Move Over Tesla. Here Comes Cadillac.

[Bladerskb]

This document was useful, I think people were looking for this in other threads discussing the different levels.

However, I did notice in the document it says that L3 does disengage on page 20:
"• Disengages an appropriate time after issuing a request to intervene
• Disengages immediately upon driver request"

SAE doesn't seem to use "handover" in its terminology. Rather it talks about "request to intervene", and disengagement happens after request to intervene. So SAE's document seems to match @EinSV's characterization. The "handover" is a valid disengagement event, even under L3 (and SAE covers both cases, where driver manually disengages or the system disengages after the timer expires).

It doesn't match his characterization at all though.
I have always said that a Level 3+ system never immediately disengages by it self. He is saying it does disengage immediately by itself.

Since we are talking about "disengagement" as refereed to in the CA DMV regulations, then a level 3+ car doesn't do that. It however hands over the control of the car when the driver either takes over or when it request for the driver to take over after giving them adequate time.

This contrasts the CA DMV definition of disengagement. Its not that the DMV is wrong its that the CA DMV regulation is for test cars which are NOT L3 and only L2 and which all immediately disengage. The SAE document however is not referring to test cars but actual designated level cars.

CA DMV disengagement statement include immediate events induced by the system and like i said "there are no system induced immediate disengagement in a Level 3+ car, these are system failures."

Handover is NOT a valid disengagement event as characterized by the CA DMV regulations as it describe an immediate disengagement event based solely on safety reasons. In the SAE however handover happen for other reasons and are not immediate in L3 cars.

The word "disengage" is a blanket word which has completely two different meanings in the SAE document versus the CA DMV rules.
One is talking about deployed cars while one is talking about test cars.

The SAE matches my statements 100%. Also request to intervene = handover, its a more appropriate term the industry uses than to say "request to intervene".

Probably talking about this:
"When in doubt, however, the CT6 defers to analog driving. For long stretches between New York’s Cadillac House and Washington, on pristine sections of the New Jersey Turnpike, the system stubbornly declined to launch. "
Cadillac Finally Has an Answer to Tesla’s Autopilot

Yeah but that's one article...

 

[malcolm]

Tesla claims it doesn't need LIDAR but I don't think there's any car companies that would agree that at all.

I think there's a marketing advantage in being able to claim that a system either uses LIDAR-derived terrain scans or the vehicle itself is fitted with LIDAR units.

But the challenge of being able to process data in real time and track moving and changing potential hazards (other vehicles, the state of traffic lights, pedestrians) is much the same, whether you use cameras or LIDAR. The data still has to be crunched in meaningful ways and every sensing device always produces unique artifacts which must be dealt with as part of that process.

It's no surprise that these early systems are low-speed and freeway-based since this is the most easily predictable environment. And LIDAR is overkill for these situations.

A 3D point-cloud sounds high-tech and futuristic, but you still have to work out where the other vehicles and the pedestrians are travelling. Plus, it gives you nothing on traffic lights and road signs.

 

[stopcrazypp]

It doesn't match his characterization at all though.
I have always said that a Level 3+ system never immediately disengages by it self. He is saying it does disengage immediately by itself.

I don't think he was making an argument at all that a L3 system would immediately disengage by itself. His issue was with the claim that a handover event is not disengagement and also your claim that a L3 system never disengages:

WRONG!

The car gives you back control when you reach the end of an exit lane which you are still on.
This is an actual feature not an disengagement, think of it as a handover.

To help you out, a L3 car never disengages.
You need the read the SAE on what handover and disengagement are.


1,000 miles without disengagement is disengaging freq? or 40 miles from another reviewer?

The SAE document seems to contradict this as it specifically says L3 cars can disengage and not only that, it explicitly says that a handover (or "request to intervene") is a disengagement event:
"Level 3 – Conditional Driving Automation...
• Disengages an appropriate time after issuing a request to intervene
• Disengages immediately upon driver request"

However, I don't think the original conversation really requires this issue over terminology to be pushed that far. For the purposes of evaluating a level 2 system, the percentage the system remains engaged matters more. For example, a system that can safely drive through a construction zone should be evaluated higher than one that does a "handover" before reaching the construction zone. In the comparison tests I linked before, I imagine the "handover" would also be counted as a interruption (note I'm deliberately not using the term in question), where an interruption would simply be a change of state from engaged to disengaged.

