Autonomous Car Progress

[S4WRXTTCS]

Kind of a weird answer there...

"Sure You can pick any address, it just might not actually go to that actual address...."


Which sounds a lot like the Waymo behavior previously reported where it'd just refuse to let you be picked up at "hard" locations and instead you'd have to walk a little bit to get to an "easier" pickup.

SF has a lot of rules regarding where you can be picked up so it makes sense to me.

 

[stopcrazypp]

I was being sarcastic. I guess the eye rolling emoji is too subtle. haha.

I got the sarcasm, but the bus example doesn't really address the point either. Buses have regularly spaced designated stops that allow for easy loading and unloading of disabled passengers without blocking traffic, but ride share cars are not allowed to use them, so it's moot for them even if they exist for buses.

Rideshare cars also don't get the same exemptions that taxis do for loading/unloading such passengers (for example when it stops at bike lanes).

Double parking is selectively enforced, as it should be. If you double park on a street with light traffic I doubt SFPD would give you a ticket. If you do it on a thoroughfare and hold up traffic you might. When I take an Uber I walk somewhere appropriate for pickup. Until the Tesla Bot can put you in a wheelchair and push you to the vehicle I don't think picking up passengers anywhere is a realistic expectation.

The issue is SFMTA has zero tolerance for self driving cars breaking such rules, as per their dispute with Cruise. Basically they can lay their foot down and say you can't even program the car to allow it to do that. A similar thing can be seen in the whole rolling stop recall for Tesla. So even though Uber/Lyft and other drivers can get away with it, the self driving cars can't.

 

[Daniel in SD]

I got the sarcasm, but the bus example doesn't really address the point either. Buses have regularly spaced designated stops that allow for easy loading and unloading of disabled passengers without blocking traffic, but ride share cars are not allowed to use them, so it's moot for them even if they exist for buses.

Rideshare cars also don't get the same exemptions that taxis do for loading/unloading such passengers (for example when it stops at bike lanes).

The issue is SFMTA has zero tolerance for self driving cars breaking such rules, as per their dispute with Cruise. Basically they can lay their foot down and say you can't even program the car to allow it to do that. A similar thing can be seen in the whole rolling stop recall for Tesla. So even though Uber/Lyft and other drivers can get away with it, the self driving cars can't.

My point with the bus example was that elderly and disabled people currently travel more than a block to take transportation. It is unlikely that transport for people with extreme mobility issues will be automated any time soon (unless Tesla Bot is on schedule ).
Accessibility issues aren't really autonomous vehicle specific problems. Cruise does plan on making a wheelchair accessible Origin:

Cruise disagrees with the SFMTA. https://www.cpuc.ca.gov/-/media/cpuc-website/divisions/consumer-protection-and-enforcement-division/documents/tlab/av-programs/phase-i-av-deployment-program-al-status/20211206-cruise-llcs-reply-to-protest-and-comments-to-application-for-driverless-deployment-permit.pdf]

GM's Cruise disputes San Francisco concerns on stops, says 'double parking' legal

General Motors Co's self-driving technology unit Cruise on Monday pushed back on claims from San Francisco authorities that its robotaxis are illegally double-parking, saying they have the right to block traffic for quick stops.

www.reuters.com

And they are double parking:

 

[diplomat33]

According to Aurora CEO, we are in the "pre-Model T" phase of autonomous driving:

https://twitter.com/x/status/1521965985267335170

 

[diplomat33]

Mercedes will start selling their L3 system in Germany in 11 days, starting on May 17, 2022:

Mercedes-Benz today released details of the sales launch of DRIVE PILOT in Germany. The system for conditionally automated driving (SAE Level 3[1]) can be ordered from May 17, 2022 as an optional extra for the S-Class for 5,000 Euro and for the EQS for 7,430 Euro (Driver Assistance Package Plus: 2,430 Euro and DRIVE PILOT: 5,000 Euro) excl. VAT. This makes Mercedes-Benz the first car manufacturer in the world with an international valid certification for conditional automated driving, to offer such a system as an option ex-works for vehicles from series production. DRIVE PILOT enables customers to hand the driving task over to the system under certain conditions in heavy traffic or congestion situations on suitable motorway sections in Germany up to a speed of 60 km/h.

Source: Conditionally automated driving: Mercedes-Benz announces sales launch of DRIVE PILOT | marsMediaSite

 

[EVNow]

What happens if people are violent against AVs ?

Since people are getting confused ...

"This is what happens when people get violent against AVs"

https://twitter.com/x/status/1522322748315291648

 

[2101Guy]

imma go out on a limb and guess the same thing that happens when people damage non AV's...arrested and charged.
(is life really this hard?)

 

[EVNow]

Mercedes will start selling their L3 system in Germany in 11 days, starting on May 17, 2022:



Source: Conditionally automated driving: Mercedes-Benz announces sales launch of DRIVE PILOT | marsMediaSite

They give a page long presser but won't give any details about how it was tested or disengagement rates.

Hope they tested for more than 5 minutes as required by Regulation 157 which they term "demanding". What a fraud.

Mercedes-Benz is the first automotive company in the world to meet the demanding legal requirements of UN Regulation No. 157 for a Level 3 system.

 

[Doggydogworld]

What happens if people are violent against AVs ?

They get arrested. Or is this some kind of trick question?

As Tesla owners have repeatedly demonstrated, constant surveillance is a great aid to law enforcement. Privacy advocates are less enthused.

 

[EVNow]

They get arrested. Or is this some kind of trick question?

As Tesla owners have repeatedly demonstrated, constant surveillance is a great aid to law enforcement. Privacy advocates are less enthused.

That was the "lede" for the news item posted.

