Here's What's Missing from Self-Driving Cars: TRUST

[verygreen]

And remember, Audi's system is not released, and no laws are in place right now that would allow it.

This is not true.
There are numerous states in US that allow cars that drive themselves.
I only looked into what TN did in some detail and it specifically allows cars that drive themselves and humans inside (must be at least one) do not need to be licensed to drive, could be minors. The autopilot is the driver for the law purpose, the accidents could be reported by manufacturer staff after getting a report from the car that it got into an accident and such.
Here's the whole bill: https://openstates.org/tn/bills/110/SB 151/documents/TND00072921/

 

[NerdUno]

Funny that Cadillac didn't seem to need regulatory approval for today's Super Cruise drive by CNBC...

General Motors shows off new hands-free ’Super Cruise’ system in highway demo

 

[SomeJoe7777]

This is not true.
There are numerous states in US that allow cars that drive themselves.
I only looked into what TN did in some detail and it specifically allows cars that drive themselves and humans inside (must be at least one) do not need to be licensed to drive, could be minors. The autopilot is the driver for the law purpose, the accidents could be reported by manufacturer staff after getting a report from the car that it got into an accident and such.
Here's the whole bill: https://openstates.org/tn/bills/110/SB 151/documents/TND00072921/

1. Is this bill a proposed law, or did it actually pass and get signed?
2. Is the bill applicable to all autonomous vehicles in general or only to those vehicles/companies that have been pre-approved in some fashion? e.g. Google had to get special permission to do tests of their autonomous vehicles.

To clarify what I said earlier, the laws in some states probably would not prohibit a level 3 or higher autonomous vehicle from being sold or driven there. But I'm fairly certain you would need changes in other laws to shift the responsibility for an accident from the human driver to the vehicle's automatic systems (and therefore the company who built it). To me, if you really want to be able to read a book while the car drives, that law that shifts the responsibility has to be in place otherwise you're taking a gigantic risk.

 

[whitex]

And remember, Audi's system is not released, and no laws are in place right now that would allow it. Are we going to have the same zealous outrage on this forum when Audi misses this "end of the year" promised deadline?

Has Audi sold people cars and charged them $8,000 for those features telling them they are expected to work at the end of the year? If not, then why would people be outraged? Selling a $100K car promising it will do something and then not delivering on it in a timely manner is what causes outrage. When you take people's money, that's when they get mad.

 

[verygreen]

1. Is this bill a proposed law, or did it actually pass and get signed?
2. Is the bill applicable to all autonomous vehicles in general or only to those vehicles/companies that have been pre-approved in some fashion? e.g. Google had to get special permission to do tests of their autonomous vehicles.

Passed and signed in May I believe. Based on this Knoxville, TN is getting some self-driving buses (as a pilot for now) later this year.

In order for a manufacturer to participate in a SAVE project, it must submit a letter to the department prior to operating any ADS -
operated vehicles on the public roads or highways. The letter required by this section must include the geographical areas in which the participating fleet will operate and provide a certification by the manufacturer that:
The vehicles in the participating fleet are owned or controlled by the man ufacturer and are equipped with all of the following:
(A) An automated driving system;
(B) Automatic crash notification technology; and
(C) A data recording system ...

(the owned part is moot I think since it could be argued that car under ADS control is under control of the manufacturer).

"SAVE project" means an initiative by a manufacturer that makes ADS - operated vehicles available to the public for operation on the public roads and highways of this state in a manner that is determined by the manufacturer that owns the vehicle and that is in compliance with this chapter. A SAVE project includes, but is not limited to, making an on - demand ADS - operated vehicle network available to the public.

 

[Canuck]

I will continue to use Tesla AP mainly for testing purposes and to satisfy my technological curiosity.

So what percentage do you use it for that purpose? That's probably the only reason I'd use it too but since I don't have it I don't know how often I would actually use it.

I wouldn't trust it to drive me around while not paying attention, however.

Nor should anyone, ever. Even a momentary lapse of attention could kill you...

 

[whitex]

An interesting article of why Google (now Waymo) decided Level 3 autonomy is practically unsolvable, so they moved to skip it altogether.
Robot Cars Can’t Count on Us in an Emergency

 

[oktane]

So what percentage do you use it for that purpose? That's probably the only reason I'd use it too but since I don't have it I don't know how often I would actually use it.



Nor should anyone, ever. Even a momentary lapse of attention could kill you...

I use it on the highway just secretly hoping one day it knocks my socks off in performance. I'd say about 20% of my miles I drive on AP2 with my hands on the wheel. It doesn't reduce stress for me since I don't entirely trust it. I'm hoping in the next 5 years to have a car I can nap in on the way to/from work. I consider this a slight preview so it is entertaining to me to test, but I know it's not reliable to be a serious feature yet.