 

[Tam]

...LIDAR is overkill for these situations....

My impression is Radar still has the problem of detection since world war 2 as it could mis-interpret thousands of tiny harmless aluminum chaff as big and dangerous enemy airplanes.

Tesla updated the concern with the blog and explained it with and example of a harmless soda concave bottom that could be misinterpreted by Radar as something big and dangerous.

Radar cars still have a problem of avoiding rear-ending into a stopped vehicle in front from time to time.

LIDAR folks say that's not a problem and it could easily avoid the fatal Autopilot in Florida.

 

[AnxietyRanger]

I disagree. For example, if you design a system around a rotating lidar (or multiple) mounted high on the vehicle vs ground level the door mounted lidar sensors in the Nissan vs mid mounted bumper lidar sensors in the Delphi, switching could add years of integration work. The vehicle and object signatures can be quite different (as well probabilities of occlusions).

I think the ones serious about fielding a consumer ready system needs to already be thinking about sensor integration (Nissan may be one example).

Of course you disagree, how could it be any other way.

Put let's not waste time on disagreements on what happens behind the scenes - those are mostly driven by biases at this stage anyway, due to lack of information - and look at a different angle:

Could large roof-mounted suites work?

Google/Waymo and Uber, neither probably are looking at the consumer car sales market in the first stage. I can totally see self-driving taxis work well with a roof sensor suite - and those can be more easily retrofitted to cars built by others as well, in the case of non-car-companies. Such a large and visible sensor suite might even add to the appeal and confidence in first self-driving commercial fleet cars.

As for whether or not consumers would accept a roof-mounted suite like that, I'm not 100% confident they wouldn't. If I could buy a Level 5 car today and getting that meant I needed to accept a massive Uber suite on my roof, I would be at the dealership at this minute instead of writing this message.

It is the Level 5 thing that would be such a massive differentiator that I would accept the car looking different.

 

[AnxietyRanger]

Frankly, looking at the latest exchanges, I think this thread has taken another turn to partisan bickering where benefit of the doubt is only given to those one "one's" side.

For example, it is quite easy to see both Bladerskb's and EinSV's real points underneath the crud, but neither seem truly willing to transcend their differences and reconcile. Pointless. And the accusations about "team accounts" persist - even more pointless.

For Supercruise, examples of failures are sought and ignores at a frequency 100% related to one's position on the partisan divide. How optimistic or pessimistic one is seems to depend completely on this breech.

Same with the sensor minituriazation. FUD about the viability of Tesla's competiton's approach is easily thrown around, contrary examples like Nissan are curtly acknowledged or not acknowledged at all. Pointless.

I think we are better off admitting that for an active core of participants, our biases are too strong to get anywhere sensible in a convo like this - even the conversation itself just feeds those biases, because the need to counter the bias of others just adds to our own.

I will try to steer clear of adding to my own biases by bickering in a thread like this. Of course I will fail.

 

[lolachampcar]

Ok, I really do not like Cadillacs. There, I said it

Now that my bias is out of the way, does anyone here think Cadillac would have done what they have done had it not been for Tesla pushing the envelope soooooo hard? I for one do not. Tesla is too small to "convert the world" but what they seem to be doing is showing that the previously impossible by then common standards is indeed possible; its just not easy. There are now multiple BeVs with decent utility. Autopilot seems to be following the same path. Even if Tesla goes away one day their legacy alone will have been worth the effort.

 

[lolachampcar]

Oh, and let me drop a floater in the punch bowl.

As non PC as this is, I would prefer we loose a few people to accidents where bleeding edge technology is mis-used while advancing the technology then putting padded walls around everything and moving forward at a snail's pace. There are extremes but I'm not talking about those; I'm talking about reasonable risk taking in the center of the curve.

Sure we will loose a few people to the Darwin award process but just think of the lives saved by moving FSD forward even six months! That far outweighs small losses along the way. I realize that is a brutal way of looking at life but it is also a compassionate way of living life.

 

[AnxietyRanger]

Oh, and let me drop a floater in the punch bowl.

As non PC as this is, I would prefer we loose a few people to accidents where bleeding edge technology is mis-used while advancing the technology then putting padded walls around everything and moving forward at a snail's pace. There are extremes but I'm not talking about those; I'm talking about reasonable risk taking in the center of the curve.

