Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Karpathy leaves - what's next for AP/FSD?

This site may earn commission on affiliate links.
From the Q2 2022 Earnings Report, Tesla refers to FSD Beta as "City Streets supervised autonomy":

FYIsBlfWYAIlZOX


Also interesting that the table lists "robotaxi and others" as being "in development":

FYIuIMxXEAAQaL3
 
Last edited:
Can't wait for FSD to be complete, and to turn out to be pretty shatty just like the current AP is, and then they will announce FSD 2.0 or Next-Gen FSD, which will require $5k worth of upgrades. I mean for the love of god, they're charging $2k for an "acceleration package" to shave off 0.5 seconds off your 0-60 time.
 
From the Q2 2022 Earnings Report, Tesla refers to FSD Beta as "City Streets supervised autonomy":

FYIsBlfWYAIlZOX


Also interesting that the table lists "robotaxi and others" as being "in development":

FYIuIMxXEAAQaL3

Yea sounds like they just want people to forget that FSD was supposed to be robotaxi-lite and was just supervised Driver assistance the whole time! It kinda pisses me off, but whatever my fault for supporting Elon’s kickstarter
 
...

Personally, I think Tesla will deliver "autosteer on city streets" on HW3 and say that FSD has been achieved and is done. Then they will announce HW4 with new sensors (upgraded cameras + radar) and promise "Next level FSD". Customers will need to trade in their cars to get HW4. That way, Tesla can milk customers for more money.

My feeling is Tesla will try to run out the clock on this, and not release "Autosteer on City Streets" out of Beta. At some point Regulation is going to happen, either in response to NHTSA findings or because of other companies getting >=L2 systems on the road. At this point Tesla may not be able to pass the standards and they will have to announce that "Regulations have unfairly impacted FSD". That will put FSD on a much longer path to pass the standards - probably never, but at least it's "Regulations" fault. No money back for Tesla customers.

If they release "Autosteer on City Streets" at roughly the level it is now I think there will be more anger by customers than if they just continue to spin it out forever. Of course they could widely release it as-is, call it Beta, and say they will continue to work on it. It's still likely to fall afoul of "Regulations" at some point, or possibly IEEE P2846 assuming it gets released.
 
My feeling is Tesla will try to run out the clock on this, and not release "Autosteer on City Streets" out of Beta. At some point Regulation is going to happen, either in response to NHTSA findings or because of other companies getting >=L2 systems on the road. At this point Tesla will not be able to pass the standards and they will have to announce that "Regulations have unfairly impacted FSD". That will put FSD on a much longer path to pass the standards - probably never, but at least it's "Regulations" fault. No money back for Tesla customers.

If they release "Autosteer on City Streets" at roughly the level it is now I think there will be more anger by customers than if they just continue to spin it out forever. Of course they could widely release it as-is, call it Beta, and say they will continue to work on it. It's still likely to fall afoul of "Regulations" at some point, or possibly IEEE 2846 assuming it gets released.
It’ll forever be beta like the auto lights and auto wipers lol
 
  • Like
Reactions: Revolver3131
[Devil's Advocate/Counter point debate]That was said about Steve Jobs also and.........It could be good time for Tesla to move beyond Musk. Just like Jobs, Musk has laid the foundation and launched the compony on a near vertical axes. Also just like Jobs Musk is likely insufferable to work for. Would be great if Tesla had a "Tim Cook" to now professionally and fiducially lead Tesla into the stratosphere like Cook did with Apple. Drama and seat of your pants/pivot on a dime autocratic decisions may be great strategies for a startup/growth compony but it is starting to become an anchor holding Tesla back.

Bear in mind that the insufferable Steve Jobs was *before* he got booted out. That experience humbled him considerably, and the return of S.J. saw a very different personality. Of course, in the years between those two periods, the company nearly bit the dust, and survived only because of a giant bailout by Microsoft (largely to avoid the extra scrutiny that would have resulted if Windows had become a monopoly).

