New AU FSD vrs USA FSD post June 21

[paulp]

A whole coke? Are we talking a fancy glass bottle one? Such confidence

It would have to be in a tesla decanter at least

 

[Wol747]

>>lidar system is like a set of guarantees on the vehicles position relative to the environment. With a lidar stack, you can guarantee that it won’t hit a curb (assuming HD map is up to date) as it has a ~mm accurate 3D map of where the curb is ahead of time.<<

I would hope the car ISN’T using mapping in that way! Even tectonic plate movement in many parts of the world would make maps invalid at that accuracy quite quickly!
And keeping global mapping current enough would cost billions a year.

 

[Wol747]

>>I find it so funny how people demand that AP improves constantly yet are not willing to accept the cost to this improvement.<<

I *don’t* find it particularly amusing that so many of the things that we’ve paid for either don’t work or are randomly unreliable.
It would be nice if a company that sells something respect its customers sufficiently to at the very least make its present software do what it says on the tin instead of - or alongside - bragging about what will/might come along next week/month/year.

 

[Shock-On-T]

A whole coke? Are we talking a fancy glass bottle one? Such confidence

I think I will stretch to a glass bottle.

 

[GrimRe]

I would hope the car ISN’T using mapping in that way! Even tectonic plate movement in many parts of the world would make maps invalid at that accuracy quite quickly!
And keeping global mapping current enough would cost billions a year.

Let's say there is some civil works that happen in between when the mapping was done and when the car drives. If that civil works extended a curb by a few CMs then all of a sudden you will be clipping that curb every time. That is of course unless you have a camera that looks for the curbs and when it sees that the curb has changed it will adjust the car's path to avoid it. Well then why do even need LIDAR in the first place if you require a back system that must be perfect in case the HD map is out of date?

Lidar is line-of-sight. Can't see through fog or bounce under car in front to see occluded second vehicle.
Anyway, Tesla is not replacing radar with lidar, it's just replacing radar with improved visual software. That improved visual software relies on the cameras, which aren't ready yet.
Let's put it another way: if we ever see a fleet of robotaxis on AP3 I will buy you a coke.

You are just as likely to catch a glimpse of the second occluded vehicle and use a NN to predict its trajectory than you are being able to reliably filter radar returns that have bounced under the lead car.

Tesla are an incredibly data driven company. They would see how many times that radar bouncing under the car feature was actually useful versus noise. My guess is that as NN processing closes in on being superhuman the fragility of radar made it counterproductive in driving policy decisions.

As for the fog, this keeps getting brought up and I really don't understand it. How can you drive properly though fog without vision anyway? You still need to see the road infrastructure (lights, lines, curves etc) and not just the lead car. It would be positively reckless to drive any faster than a skilled human driver in foggy conditions no matter how good your radar suite is. If the lead car all of a sudden pops into existence and the vision system is late to pick it up relative to radar return then its of no consequence as the car is already travelling slow enough for the conditions and is superhuman in response times.

Think about it this way: if you were driving in foggy conditions but you had a heads up display that could tell you the exact distance to a moving object in front of you, you would feel a lot more comfortable driving right? The reason this is of additive assistance is because you are using a human brain to construct the driving policy. Your brain has limited ability to focus on multiple things at once so if all your brain space is being used to try and track a lead car through heavy fog, you will feel very vulnerable to other situational elements (curve of road, traffic controls etc). Being able to offload this tracking to a HUD and free up your brain to focus on whether a stop sign is coming up seems like a very useful ability.

However this is a shortcoming of your brain (not you personally, all humans). If you had the ability to pay perfect attention to every pixel 360 degrees around the car with a ~100ms reaction time (versus ~2500ms), this HUD of the radar response becomes unnecessary noise as your limiting factor is the vision system being able to make sense of the wider environment anyway. It doesn't matter that you can't see the lead car through the fog 50 metres away when you need to be able to read a stop sign that is only 5 metres away. If the lead car does appear 5 metres away you have the ~100ms reaction time to update the driving policy to prevent a collision.

