EAP and FSD pricing give hints to likely release readiness

[KyleDay]

Tin foil hat time-

After some thinking, and some late night crying because Elon said absolutely zip about EAP/FSD at the Model 3 reveal, I've come to accept that Tesla is very far away from an FSD release. Here's why:

Think for a moment about the features that EAP promises vs the features that FSD promises. Disregard what is "likely" or "over-hyped", but just read the feature descriptions of each package on Tesla's website. Ask yourself which set of features would bring more true value to the owner in terms of benefit for the cost.

Then, notice that EAP is $5k, and FSD is $3k. FSD promises significantly more benefits for the driver, yet it is 60% of the cost of EAP?

Granted, you have to buy EAP to get FSD, but the pricing doesn't seem to make sense when matched up against potential features.

I believe this is a clear signal that FSD is a pipe dream and that they're over charging for EAP. If both actually existed, they'd likely price EAP at $3k and FSD at $5k.

 

[JeffK]

Tin foil hat time-

After some thinking, and some late night crying because Elon said absolutely zip about EAP/FSD at the Model 3 reveal, I've come to accept that Tesla is very far away from an FSD release. Here's why:

Think for a moment about the features that EAP promises vs the features that FSD promises. Disregard what is "likely" or "over-hyped", but just read the feature descriptions of each package on Tesla's website. Ask yourself which set of features would bring more true value to the owner in terms of benefit for the cost.

Then, notice that EAP is $5k, and FSD is $3k. FSD promises significantly more benefits for the driver, yet it is 60% of the cost of EAP?

Granted, you have to buy EAP to get FSD, but the pricing doesn't seem to make sense when matched up against potential features.

I believe this is a clear signal that FSD is a pipe dream and that they're over charging for EAP. If both actually existed, they'd likely price EAP at $3k and FSD at $5k.

I'd say the majority of my driving personally is freeway/highway driving. That's worth more to me than FSD to a degree.

That said, there are rumors we might see the first FSD feature in the next week or two. Take it with a grain of salt.

 

[KyleDay]

It's just not priced right compared to the features it promises. To me that's a clear indicator that it will be limited if any actual features ever show up.

 

[JeffK]

It's just not priced right compared to the features it promises. To me that's a clear indicator that it will be limited if any actual features ever show up.

I don't think so... if that was the case there'd be an eventual lawsuit. I'm quite sure that Tesla is confident they can deliver even if they had to upgrade the computer hardware for free.

 

[calisnow]

Tin foil hat time-

After some thinking, and some late night crying because Elon said absolutely zip about EAP/FSD at the Model 3 reveal, I've come to accept that Tesla is very far away from an FSD release. Here's why:

Think for a moment about the features that EAP promises vs the features that FSD promises. Disregard what is "likely" or "over-hyped", but just read the feature descriptions of each package on Tesla's website. Ask yourself which set of features would bring more true value to the owner in terms of benefit for the cost.

Then, notice that EAP is $5k, and FSD is $3k. FSD promises significantly more benefits for the driver, yet it is 60% of the cost of EAP?

Granted, you have to buy EAP to get FSD, but the pricing doesn't seem to make sense when matched up against potential features.

I believe this is a clear signal that FSD is a pipe dream and that they're over charging for EAP. If both actually existed, they'd likely price EAP at $3k and FSD at $5k.

I'll disagree and take the optimistic case - the fact that 8 months later they are still taking $3K for FSD and Elon is still on stage telling the entire world that his entire fleet since October 2016 is going to go full autonomy means Elon still believes it will happen. This isn't a promise he made and then shuffled away hoping we'd forget about - he made it. He keeps making it. He says it loudly - he continues to take money for it. Also EAP has progressed very rapidly and continuously in both performance and features since its release in January. He has a very serious AI researcher heading up his vision team.

I think it will happen - also - I think EAP personally is the bigger utility factor for most. It's long highway drives that are fatiguing. Hopping around town is not a big deal.

 

[calisnow]

I'd say the majority of my driving personally is freeway/highway driving. That's worth more to me than FSD to a degree.

That said, there are rumors we might see the first FSD feature in the next week or two. Take it with a grain of salt.