 

[Canuck]

It doesn't reduce stress for me since I don't entirely trust it.

That's what I felt when I only had it for a week but people told me I needed to give it more time. However, I don't see how more time would have made it any better since I would still have to pay constant attention regardless of how comfortable I got with it.

 

[whitex]

That's what I felt when I only had it for a week but people told me I needed to give it more time. However, I don't see how more time would have made it any better since I would still have to pay constant attention regardless of how comfortable I got with it.

Speaking from my own experience, more time does reduce the stress of driving with AP1. But that is precisely where the danger lies - you end up complacent and miss when autopilot does something wrong (or doesn't do something right). At first it was incredibly stressful and I would white-knuckle the entire drive ready to take over any time (did close to 60 miles once without touching the steering wheel - that was before the nags, but it was an exhausting experience). Then few months later I caught myself not paying enough attention, so I stopped using it other than stop and go traffic. An interesting thought from the google article I linked earlier - "What if AP was so good it only required user interaction every 10,000 miles?" Would people really be ready to take over in a split second if they only had to do that once a year? If not, are you willing to have an accident every 10,000 miles? The ironic truth is that the better Tesla's AP gets, the more dangerous it gets.

 

[MarcusMaximus]

That's what I felt when I only had it for a week but people told me I needed to give it more time. However, I don't see how more time would have made it any better since I would still have to pay constant attention regardless of how comfortable I got with it.

I still pay constant attention, but despite that, it's a lot less taxing to do long trips. The act of doing minor steering corrections isn't a big deal on short trips, but when driving multiple hours on end, it builds up(much like having to constantly adjust pressure on the pedal when driving without cruise control).

A good example is the trip between San Jose and San Diego. Without AP, that drive is hell. Since getting full speed AS, I've done that trip a few times and it becomes easy and I get in feeling good.

 

[Canuck]

A good example is the trip between San Jose and San Diego. Without AP, that drive is hell.

I assume that's because a lot of it is in stop and go traffic?

 

[oktane]

Agreed best use is stop-and-go, where even if it malfunctions and there is a crash, there may be expensive damage to cars but occupants probably OK. Complacency at high-speed can be a big problem, since conditions change quickly at high speed - even a momentary lapse in attention can be fatal.

 

[MarcusMaximus]

I assume that's because a lot of it is in stop and go traffic?

There's that if you hit LA at the wrong time, but more just endless almost empty straight freeway through the central valley. For me, at least, paying attention the whole time isn't a big deal but making all the constant small adjustments to steering for that ~5 hours gets really tiring. With AP2 I may not be able to just set it and forget it(yet), but at least my physical interactions with it are much more rare.

 

[Bladerskb]

As noted before, Tesla's "beta" system is better than other "non-beta" systems from other manufacturers. Other systems will also happily throw you into the oncoming lane, and keep fully in that wrong lane too (not just head towards the lane as in AP).
Fatal Flaw in AP2 Design

I would buy a "beta" product (marketing sense) over "non-beta" products that worked horribly. I would not buy a beta product (as in the traditional software engineering definition without the quotes).

But I don't like analogies when there are direct examples: the Mercedes Drive Pilot level 2 system is "non-beta" but it will put you directly into the oncoming lane and keep you there (as I linked above in the Norwegian test; another review linked below said the same). Just because the other automakers don't put the "beta" label doesn't mean their product works better.
The War For Autonomous Driving: 2017 Mercedes-Benz E-Class VS. 2017 Tesla Model S

Are you kidding me right now? Tesla AP 2 does the same thing. EVEN WORSE.

There are dozens of fail videos where AP has you comfortably in the other lane repeatedly.
Seriously you tesla fans are killing me with your buffoonery.


Volvo Pilot Assist 2 is FAR better than AP2 and equals and surpasses AP1 in many avenues.

Here comes your excuses and patented deflections in 3...2...1

 

[mmd]

Funny that Cadillac didn't seem to need regulatory approval for today's Super Cruise drive by CNBC...

General Motors shows off new hands-free ’Super Cruise’ system in highway demo

This is impressive, more so that the Cadillac president is doing the demo at 73 miles an hour on curvy roads of I-280/
But I don't get how the lidar mapped data is going to help when a road hazard appears, like stopped vehicle partially blocking lanes, someone cutting in front, animals, lane diversion, water logging under overpasses, etc. Hopefully the software is mature enough. An accident can happen in a few seconds, so the alert after a few seconds of distraction may not help. GM did have more autonomous miles under their belt in California autonomous driving. Still, I remain fearful.

