Analysis of the price-hike for FSD, and the options it allows Tesla

[Skotty]

For used Tesla's sold by Tesla, they can increase the value and/or price of every car without FSD by adding FSD, since they can add it at no cost to themselves. This is probably why so many used ones have FSD. Likewise, for cars not sold by Tesla, it would not be out of character for Tesla to act like no other seller can be trusted in any way, and remove FSD from any cars they think they can without getting into a lawsuit.

 

[nvx1977]

For used Tesla's sold by Tesla, they can increase the value and/or price of every car without FSD by adding FSD, since they can add it at no cost to themselves. This is probably why so many used ones have FSD. Likewise, for cars not sold by Tesla, it would not be out of character for Tesla to act like no other seller can be trusted in any way, and remove FSD from any cars they think they can without getting into a lawsuit.

But they're likely not adding $15k to the selling price of their used cars with FSD, correct? If that's the case, you'd be getting a better "deal" buying used from Tesla if FSD is important to you. May just be a technique Tesla is using to try to prevent their backlogs from growing.

 

[stopcrazypp]

But they're likely not adding $15k to the selling price of their used cars with FSD, correct? If that's the case, you'd be getting a better "deal" buying used from Tesla if FSD is important to you. May just be a technique Tesla is using to try to prevent their backlogs from growing.

Tesla isn't really in the business to sell used cars, so I think they will do whatever necessary to get the car moving fast especially if the car is taking up storage space that can be used for new cars.

 

[Tam]

So your feeling is that the price of FSD is based on competition for Autopilot? Do you suggest that a large fraction of the FSD sales were to get the EAP features rather than the promise of future FSD? In that case, I could see the point. But since you get basic Autopilot with every car in China or the USA, why does Autopilot competition affect the price of FSD the future self-driving product. (Amusing note, Future Self Driving would be a better name for the product, and might get California off Tesla's back.)

The price competition will be: crashable FSD beta against uncrashable L2 ADAS. Luminar founder Austine Russell thinks fully autonomous vehicles will be decades away:

Elon Musk's rival Austin Russell said Teslas will not become self-driving anytime soon

While Elon Musk has been promising autonomous cars are on the horizon, Austin Russell has made billions betting against him.

www.businessinsider.com

Machines are very good at repetitions. We may think a machine is "smart" because it can beat a chess game but actually, it's just repetitions: There are just finite moves in the game, and eventually, it's just repeatability.

Waymo has already demonstrated it could avoid collision for years (Others hitting Waymo doesn't count for this purpose of "collisions").

With the correct hardware configuration, software, software engineers... an uncrashable L2 ADAS technology is available now.

Xpeng in China is releasing City NGP to its beta users. It is meant to be L4 someday, but in the meantime, we can use it as a consumer L2 but it crashes less than Tesla FSD beta does.

 

[Knightshade]

it crashes less than Tesla FSD beta does.

<citation required>

 

[Tam]

<citation required>

All we can see now is the Youtube in both Tesla FSD beta and Xpeng City NGP.

The data is limited and biased because City NGP videos are done not by owners but by the press or influencers.

However, my prediction is the price competition is not between FSD beta vs. L4 like Waymo but FSD beta vs. less crashing L2 and toward Austine Russell's "uncrashable" L2.

That's why Tesla is giving a 50% discount on FSD for existing Tesla owners in China because China is starting to release FSD-beta-equivalent not as advertised L5 but as an L2 with the price of L2.

Putting it in another way, will consumers settle for City NGP style of L2 ADAS or Tesla Full Self Diving?

If so, why would someone settle for a less functional L2 when Tesla Full Self Diving someday summons your car from the Fremont factory and greet you in Detroit?