I think I will stretch to a glass bottle.

Are we talking globally or Australia? If globally, what are the time bounds? I think I'd win a bottle of coke if its globally in the next 5 years.

 

[Shock-On-T]

As for the fog, this keeps getting brought up and I really don't understand it.

The radar can see into a fog bank and slow down gracefully for a hidden obstacle.
An example of a hidden obstacle would be a robotaxi without a radar whose cameras suddenly all went gray and had to slam on its brakes because it had no way of knowing if there was a hidden obstacle ahead in the fog.

 

[paulp]

The radar can see into a fog bank and slow down gracefully for a hidden obstacle.
An example of a hidden obstacle would be a robotaxi without a radar whose cameras suddenly all went gray and had to slam on its brakes because it had no way of knowing if there was a hidden obstacle ahead in the fog.

Isn’t that why aircraft use radar? although they typically steer around rather than brake gracefully

 

[GrimRe]

The radar can see into a fog bank and slow down gracefully for a hidden obstacle.
An example of a hidden obstacle would be a robotaxi without a radar whose cameras suddenly all went gray and had to slam on its brakes because it had no way of knowing if there was a hidden obstacle ahead in the fog.

I suppose you can close your eyes and reverse parallel park using only sonar beeps by your logic?

You can’t see through fog. That much is true. However *most* of the information the autonomous car needs to drive exists in around visible light wavelength spectrum. You can’t rely on human intuition to augment radar (like humans do). The car has to have the passive optical receptors in order to make inferences about the world around it.

once you have that constraint, the radar return is like being served the salt and pepper without a meal.

 

[paulp]

I suppose you can close your eyes and reverse parallel park using only sonar beeps by your logic?

You can’t see through fog. That much is true. However *most* of the information the autonomous car needs to drive exists in around visible light wavelength spectrum. You can’t rely on human intuition to augment radar (like humans do). The car has to have the passive optical receptors in order to make inferences about the world around it.

once you have that constraint, the radar return is like being served the salt and pepper without a meal.

ShockonT never suggested tht the car would work on radar alone, so your suggestion of reverse parking with sonar and eyes closed is pointless and irrelevant.
A fully autonomous car that is capable of driving normally around a city does not exist yet. That is an undisputable fact. Your claims of how (mass use) autonomous cars work are therefore hypothetical. ‘Has to have’ for example. Only in your hypothetical theory. You have zero evidence that the future will make this a ‘has to have’.

 

[GrimRe]

ShockonT never suggested tht the car would work on radar alone, so your suggestion of reverse parking with sonar and eyes closed is pointless and irrelevant.
A fully autonomous car that is capable of driving normally around a city does not exist yet. That is an undisputable fact. Your claims of how (mass use) autonomous cars work are therefore hypothetical. ‘Has to have’ for example. Only in your hypothetical theory. You have zero evidence that the future will make this a ‘has to have’.

He wrote “drive into a fog bank” to me that means zero visibility hence my analogy. So assuming you are relying on this radar ping through the fog to make a driving policy decision then yes, you are effectively driving on radar alone in that instance because you are trusting radar over vision . If you are not relying on it, then you have to be driving within the capabilities of vision anyway so radar is redundant.

By the way; this is assuming the radar even works through dense fog which is a HUGE assumption. Infrared sensitive cameras might work better but jury still out on that one

yes there a fully autonomous cars driving around on city streets right now. Waymo have a commercial level 5 system operating right now with no driver. Zoox, MobilEye, Cruze and others have autonomous vehicles driving in multiple cities right now. Tesla FSD beta also is capable of driving through a city environment autonomously with driver monitoring. So no that’s not an “undisputable fact”. You are just demonstrating your ignorance and lack of understanding.

