Self-Driving Cars and Taxis Are Here! 5

[IRstuff]

https://abc7news.com/self-driving-cars-san-francisco-robo-taxi-waymo-cruise-car/13442069/

This is one reporter's experience with Waymo's self-driving taxis. Not mentioned in Veritassium's Waymo video, but shown in this one, is a massive, behind-the-scenes, human monitoring of all the Waymo taxi in operation, because, well, the AI is stupid.

This is a fun one

https://www.youtube.com/watch?v=nKJlF-olKmg

It shows only a tiny fraction of the ways that AI tries to "game" the human trainer. Lots of good lessons to be learned here.

The moral is that rather than an evil, "Humans must die", Skynet, we're more likely to wind up with it killing humans simply because that ups its rewards in the reinforcement learning algorithm.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[resqcapt19]

Already causing problems for emergency services.

https://www.fireapparatusmagazine.c...robotaxis-that-interfere-with-fire-apparatus/

 

[wayne440]

Seems to me like a case for adding substantial bumpers to emergency vehicles.

 

[EdStainless]

The problem in the city is that you can't just push them out of the way, there is no place for them to go.
In the 'burbs they would just push it up into a yard and drive on.
They should tow and impound them for the violation though.
That might get it issue addressed.

= = = = = = = = = = = = = = = = = = = =
P.E. Metallurgy, consulting work welcomed

 

[SWComposites]

Simple way to solve issues with “self driving”, including Level 3 and up automation:
All liability for any damage etc is born by the vehicle manufacturer, regardless of driver or passenger actions.
Each instance of vehicle stopping and blocking roadway - $100,000 fine to manufacturer.
Each instance of impeding emergency, fire, police vehicles or personnel - $1,000,000 fine to manufacturer.
Any instance of injury to any person -$1,000,000 + damages.

 

[wayne440]

That would probably solve the issue by instantly eliminating the production of "self-driving" vehicles for use on public roadways.

 

[SWComposites]

Exactly.

 

[TugboatEng]

Humans find even more creative ways to die than self-driving cars will ever be able to reproduce.

The key is to build a road that is compatible with a self-driving car. Either that or certify existing roads. However, the roads should be test driven and the car trained to work with the road (including where to pull off or pull forward when approached by an emergency vehicle).

 

[enginesrus]

SWComposites

Good idea. Like it.

Also people affected or injured by them, should be able to fine them as well.

 

[SWComposites]

Its not possible to predetermine where emergency, fire, police vehicles will stop or need to go. Nor is it possible to predetermine where construction will take place requiring road alterations.

And how will they deal with traffic signals that are inoperative? And a bunch of other non nominal conditions?

And why are all self driving cars being tested in CA cities and Phoenix? Because the weather conditions are really easy. Good luck getting one to work well in a midwest whiteout blizzard. Or in very heavy fog. Or when roads are completely iced over. And, no, its just not practical to shut them down when conditions get bad.

 

[CWB1]

And why are all self driving cars being tested in CA cities and Phoenix? Because the weather conditions are really easy. Good luck getting one to work well in a midwest whiteout blizzard.

They're not. The overwhelming majority of automotive development stateside is done in SE MI. Silicon Valley startups usually spend their first few years blustering about "CA vs Detroit" bc the publicity drives stock prices but none succeed without a local office in Detroit. We've had automated taxi and delivery services locally for years.

 

[stevenal]

So how do they handle Detroit slush obscuring the lane markings?

 

[SWComposites]

I thought GM Cruise was only testing in San Fran.
And Waymo is gong to test in LA.
And Uber was testing in Phoenix.
I can't imagine any autonomous cars being tested in midwest winters at this point.
And last I checked Tesla doesn't have a Detroit office.

 

[EdStainless]

A number of these companies don't have MI offices, but they hire firms to do the work for them.
The reason is the vehicle infrastructure is so good.
In all of these locations they are only operating at specific times.
None of these systems are ready for 'primetime'.

= = = = = = = = = = = = = = = = = = = =
P.E. Metallurgy, consulting work welcomed

 

[1503-44]

It gets even better.

https://www.industryleadersmagazine.com/faa-certification-for-the-first-flying-car-prototype/

--Einstein gave the same test to students every year. When asked why he would do something like that, "Because the answers had changed."

 

[SWComposites]

That Alef "thing" is absurd. I see no way that will ever be certified for either the motor vehicle standards or the FAA flight standards.

The FAA certification is only for a special airworthiness certificate, allowing for limited purposes that include exhibition, research and development. Meaning the company is allowed to (try to) fly it as long as they only kill their own pilot.

 

[debodine]

SWComposites and others, I tend to agree that certification is a LONG way off. Without some kind of very rigorous and automated method of creating "timing and spacing controlled lanes" in the sky, all I can see is drunks, druggies and texters killing others in 3 dimensional space instead of just 2 dimensions. I admit I am a sceptic for systems with humans in the control loop.

However, knowing how technology advances, I suspect some or all of these issues may someday be mitigated enough to make vehicles like this feasible for common travel.

 

[SnTMan]

Only "20 years" away

The problem with sloppy work is that the supply FAR EXCEEDS the demand

 

[SWComposites]

just like fusion power, its been "25 years away" for the last 50 years.

the Alef thing isn't going to get certified as it never going to be controllable, has no systems redundancy, and won't pass air vehicle or ground vehicle crash requirements, etc.

the whole autonomous air taxi air space management and control issue is another large problem that will need to be solved, somehow.

 

[drawoh]

This is up on MSM.com, but it comes from Road and Track.

People on TikTok Are Disabling Self-Driving Cars in SF by Putting Traffic Cones On Their Hoods

--
JHG