 

[diplomat33]

They give a page long presser but won't give any details about how it was tested or disengagement rates.

Hope they tested for more than 5 minutes as required by Regulation 157 which they term "demanding". What a fraud.

Reg 157 does not say only 5 mn of testing. Show me the section!

Here is the full text of Reg 157:

https://unece.org/sites/default/files/2021-03/R157e.pdf

It is 64 pages! I consider that pretty demanding.

 

[EVNow]

Reg 157 does not say only 5 mn of testing. Show me the section!

Here is the full text of Reg 157:

https://unece.org/sites/default/files/2021-03/R157e.pdf

It is 64 pages! I consider that pretty demanding.

I've quoted it in this thread ! How hard is it to search ?

If they find 64 pages "demanding" - how demanding would 500 Million miles of testing be ?

View attachment 776312

 

[diplomat33]

how demanding would 500 Million miles of testing be ?

We've been over this before. You don't need to do 500M miles of testing to validate a L3 system that is limited to such a small ODD.

 

[EVNow]

We've been over this before. You don't need to do 500M miles of testing to validate a L3 system that is limited to such a small ODD.

I showed you, you were wrong. Basic math.

 

[diplomat33]

I showed you, you were wrong.

I don't think I am wrong. You really think you need 500M miles to validate that the L3 can drive at max 30 mph on a premapped limited access highway with perfect lane lines and a lead car, and the system warns the driver to take over when needed? No, you don't. For one, that is a very limited ODD. The number of safety cases and edge cases will be smaller. Second, L3 will ask the driver to take over when needed. So L3 does not need to handle every edge case on its own. As long as the system has a reliable way of getting the driver to take over before there is a problem, your system will be ok. I agree you should do more than 5 mn of testing, but you don't need 500M miles.

 

[Dan D.]

I've quoted it in this thread ! How hard is it to search ?

If they find 64 pages "demanding" - how demanding would 500 Million miles of testing be ?



View attachment 801397

In Annex 5 Section 1, It does say that:

Until such time that specific test provisions have been agreed, the Technical
Service shall ensure that the ALKS is subject to at least the tests outlined in
Annex 5.
...
The test specifications in this document are meant to be a minimum set of tests,
the technical service authorities may perform any other test within the system
boundaries

So the tests given are the minimum, they seem hopeful that companies will feel they can do more than the minimum. The section highlighted 4.1.2 refers to test 4.1 having a minimum of 5 minutes. All the other tests in section 4 also have to be completed, none of them having a time specified.

 

[diplomat33]

View attachment 801397

My read of the test is that it is proposing one particular test scenario where the car has to stay in the lane while another car swerves in the lane or another vehicle drives too close. That individual test run would take at least 5 minutes. So you start the run on a track, a minute or two later, you have a car swerve in the lane, a few minutes after that, you have another car drive too close. And you see how your ego car responds. So that test run should last at least 5 minutes. But you would do many test runs that scenario and many test runs for other scenarios as well. The entire testing/validation process would not be 5 minutes.

 

[EVNow]

I don't think I am wrong.

Two things
- We discussed this limited ODD and how many miles you need to statistically prove your disengagement rate w.r.t. Waymo and you agreed that tiny ODD will not reduce number of miles you need to drive. When I have some time I'll search and post the link
- I've already had a long discussion with @Daniel in SD on this subject. We can follow up on that discussion instead of rehashing the discussion

Seriously, here is the question about L3 in consumer cars.

Do Merc & BMW really have production L3 in Germany ? How did they test & confirm it works ? How many miles did their testers drive ?

I think thats going to be an issue ... esp. if they are looking for < 60 km/h conditions and millions of miles of driving.

Same question for Volvo. How can they assure error rates are good if they can't test a million miles ?

The only videos of Merc I can find of L3 are the demo ones in Merc's demo track where they actually recreate slow spped traffic-jam situation.

 

[EVNow]

So the tests given are the minimum, they seem hopeful that companies will feel they can do more than the minimum.

My read of the test is that it is proposing one particular test scenario where the car has to stay in the lane while another car swerves in the lane or another vehicle drives too close. That individual test run would take at least 5 minutes. So you start the run on a track, a minute or two later, you have a car swerve in the lane, a few minutes after that, you have another car drive too close. And you see how your ego car responds. So that test run should last at least 5 minutes. But you would do many test runs that scenario and many test runs for other scenarios as well. The entire testing/validation process would not be 5 minutes.

The problem is - the "tests" mentioned are not comprehensive. To properly test
(1) you need to figure out every permutation of scenarios and test them OR
(2) you just test for millions of miles, because statistically you would have covered a lot of real world scenarios

Since (1) is not possible, responsible companies (which includes every company testing in CA, for eg.) test in real world conditions for millions of miles.

 

[diplomat33]

Two things
- We discussed this limited ODD and how many miles you need to statistically prove your disengagement rate w.r.t. Waymo and you agreed that tiny ODD will not reduce number of miles you need to drive.

There is a big difference between L3 and L4. L3 generally does not need as much testing as L4. So yes, Waymo still needs a lot of miles because of the nature of the AV being L4. Also, the ODD of Waymo is much bigger than the ODD of Mercedes' L3.

Also, disengagements don't mean the same thing in L3 as they do for L4. Disengagement rates matters more for L4 because L4 is expected to handle everything without any human intervention. L3 asks the driver to intervene on purpose and has an entire mechanism for making sure the driver takes over correctly. So disengagements for L3 are not necessarily bad like they are in L4. For example, a disengagement rate of say 1 per 100 miles would be terrible for L4 but could be perfectly acceptable for L3 if every disengagement was a safe one where the driver took over correctly. With L3, you only need to measure unintended disengagements. With L4, you need to measure all disengagements.