In none of those autonomous car systems is radar or lidar ever used exclusively to drive the car. However there are already thousands of Tesla’s without radar driving autonomously on highways right now (I.e navigate on autopilot for June built model y and model 3). Soon (as soon as a few weeks) there will be FSD beta cars driving around in cities not using radar at all either.

On the Tesla front, there are millions of cars with HW3 that are able to evaluate the camera performance versus radar. I would therefore call the data on this “mass use”.

 

[paulp]

He wrote “drive into a fog bank” to me that means zero visibility hence my analogy. So assuming you are relying on this radar ping through the fog to make a driving policy decision then yes, you are effectively driving on radar alone in that instance because you are trusting radar over vision . If you are not relying on it, then you have to be driving within the capabilities of vision anyway so radar is redundant.

By the way; this is assuming the radar even works through dense fog which is a HUGE assumption. Infrared sensitive cameras might work better but jury still out on that one

yes there a fully autonomous cars driving around on city streets right now. Waymo have a commercial level 5 system operating right now with no driver. Zoox, MobilEye, Cruze and others have autonomous vehicles driving in multiple cities right now. Tesla FSD beta also is capable of driving through a city environment autonomously with driver monitoring. So no that’s not an “undisputable fact”. You are just demonstrating your ignorance and lack of understanding.

In none of those autonomous car systems is radar or lidar ever used exclusively to drive the car. However there are already thousands of Tesla’s without radar driving autonomously on highways right now (I.e navigate on autopilot for June built model y and model 3). Soon (as soon as a few weeks) there will be FSD beta cars driving around in cities not using radar at all either.

On the Tesla front, there are millions of cars with HW3 that are able to evaluate the camera performance versus radar. I would therefore call the data on this “mass use”.

There is no level 5 vision only autonomous car driving aound the streets. Apologies for not making it clear that my point was for vision only systems. You also may like to review the tesla response to its FSD beta to the USA DMV enquiry. Tesla described it as level 3 at best. In any case, I’m fairly sure you’ll find all the brands you mentioned are supervised under strict DMV licencing controls. They therefore are not level 5 autonomous, but rather systems in development and testing to be autonomous, hopefully.

For you to claim ”However there are already thousands of Tesla’s without radar driving autonomously on highways right now (I.e navigate on autopilot for June built model y and model 3).” is absurd. They are advanced lane keeping systems and are no-where near autonomous. Indeed its that notion that teslas are autonomous that have gotten a lot of people seriously injured or killed, but tesla cannot be blamed for stupidity.

Also, “You are just demonstrating your ignorance and lack of understanding.” does nothing other than convince me you are pretending to be knowledgable. Its a known response in such circumstances. You may also choose to note that no-one during this discussion other than you has made snide remarks to anyone else.

Do you own a tesla car?

 

[GrimRe]

There is no level 5 vision only autonomous car driving aound the streets. Apologies for not making it clear that my point was for vision only systems. You also may like to review the tesla response to its FSD beta to the DMV enquiry. Tesla described it as level 3 at best. In any case, I’m fairly sure you’ll find all the brand you mentioned are supervised. They therefore are not level 5 autonomous, but rather systems in development.

Also, “You are just demonstrating your ignorance and lack of understanding.” does nothing other than convince me you are pretending to be knowledgable. Its a known response in such circumstances. You may also choose to note that no-one during this discussion other than you has made snide remarks.

Do you own a tesla?

You did not write level 5 vision only autonomous car. You wrote "A fully autonomous car that is capable of driving normally around a city does not exist yet." As that is patently not true, it is reasonable to say that the person making such a claim is ignorant and lacks understanding in the area of autonomous vehicles.