Would Tesla be better off if the board took Musk down a peg or two? Probably. Is kicking him out entirely a good idea? Only if you know somebody who can replace him who doesn't think that selling cars is just like selling soda, or that having as many slightly different configurations as possible will make it easier to get your cars into stores, or....
 
  • Like
Reactions: diplomat33
Sadly I'm not in the mood to spend $76 to view the IEEE P2846 Standard. At least there's a free white paper talking about it here.

Is the industry using this draft standard, or does it have to come out of "Draft" first before it's a standard? Is Tesla?

Tesla is not using it. Tesla was not even a member of the group that drafted the standard. I think other AV companies may be using it by choice. The standard is heavily based on Mobileye's RSS so it is likely Mobileye is using the standard. The other members of the group are probably using the standard too since they helped draft it. But AFAIK, the standard is not required since it has not been formally adopted by regulators yet.
 
  • Informative
Reactions: jebinc
I am not sure what you mean. In the future, the costs could come down enough that LIDAR is competitive with cameras. That’s just plainly obvious.

It's not obvious to me that LIDAR costs will ever come down enough, as it's a physics problem and not something that moore's law will help at some level.

The lidar problem requires counting and gating photons, literally. There's a large noise level and individual photon counting is necessary.

Inexpensive lidar, like what you can get on the top level iPhones works because it has short range, and because it rescans the area enough times to build up signal to noise ratio, i.e. slow and for static scenes.

What's necessary for driving is the opposite: very long range and rapidly updated. That isn't cheap, as it needs high power and quality physical optics which can be very accurately slewed, plus high speed photon detectors.

The lidar problem requires power proportional to R^4 for detection at range R and given reflectivity.

The future is AESA like imaging radar, not lidar. That is benefitted by Moore's laws and integrated semiconductor chips can get fabbed and packaged with conventional equipment, unlike fiddly very fragile lidar. Radars will slew with electronics changing phasing, not mirrors and lenses physically deforming or tilting. Conversion of electrical power to RF output power can now be quite efficient thanks to decades of mobile phone work. Conversion of electrical power input to laser optical output is far less efficient.
 
It's not obvious to me that LIDAR costs will ever come down enough, as it's a physics problem and not something that moore's law will help at some level.

The price has already come down enough IMO. Baidu just announced their next gen AV, the Apollo RT6, that has 8 lidar. And the unit production cost of the vehicle is only $37k. At that price, the vehicle would be affordable to sell to consumers. So clearly the cost of lidar has come down enough when the entire vehicle only costs $37k to produce and it has 8 lidar.


And companies have engineered new lidar that has no moving parts. They are much cheaper.


I do agree with you about imaging radar but it's not either/or. AV companies use both imaging radar AND lidar to compliment each other.
 
Last edited:
  • Informative
Reactions: _Redshift_
It's not obvious to me that LIDAR costs will ever come down enough, as it's a physics problem and not something that moore's law will help at some level.

The lidar problem requires counting and gating photons, literally. There's a large noise level and individual photon counting is necessary.

Inexpensive lidar, like what you can get on the top level iPhones works because it has short range, and because it rescans the area enough times to build up signal to noise ratio, i.e. slow and for static scenes.

What's necessary for driving is the opposite: very long range and rapidly updated. That isn't cheap, as it needs high power and quality physical optics which can be very accurately slewed, plus high speed photon detectors.

The lidar problem requires power proportional to R^4 for detection at range R and given reflectivity.

The future is AESA like imaging radar, not lidar. That is benefitted by Moore's laws and integrated semiconductor chips can get fabbed and packaged with conventional equipment, unlike fiddly very fragile lidar. Radars will slew with electronics changing phasing, not mirrors and lenses physically deforming or tilting. Conversion of electrical power to RF output power can now be quite efficient thanks to decades of mobile phone work. Conversion of electrical power input to laser optical output is far less efficient.

You might be right. I probably should not have said ‘plainly obvious’.

If the costs come down, though - they may not match that of a camera sensor, but the advantages might be enough to deploy them at that cost.

I worked as an advisor to a startup years ago which was doing something in this area, and know a little about it, but I got out after a short while from that role.