Sure we will loose a few people to the Darwin award process but just think of the lives saved by moving FSD forward even six months! That far outweighs small losses along the way. I realize that is a brutal way of looking at life but it is also a compassionate way of living life.

I don't dislike this view. Frankly, I guess if I really had an opinion on it, I would agree. Haven't thought about it too much, but this message got me thinking...

I guess there is the point, though, that what if those bleeding edge losses end up setting FSD back a decade instead... because everyone becomes super-careful after some visible losses and lawsuits that follow?

Or, even worse, what if an Autopilot accident were big enough to be a mission killer for the likes of Tesla? Would it be worth it to set back BEV progress at the same time, just for the sake of Autopilot progress?

How much did the Brown Incident play a part in the following?
- AP1 being more restricted today than a couple of years ago
- MobilEye ditching Tesla, making AP2 less today than it could have been...
- ...and setting back Tesla's progress as FSD resources likely were redirected to EAP work
- OTOH, Cadillac adding belatedly driver monitoring to Supercruise

What if Brown Incident would have instead been The Browns Incident with a family of seven on board?

So, I guess a healthy balance is called for... the trick is to find the right one. Luck, in the end, will probably play a part...

 

[lolachampcar]

AR,
More than balance, I'm a fan of proper assessment of risk. Somehow we have forgotten that life is full of it (risk).
One simple example is terrorism. People are all wigged out over it yet the odds of being exposed to it are tiny in comparison to being involved in a car accident. It makes no sense. In addition, all that wigging out has the desired affect thus increasing the risk. Go figure

 

[EinSV]

@stopcrazypp, thank you for clarifying the situation with Level 3 disengagements. As you note, Level 3 is a side issue here where the focus is Level 2 but confirmation that the SAE, like CA DMV, defines system-initiated handovers as disengagements, contrary to the incorrect statements from the @Bladerskb team, is helpful.

I also think your description of the relevant standard for this aspect of evaluating Level 2 systems is spot on:

However, I don't think the original conversation really requires this issue over terminology to be pushed that far. For the purposes of evaluating a level 2 system, the percentage the system remains engaged matters more. For example, a system that can safely drive through a construction zone should be evaluated higher than one that does a "handover" before reaching the construction zone. In the comparison tests I linked before, I imagine the "handover" would also be counted as a interruption (note I'm deliberately not using the term in question), where an interruption would simply be a change of state from engaged to disengaged.

 

[McRat]

Ok, I really do not like Cadillacs. There, I said it

Now that my bias is out of the way, does anyone here think Cadillac would have done what they have done had it not been for Tesla pushing the envelope soooooo hard? I for one do not. Tesla is too small to "convert the world" but what they seem to be doing is showing that the previously impossible by then common standards is indeed possible; its just not easy. There are now multiple BeVs with decent utility. Autopilot seems to be following the same path. Even if Tesla goes away one day their legacy alone will have been worth the effort.

First press demonstration of Super Cruise early 2012. First announcement of Autopilot 2013?

By the way, is a Tesla Model S even safe aerodynamically at 200mph or will it spin, lift, and roll like a Insight or many other low Cd designs do? What keeps the frontal pressure from spinning the car at high speeds? GM sells 3 models that are good to 200 mph or higher without risk of aero failure. One is a Cadillac.

 

[AnxietyRanger]

I find if I keep both hands on the wheel the "forces" cancel themselves out and the car will nag me. I've adopted a one handed approach, with my hand more or less at the 9 or 3 position. The resting weight of just one hand seems to keep it happy most of the time.

An update:

This was a good tip from @Pale_Rider, I have been able to lessen the amount of nags I get significantly by using just one hand - when I dare. The tip does work.

Now, of course, this is counter-intuitive. I would assume Tesla would like me to have both hands on the wheel... hope they find a solution.

 

[stopcrazypp]

Of course you disagree, how could it be any other way.

Put let's not waste time on disagreements on what happens behind the scenes - those are mostly driven by biases at this stage anyway, due to lack of information - and look at a different angle:

Could large roof-mounted suites work?

Google/Waymo and Uber, neither probably are looking at the consumer car sales market in the first stage. I can totally see self-driving taxis work well with a roof sensor suite - and those can be more easily retrofitted to cars built by others as well, in the case of non-car-companies. Such a large and visible sensor suite might even add to the appeal and confidence in first self-driving commercial fleet cars.