So now you are just moving the goalposts by changing what you said to "There is no level 5 vision only autonomous car driving aound the streets."? Even on your do-over you are still wrong. Maybe try changing what you said a 3rd time and see if that sticks? Let me guess, next time it will be "bUt sAfEtY dRiVeRs!!!". Hate to burst your bubble but having a safety driver does not change the SAE level and on that point, the SAE levels are bullshit. Waymo, Cruze, MobilEye, Zoox, Tesla are not developing a level 1,2,3,4,5 system, they are all aspirationally trying to solve autonomous vehicles that are on a statistical basis 10x safer than a human. No human is even Level 5 and is by definition impossible to demonstrate, so using that as some sort of "gotcha" when arguing, again, shows ignorance in what the industry views are.

On the DMV emails, what Tesla have done, brilliantly mind you, is the create a fully autonomous system under the legal definition of a Level 2 ADAS. That way they can accelerate the development by getting into consumer's hands faster, collect more data and then iterate faster to improve. A virtuous cycle nobody else has. The only way you get from Level 2 to "Level 5" is by collecting enormous amounts of data and then demonstrating that under computer control, the KMs per accident are 10x that compared to under human control. Only then regulators might allow it to be driverless with certain constraints to increase safety even further. Note 10x safer than human doesn't mean perfect, far from it, but once proven it would be highly unethical to not increase autonomous robotaxi usage as much as possible in order to save as many lives as possible.

MobilEye Vision Only fully autonomous 20 minute drive in a busy street:

"12 camera sensor suit no LIDAR, no radar"

where is the LIDAR & radar only drive if radar and LIDAR as so foundational to solving fully autonomy?

Also Tesla have ADAS drivers in the U.S in soon the E.U testing "Vision Only" FSD (i.e FSD Beta v9) right now. There are general public drivers who are using NoAP which is arguably Level 4 on the highway, vision only, right now. In a matter of weeks (hopefully), you will see dozens of YouTube videos of V9 FSD Beta doing supervised autonomous drives through city streets-- again-- vision only FSD.

By the way, what specifically leads you to think I am pretending to be knowledgable?

yes I have a Tesla (Model 3 SR+) with FSD. I note you refuse to purchase FSD. Seems to me you are trying to justify your position in having not bought it?

 

[paulp]

You did not write level 5 vision only autonomous car. You wrote "A fully autonomous car that is capable of driving normally around a city does not exist yet." As that is patently not true, it is reasonable to say that the person making such a claim is ignorant and lacks understanding in the area of autonomous vehicles.

So now you are just moving the goalposts by changing what you said to "There is no level 5 vision only autonomous car driving aound the streets."? Even on your do-over you are still wrong. Maybe try changing what you said a 3rd time and see if that sticks? Let me guess, next time it will be "bUt sAfEtY dRiVeRs!!!". Hate to burst your bubble but having a safety driver does not change the SAE level and on that point, the SAE levels are bullshit. Waymo, Cruze, MobilEye, Zoox, Tesla are not developing a level 1,2,3,4,5 system, they are all aspirationally trying to solve autonomous vehicles that are on a statistical basis 10x safer than a human. No human is even Level 5 and is by definition impossible to demonstrate, so using that as some sort of "gotcha" when arguing, again, shows ignorance in what the industry views are.

On the DMV emails, what Tesla have done, brilliantly mind you, is the create a fully autonomous system under the legal definition of a Level 2 ADAS. That way they can accelerate the development by getting into consumer's hands faster, collect more data and then iterate faster to improve. A virtuous cycle nobody else has. The only way you get from Level 2 to "Level 5" is by collecting enormous amounts of data and then demonstrating that under computer control, the KMs per accident are 10x that compared to under human control. Only then regulators might allow it to be driverless with certain constraints to increase safety even further. Note 10x safer than human doesn't mean perfect, far from it, but once proven it would be highly unethical to not increase autonomous robotaxi usage as much as possible in order to save as many lives as possible.

MobilEye Vision Only fully autonomous 20 minute drive in a busy street:

"12 camera sensor suit no LIDAR, no radar"

where is the LIDAR & radar only drive if radar and LIDAR as so foundational to solving fully autonomy?

