car they're driving in. There is no other possibility. That is completely and unequivocally unacceptable under all circumstances. If somebody other than me wants to control how my car moves, they are going to buy it for me with their own money (stealing it from me and then giving it back doesn't qualify) and pay for the upkeep.

 

Huh?

Don't be ridiculous. I'm not arguing that the only way to have self driving cars available is to abolish the human-driven cars. We can have both, mixed together on the same roads. The self driving systems don't need to be perfect, they just need to be able to demonstrate that they are less dangerous than the drivers they are replacing or substituting for, in the circumstances where they are operating.

 

It's not ridiculous at all. You can't remove the human element without, well, removing the human element. It's a logical absurdity.

The very people that cause accidents because of their driving are the very people that are going to override an autonomous car when it's not doing what they want. What they want is to be aggressive, cut people off, speed, pass on the shoulder, etc. The only way to prevent those people from driving like that and causing accidents is to make cars so that their self driving cannot be overriden.

 

Are you claiming that adding an override-able machine element will necessarily make the rouge humans worse than they already are? I don't agree.

I'm not calling for a removal of the human element, but only for a machine element to be less dangerous than the human elements they are substituting for. When the humans choose to retain control, then we'll just keep the same carnage we have now.

 

No. I'm saying crummy drivers are crummy and unless you prevent them from driving crummy they're going to drive crummy. If you give them a car that goes "too slow" in their opinion, they're going to override it to go faster. If the car is sitting in traffic when the driver feels like it should be passing on the shoulder, they're going to override it to pass on the shoulder. The same for every horrible driving habit these people have.

I don't think the impact to safety will be worth the risk of submitting our autonomy to even the potential of outside control. "Nobody's going to do that" isn't good enough. Only "there's no possible way for someone to do that" is good enough.

 

There are ways to handle the human pilot complication. Of course, none of them are perfect.

Start with the fact that in most states, you can pass a driving test with only 75% correct answers. And you can keep driving after a DUI, multiple tickets, accidents, etc.

How do you fix that? You can, for instance, change the laws so that getting a license requires that you ace the driving test, both driving and written. Couple that with revoking that license after a single ticket and you will have much more cautious drivers to contend with. This will not keep people from being able to use cars. It will require a level of skill and ability to drive one.

Less politically charged solutions might be to segregate the humans from the automatic cars. Easiest to do on the freeways. Not so easy to do in downtown San Francisco.

BTW, San Francisco is considered a very tough environment for an autonomous car, yet Google's cars were tested there for millions of miles WITHOUT an accident caused by the Google car.

 

Now let's put regular drivers with crummy driving records in those Google cars and see how they do.

 

No need. The accidents (there were some) were caused by bad drivers running into the Google cars.

 

What I mean is the cars had a good record because of the drivers in the car. Put the bad drivers in the car and watch the safety margin evaporate.

 

I think that you misunderstand the level of automation that the Google cars had reached before spinning the project off to the Waymo subsidiary. All those miles were clocked with a safety driver, but the car was in control. The number of driver interventions were amazingly few, considering the San Francisco topology, traffic congestion, etc.

 

I don't misunderstand at all. The reason the number of driver interventions were amazingly low is because they were good drivers. The number of interventions will increase as the recklessness of the driver increases. Why? Because the more reckless is the driver, the larger the gap between what the car is programmed to do and what the driver wants it to do. As that gap widens, so too does the likelihood of intervention.

 

That's an interesting theory, but what (other than your intuition) backs it up? The data they accumulated showed the cars did extremely well. There is no mention of recruiting extra highly skilled safety drivers. As a matter of fact, some of the miles were accumulated from random Stanford University students who just wanted to take a thrill ride. Stanford was a partner at the time.

Keep in mind that in research such as this, each intervention is scrutinized and examined for possible upgrades / fixes to the logic. A guy who hits the panic button so that he can drive on the sidewalk would be kicked from the program pretty quickly.

 

If this is what is going to happen, shouldn't it already be showing up in Tesla fatality rates?

Despite plenty of stories of crummy Tesla drivers abusing their misnamed driver assists, killing some of them, it appears that their overall fatality rate is lower than for other cars, not higher or similar.

You are using lack of perfection to block improvements.

I'm not aware that all improvements require a data connection. Maybe some features do, but not all. Don't force your tin foil hat onto everyone else.

 

I'm using cost-benefit analysis to block "improvements." Some things have an infinite ratio, like things that have a real potential to be abused to cripple fundamental human rights like freedom of movement.

 

That is called a PASSENGER TRAIN !!!
And as a country, we have made a HUGE mistake with letting most of that disappear.

 

Call me when they make a passenger train that goes to my friend's house, the store I need to stop at to get some bananas, and then to the trailhead where I'm going to hike - in that order and at the time I want to do it.