As for whether or not consumers would accept a roof-mounted suite like that, I'm not 100% confident they wouldn't. If I could buy a Level 5 car today and getting that meant I needed to accept a massive Uber suite on my roof, I would be at the dealership at this minute instead of writing this message.

It is the Level 5 thing that would be such a massive differentiator that I would accept the car looking different.

I find it hard to have a conversation when every post you just trot out the standard "bias" line.

As for large roof-mounted suites, sure, it would work fine for taxis. For those, height clearance, roof racks, and aero matters less because it will all be managed by a third party. However, for consumer vehicles, I don't find it likely they will be acceptable.

Given you have a Model X, I'm sure you are no stranger to height clearance issues. The X height is ~85 inches (216 cm) when doors fully open. This is tight for some cases, but at the very least you can enter the garage first while the car is only 66 inches tall and clear the garage door and open when inside (where there is more clearance) The standard garage door height in the US (including mine) is 7 feet tall (84 inches).

The Waymo Pacifica is about 90 inches tall (estimated from straight on photo and known height of 69.9 inch for standard Pacifica). It won't be able to get into most home garages.

I think it's telling that the automotive suppliers (for example Delphi, Bosch, Magna, Continental) are working on sensors that can be integrated into the car body, not on ones you just slap on the roof and call it a day.

(Note: Magna L3 Cadillac ATS in front, Continental L3 Chrysler 300 in back)

Magna MAX4

 

[AnxietyRanger]

I find it hard to have a conversation when every post you just trot out the standard "bias" line.

Don't mind me - it is just a personal epiphany I've had recently and wanted to share. Maybe it helps someone put something in perspective. It is just so... expected once you see it. The responses from us. Of course you would disagree, right... I'm sure I have my moments of the opposite.

As for the topic, of course manufacturers are working on sensor minituriazation and integration - and that has a value in the consumer market. IMO that is pretty much everyone except some self-driving pioneers with roof kits...

That said, if you don't take the highest possible car as an example - and say, slap that Uber kit on a Model S - I'd buy one today if it did Level 5. I'm pretty sure I wouldn't be alone.

 

[lolachampcar]

McRat,

Lots of people have been working on lots of things. The question for me is has Tesla's (possibly reckless) approach pushed others to do more sooner? If only looking at the BeV I think it is clear they have. I just extended the possibility that the same can be said for autopilot type systems.

I concede that other cars, including GM, are more stable at speeds unachievable by Tesla You win.

 

[AnxietyRanger]

McRat,

Lots of people have been working on lots of things. The question for me is has Tesla's (possibly reckless) approach pushed others to do more sooner? If only looking at the BeV I think it is clear they have. I just extended the possibility that the same can be said for autopilot type systems.

I concede that other cars, including GM, are more stable at speeds unachievable by Tesla You win.

Ironically (from an "Level 5 capable" FSD-seeking perspective that is) Bladerskb made recently the point that the "reckless" approach has made high-speed Level 2 the target - instead of more robust, but perhaps more limited-scenario Level 3...

I'm sure it is not just Tesla that has contributed to this, after all the talk of waiting until Level 4 is not new, but IMO it would be unfortunate if we don't see Level 3 cars from many manufacturers in the meanwhile.

Level 3 is just a different beast... even a nicely working Level 2 is no substitute. So, too bad if the industry is pivoting towards Level 2 instead. The best compromise, of course, would be great Level 2 with increasing amount of Level 3 modes as well... until we get to Level 4+.

 

[d21mike]

This is why I got the Tesla. Audi A8 Level 3 up to 37 mph. This is the 2019 car (not now) and this is what you can buy then. Not that impressed. To me, my Tesla would have not problem is this traffic jam situation.

2019 Audi A8 Level 3 self-driving real world test

 

[lunitiks]

Some people fall asleep with their eyes open. Does Super Cruise account for that?

 

[Phrixotrichus]

This is why I got the Tesla. Audi A8 Level 3 up to 37 mph. This is the 2019 car (not now) and this is what you can buy then. Not that impressed. To me, my Tesla would have not problem is this traffic jam situation.

2019 Audi A8 Level 3 self-driving real world test

another one who doesn`t get what a massive difference the fact makes, that audi takes RESPONSIBILITY during the time the system is active. The difference between lvl 2 and 3 is enormous.
The failure rate to be able to do that has to be incomparably better than what the current Tesla AP delivers.....