Also Tesla have ADAS drivers in the U.S in soon the E.U testing "Vision Only" FSD (i.e FSD Beta v9) right now. There are general public drivers who are using NoAP which is arguably Level 4 on the highway, vision only, right now. In a matter of weeks (hopefully), you will see dozens of YouTube videos of V9 FSD Beta doing supervised autonomous drives through city streets-- again-- vision only FSD.

By the way, what specifically leads you to think I am pretending to be knowledgable?

yes I have a Tesla (Model 3 SR+) with FSD. I note you refuse to purchase FSD. Seems to me you are trying to justify your position in having not bought it?

In relation to your 2nd last paragraph, have another read of my post. Its adequately explained.
I have not purchased FSD and am in no way having to justify that choice, and the reason for that choice to you or indeed anyone.
But given I own two debt free model S, multiple tesla batteries, and a significant solar array, it could be that I’m more interested in tesla’s pursuit of environmental outcomes than systems that supposedly make me, a driver that has never crashed, safer for the roads.
But beyond that I’m a city dweller living in a city where we dont really have freeways. So where exactly would I use noa again? I already have lane keeping and advanced summon.

 

[GrimRe]

But given I own two debt free model S, multiple tesla batteries, and a significant solar array, it could be that I’m more interested in tesla’s pursuit of environmental outcomes than systems that supposedly make me, a driver that has never crashed, safer for the roads.

Cool story bro.

Maybe if the mission of Tesla is accelerate the world’s transition to sustainable energy it would make sense that each EV is utilised for as many KMs as possible (I.e autonomous ride hailing) and not just appease you personally?

What are your multiple debt free model s’ doing to contribute to “environmental outcomes” if they spend 90% of their useful life parked in your garage? If somebody could increase the net environmental outcome of their Tesla 10 fold by allowing others who are less fortunate than you to travel emission free — shouldn’t you be cheering that outcome on over selfishly wanting your own Wishlist of free software enhancements to be prioritised ?

Perhaps the strategy to achieve this autonomous vision is to create a compelling ADAS to generate enough revenue to drive a flywheel of progress to an ultimate goal of full autonomy? There are, shocking I know, people out there who have paid the cost for FSD and are completely satisfied with what they have now and they haven’t even seen the final product yet?

If you placed Tesla’s ambition for environmental outcome above your own personal benefit, maybe you wouldn’t be so critical of FSD and it’s progress. This is something that nobody has ever done before, it’s incredibly difficult to estimate with any certainty the timelines of how things will progress. Is an iota of respect and credit deserved for the thousands of men and women making it happen? There are constant technical challenges and set backs yet they preserve despite the drone of clueless neo-luddities telling them how to do their jobs.

Btw paulyp before you have a giant sook, I am not calling you a neo-Luddite there are people much worse than you, however you really should try to do better if you really want to convince other people—and not just yourself— that your interest in the environmental outcome of Tesla is not contingent on it benefiting you personally.

 

[paulp]

Cool story bro.

Maybe if the mission of Tesla is accelerate the world’s transition to sustainable energy it would make sense that each EV is utilised for as many KMs as possible (I.e autonomous ride hailing) and not just appease you personally?

What are your multiple debt free model s’ doing to contribute to “environmental outcomes” if they spend 90% of their useful life parked in your garage? If somebody could increase the net environmental outcome of their Tesla 10 fold by allowing others who are less fortunate than you to travel emission free — shouldn’t you be cheering that outcome on over selfishly wanting your own Wishlist of free software enhancements to be prioritised ?

Perhaps the strategy to achieve this autonomous vision is to create a compelling ADAS to generate enough revenue to drive a flywheel of progress to an ultimate goal of full autonomy? There are, shocking I know, people out there who have paid the cost for FSD and are completely satisfied with what they have now and they haven’t even seen the final product yet?