That would dovetail with the continued release schedule of improvements/features since January. My spreadsheet shows 2.5 weeks remaining from yesterday until the next autopilot update/improvement - based on the average days between releases for the last 6 months.

 

[calisnow]

@kdday - furthermore in the "FSD will happen" camp is this. FSD is a two part problem as I understand it 1 - object recognition. 2 - path planning.

• Object recognition ie "stay in lane and don't hit that shopping cart" has been on a continued improvement path very quickly since January. Given its continual improvement let's assume they'll have that to an acceptable level within a year.
• Path planning / decision making - this is the logic/reasoning problem which they could have been working on for several years now in secret using software agents and/or cars on private courses. This is the FSD nut and probably has at least some coded logic along with some neural network based "learned" outcomes not specifically coded. - "You see two vehicles in opposing lanes - do you slow down, speed up, move to the shoulder etc."

It's possible that that the path planning coding is largely completed and what we are actually waiting on for some FSD features to get turned on is for "Tesla Vision" to reach a sufficiently accurate state of object/lane recognition that Musk decides to give the green light for some early/cautious/low-speed FSD feature releases.

 

[Bumper]

Keep in mind FSD is "dependent upon extensive software validation and regulatory approval, which may vary widely by jurisdiction". I'm no lawyer but if it is not legal for your customers to use FSD, it may not be the best idea to hand them a "Beta" of it.

We also know that Telsa is capturing tons of data from our cars and processing it in their data centers. I suspect they are able to gather data from all 8 cameras even though only a few are actually being used by AutoPilot currently.

If I were Tesla, I would have a firmware version that gathers everything like a flight data recorder. Every user input and every camera angle, plus overlay what choices and corrections the AI would have made. These cars would likely be driven by employees. Then crunch that data and look for places where the user made an input that the computer would have done a different one. Focus on those to improve the AI.

But even if FSD was 90% working, they have not cleared "Regulatory Approval" beyond a handful of test vehicles on a waiver situation and a limited geographic area.

 

[Bladerskb]

@kdday - furthermore in the "FSD will happen" camp is this. FSD is a two part problem as I understand it 1 - object recognition. 2 - path planning.

• Object recognition ie "stay in lane and don't hit that shopping cart" has been on a continued improvement path very quickly since January. Given its continual improvement let's assume they'll have that to an acceptable level within a year.
• Path planning / decision making - this is the logic/reasoning problem which they could have been working on for several years now in secret using software agents and/or cars on private courses. This is the FSD nut and probably has at least some coded logic along with some neural network based "learned" outcomes not specifically coded. - "You see two vehicles in opposing lanes - do you slow down, speed up, move to the shoulder etc."

It's possible that that the path planning coding is largely completed and what we are actually waiting on for some FSD features to get turned on is for "Tesla Vision" to reach a sufficiently accurate state of object/lane recognition that Musk decides to give the green light for some early/cautious/low-speed FSD feature releases.

You are really naive if you think all FSD is is basic object recognition and rudimental path planning.

 

[kbM3]

@kdday - furthermore in the "FSD will happen" camp is this. FSD is a two part problem as I understand it 1 - object recognition. 2 - path planning.

• Object recognition ie "stay in lane and don't hit that shopping cart" has been on a continued improvement path very quickly since January. Given its continual improvement let's assume they'll have that to an acceptable level within a year.
• Path planning / decision making - this is the logic/reasoning problem which they could have been working on for several years now in secret using software agents and/or cars on private courses. This is the FSD nut and probably has at least some coded logic along with some neural network based "learned" outcomes not specifically coded. - "You see two vehicles in opposing lanes - do you slow down, speed up, move to the shoulder etc."

It's possible that that the path planning coding is largely completed and what we are actually waiting on for some FSD features to get turned on is for "Tesla Vision" to reach a sufficiently accurate state of object/lane recognition that Musk decides to give the green light for some early/cautious/low-speed FSD feature releases.

It is not just object recognition. It is probable path planning of all of the non-stationary objects detected by the sensors. And determining how those paths may intersect your vehicle's path.

The whole thing is an unbelievably complex problem. I cannot imagine that they are close to solving it. But, I also have trouble imagining that Elon Musk is so off base. It seems like it nearly requires human level cognition to handle never before encountered circumstances.