GM seems to be getting pushed to the edge by Tesla's bold marketing to bring this to market.

 

[Cosmacelf]

There is a very simple and logical reason to believe certain companies will be further along than Tesla in autonomous driving. They have been at it longer and taken a very serious and thorough route to it.

Don't confuse the driver aids with the autonomous projects of those companies, they are absolutely separate.

I don't know all companies and their efforts, but here is e.g. Audi:

I believe it was around 2010 or so, when Audi first showed their autonomous driving system - it took the whole boot, but it was the system that will ship in the new Audi A8 later this year.

Audis did coast to coast already in 2015, though not all the city parts. But that was in 2015.

What I've seen from Audi over the years, already many, many years ago, is infinitely superior to Tesla's FSD video. I simply trust what I've seen a lot more than the Tesla video.

I disagree. Tesla has learnt a lot simply because they have a released product. In addition Elon also has the foresight to fund a research AI company called OpenAI, which they are already drawing upon. Autonomous driving is a very hard unsolved AI problem. Which companies have actually solved basic R&D problems in the last 5-10 years. SpaceX and Tesla come to mind, the other auto makers do not.

 

[Cosmacelf]

Funny that Cadillac didn't seem to need regulatory approval for today's Super Cruise drive by CNBC...

General Motors shows off new hands-free ’Super Cruise’ system in highway demo

Thanks for the link. Well worth the time to view it. It looks like Cadillac will be at AP1 parity this fall, with the additional safety system of an eye tracking system. Kudos to GM. I believe GM is using Mobile Eye as well.

 

[Bladerskb]

Now wait a minute, was the video a lie or was the video a mistake because the AEB system wasn't installed on that car? i.e. Either someone published that video as a joke or to be funny, having staged the entire thing, OR the salesman was legitimately attempting to demo the AEB system, didn't realize that the car he was demonstrating it with was not equipped, and therefore got hit. Just clarify for me what the truth is.

The later

Furthermore, I still haven't seen a Volvo lane-keeping system that I could actually use on the interstate, which is the main point here. If there's a video of it that shows how it can match AP1, please let me see it.

Its a myth that tesla autopilot doesn't have any competition and this is why Tesla continues to be a breeding ground for myths.

For example 2017 Volvo Pilot Assist 2 (It even handles curves better than AP1 and AP2). There's no dumb braking and not recognizing car infront, etc... Obviously Volvo doesn't have the cult following that Tesla has so there are no thousands of video of their pilot assist 2 randomly driving around.

rain/snow/sleet

local road

huge bend

I'm not going to be called a liar or be accused of "making stuff up". The figure I cited is exactly what I read. If that's inaccurate, so be it, but that's not my doing.

Audi's own web site states 35 MPH "under certain conditions": Audi Newsroom

And remember, Audi's system is not released, and no laws are in place right now that would allow it. Are we going to have the same zealous outrage on this forum when Audi misses this "end of the year" promised deadline? Or when it's limited to a lower speed than the "promised" 35 MPH upon release? We'll see.

What you said weren't even close which means you completely fabricated it. Audi system comes out next month, infact actually in just 8 days. Again as pointed out by others, car companies outside of tesla don't sell promises for $$$$ thousands of dollars while the customer don't see anything even resembling those promises till 1-2 years later. Tesla is the only one selling promises, others actually sell finished products.

Is that so? Where are all the deaths? Tesla's AP has been used for well over 100M miles of driving and to date, there is one fatality that in my opinion was due mostly to distraction, not the hardware, software, or design. Where is your evidence that AP will "kill you", because 100K Tesla owners have 100M+ miles of AP driving that disagrees with that assessment.

2 people have already been confirmed dead at the hands of autopilot, thats two too many. Also there have been hundreds of AP accidents. Many have been posted in this forum and on the tesla subreddit.

 

[mmd]

I disagree. Tesla has learnt a lot simply because they have a released product. In addition Elon also has the foresight to fund a research AI company called OpenAI, which they are already drawing upon. Autonomous driving is a very hard unsolved AI problem. Which companies have actually solved basic R&D problems in the last 5-10 years. SpaceX and Tesla come to mind, the other auto makers do not.

It's not necessary nor is it necessarily better to build everything by yourself. Automakers, especially smaller ones, can just buy the system from other OEMs, like Waymo, Delphi (who did LA-NY self driving demo last year) and others. It's the IBM vs. Microsoft-Intel model. Componentization is the key .
Automakers can focus on building solid, reliable cars most efficiently while self driving OEMs can build the self driving part.

Here is one from Waymo more than a year ago:
Google’s artificial intelligence breakthrough may have a huge impact on self-driving cars and much more

Google’s self-driving cars just got way better at driving themselves