If you placed Tesla’s ambition for environmental outcome above your own personal benefit, maybe you wouldn’t be so critical of FSD and it’s progress. This is something that nobody has ever done before, it’s incredibly difficult to estimate with any certainty the timelines of how things will progress. Is an iota of respect and credit deserved for the thousands of men and women making it happen? There are constant technical challenges and set backs yet they preserve despite the drone of clueless neo-luddities telling them how to do their jobs.

Btw paulyp before you have a giant sook, I am not calling you a neo-Luddite there are people much worse than you, however you really should try to do better if you really want to convince other people—and not just yourself— that your interest in the environmental outcome of Tesla is not contingent on it benefiting you personally.

Bit of a lesson in life for you. Before you write rubbish. Stop. Relax. Go for a walk. Gain some perspective. think about the fact that you dont know squat about the person you are about to write rubbish about. Then type furiously to get it out of your head. Then press delete.
The pause also gives you time to contemplate what happens to the renewable energy that my cars dont need from my solar when not in use, and the benefits that creates, and how a robotaxi will charge when used all day. Bit of nice dirty grid dc power presumbly, supplemented by the exported energy my car isnt using.

 

[meloccom]

Interesting discussion @GrimRe & @paulp.
You have both made your points.
Time to take a breath.

 

[GrimRe]

The pause also gives you time to contemplate what happens to the renewable energy that my cars dont need from my solar when not in use, and the benefits that creates, and how a robotaxi will charge when used all day. Bit of nice dirty grid dc power presumbly, supplemented by the exported energy my car isnt using.

It doesn't matter how an EV is charged, its per KM emissions will always be lower than a combustion engine. Therefore, in order for an EV to be most effective at reducing overall emissions it needs to increase its share of the global total of all KMs travelled.

In other words: A model S driven exclusively off solar energy but only driving 5,000 KMs per year is worse for the environment than a Model S driven exclusively off grid power and driving 100,000 KMs. Reason is simple: the model s driving 100,000 KMs is converting 20x more KMs to zero tail pipe emissions and is therefore responsible for a greater reduction in emission than one that is sitting in someone's garage all day.

While your car is at home being charged by solar panels, there are trips being taken in polluting cars. If you add up all the emissions from those trips and then deduct the grid emissions of the kWh's of energy you put into the car via solar that day, you still come out far worse for the environment compared to having made those trips via EV and then add in any electricity generation emissions. This is especially true as electricity generation emissions are trending towards zero but tailpipe emissions have largely plateaued.

No one is forcing you do make your Model S a robotaxi. Its a big 'if' this even happens and a personal preference that's your prerogative. There are also people in non-metro / regional areas where having their own EV makes the most sense even if they keep it in their garage most of them and that's fine-- However don't make it sound like a robo taxi service is counterproductive to the Tesla mission when the opposite is true.

 

[GrimRe]

BTW for anyone who's interested in more info straight from the horse's mouth, here is a talk from a few days ago from Tesla's AI Direction Andrej Karpathy specifically discussing why removal of radar leads to safer autonomous vehicles. The question about radar being able to see 2 cars ahead was asked as well. I don't want to spoil it, but suffice it to say, you are not smarter than Andrej when it comes to Autonomy or vision based ADAS.

 

[cafz]

BTW for anyone who's interested in more info straight from the horse's mouth, here is a talk from a few days ago from Tesla's AI Direction Andrej Karpathy

No matter what you think of FSD, that's a fascinating talk and well worth your time. Thanks for posting it!

 

[Wol747]

>>In other words: A model S driven exclusively off solar energy but only driving 5,000 KMs per year is worse for the environment than a Model S driven exclusively off grid power and driving 100,000 KMs. Reason is simple: the model s driving 100,000 KMs is converting 20x more KMs to zero tail pipe emissions and is therefore responsible for a greater reduction in emission than one that is sitting in someone's garage all day.<<

Can I buy some of that stuff, please?