Just one silly example. It has to know that it is safe to hit a plastic bag floating around in the air. You cannot just code into the software some kind of plastic bag exception. I do not know how they would handle this kind of case. Maybe by the way the object moves they can tell that it is light enough that it is safe to hit? Maybe they will use the radar when the objects are in front of the vehicle and thus detectable by the radar?

 

[Saghost]

Keep in mind FSD is "dependent upon extensive software validation and regulatory approval, which may vary widely by jurisdiction". I'm no lawyer but if it is not legal for your customers to use FSD, it may not be the best idea to hand them a "Beta" of it.

We also know that Telsa is capturing tons of data from our cars and processing it in their data centers. I suspect they are able to gather data from all 8 cameras even though only a few are actually being used by AutoPilot currently.

If I were Tesla, I would have a firmware version that gathers everything like a flight data recorder. Every user input and every camera angle, plus overlay what choices and corrections the AI would have made. These cars would likely be driven by employees. Then crunch that data and look for places where the user made an input that the computer would have done a different one. Focus on those to improve the AI.

But even if FSD was 90% working, they have not cleared "Regulatory Approval" beyond a handful of test vehicles on a waiver situation and a limited geographic area.

They definitely have screenshots from all eight cameras - in the AP2 capabilities thread folks have intercepted a few for analysis.

I'm pretty sure that Tesla is doing the data comparison on most or all of the customer fleet already to some extent - remember the discussion last year about AP2 being turned on in shadow mode by one firmware update? I think what Elon called shadow mode is what you're calling flight data recorder mode.

 

[calisnow]

It is not just object recognition. It is probable path planning of all of the non-stationary objects detected by the sensors. ...

I know - I said that.

 

[bob_p]

From a hardware perspective, EAP should (eventually) use most of the AP2 hardware - 4 of the 8 cameras, front looking radar, 12 proximity sensors and the NVidia Drive PX2 processor. FSD activates the remaining 4 cameras.

So pricing $5K for EAP - to use most of the hardware and then an additional $3K for the future FSD may make sense, from a hardware perspective.

Plus, as Tesla gets closer to FSD, it's possible EAP will shift from using the AP1 software (which is probably where we are today) to using the FSD software, with some limitations and adding the requirement to periodically touch the steering wheel.

 

[JeffK]

FSD may need to meet regulatory approval but the subset of features can be enabled ahead of time simply requiring drivers to be in the car and aware of their surroundings.

 

[TIppy]

Then, notice that EAP is $5k, and FSD is $3k. FSD promises significantly more benefits for the driver, yet it is 60% of the cost of EAP?

Could be $3500 for hardware, $1500 for EAP and $3000 for FSD.

 

[larmor]

Once EAP becomes active, the add on price will rise-- this is just a business decision and a treat for early adopters' data being mined. Why- many companies, tech and auto are working on this problem and there is a lot of money behind this (VC and otherwise to tackle this). The most immediate casualty (ironic choice of words) of EAP will be car insurance...

 

[clostridium]

Could be $3500 for hardware, $1500 for EAP and $3000 for FSD.

Except the hardware goes into all the cars doesn't it? You can OTA upgrade to EAP and FSD.

 

[TIppy]

Except the hardware goes into all the cars doesn't it? You can OTA upgrade to EAP and FSD.

Sure it does. The cost per unit is much cheaper that way. The have a pretty good idea what percentage of buyers will choose to buy autopilot, and price the hardware appropriately. Why should non-ap buyers subsidize ap buyers?

 

[JeffK]

Why should non-ap buyers subsidize ap buyers?

Because they get the safety features provided by the AP hardware. The real question is why should AP buyers subsidize non-AP buyers.

I honestly think the hardware is included in the base price of the car. The current pricing is by value and that's justifiable right now due to the lack of real competition.

 

[TIppy]

Because they get the safety features provided by the AP hardware. The real question is why should AP buyers subsidize non-AP buyers.

I honestly think the hardware is included in the base price of the car. The current pricing is by value and that's justifiable right now due to the lack of real competition.

The level of hardware isn't required for the saftey features. They could apportion the bill between base and autopilot buyers.