 

[momo3605]

The price competition will be: crashable FSD beta against uncrashable L2 ADAS. Luminar founder Austine Russell thinks fully autonomous vehicles will be decades away:

Elon Musk's rival Austin Russell said Teslas will not become self-driving anytime soon

While Elon Musk has been promising autonomous cars are on the horizon, Austin Russell has made billions betting against him.

www.businessinsider.com

Machines are very good at repetitions. We may think a machine is "smart" because it can beat a chess game but actually, it's just repetitions: There are just finite moves in the game, and eventually, it's just repeatability.

Waymo has already demonstrated it could avoid collision for years (Others hitting Waymo doesn't count for this purpose of "collisions").

With the correct hardware configuration, software, software engineers... an uncrashable L2 ADAS technology is available now.

Xpeng in China is releasing City NGP to its beta users. It is meant to be L4 someday, but in the meantime, we can use it as a consumer L2 but it crashes less than Tesla FSD beta does.

I'm all for competition and seeing others release L2/L4 software, but this video doesn't instill a ton of confidence. I know it's using HD maps, but it looks very obviously like a pre-mapped route for the video.

A couple of really basic noticeable omissions:
- It couldn't handle a cone and stopped car in the street, this means in any type of downtown situation with a stopped car in a lane, this will likely fail
- The driver never actually takes the car down smaller residential or downtown streets, he just roams around on larger multi-lane expressways. Not a single pedestrian interaction
- Car freaks out and demands you take control when another car overtakes it.

This is simple stuff that I'd argue even FSDb has been shown to do a better job at.

 

[Knightshade]

All we can see now is the Youtube in both Tesla FSD beta and Xpeng City NGP.

So when you wrote "it crashes less than Tesla FSD beta does." that wasn't actually supported by any real facts or data.

Thanks for clarifying!

 

[Dewg]

The price competition will be: crashable FSD beta against uncrashable L2 ADAS. Luminar founder Austine Russell thinks fully autonomous vehicles will be decades away:

Elon Musk's rival Austin Russell said Teslas will not become self-driving anytime soon

While Elon Musk has been promising autonomous cars are on the horizon, Austin Russell has made billions betting against him.

www.businessinsider.com

Machines are very good at repetitions. We may think a machine is "smart" because it can beat a chess game but actually, it's just repetitions: There are just finite moves in the game, and eventually, it's just repeatability.

Waymo has already demonstrated it could avoid collision for years (Others hitting Waymo doesn't count for this purpose of "collisions").

With the correct hardware configuration, software, software engineers... an uncrashable L2 ADAS technology is available now.

Xpeng in China is releasing City NGP to its beta users. It is meant to be L4 someday, but in the meantime, we can use it as a consumer L2 but it crashes less than Tesla FSD beta does.

Interesting to see other ADAS systems in play. With all the advanced sensors, it still has phantom braking (4:20), steering wheel nag (3:54), visualization issues like Tesla's dancing semi trucks (6:19), and similar issues with construction / unexpected obstacles (9:16). The planner works very well - was impressed how it handled lane selection.

One thing I notice in other ADAS and AV systems is how smooth the visualizations are, particularly with the pathing projection line. I think Tesla's approach is great from a nerd perspective, but it's disconcerting for the average driver. We get to see the planner's projection line as it moves through the neural nets in almost real-time. It's why the grey and blue line jumps around as you're driving. If Tesla stopped showing us that near real-time line, and instead showed us an average over the last 10 frames, the line would appear much smoother and more "confident".

Overall, very impressive. I can't wait to see how Xpeng and NIO improve their systems. I also can't wait to see how Mercedes does with L3 in the US market.

 

[Tam]

So when you wrote "it crashes less than Tesla FSD beta does." that wasn't actually supported by any real facts or data.

Thanks for clarifying!

Correct! The better phrasing should be "it crashes less than Tesla FSD beta does on available youtube so far."

I've seen FSD beta collides with obstacles on youtube, but I've seen none on Xpeng City NGP.

I am not in China, I am not an influencer. I am not allowed to drive it. I only got available on youtube.

 

[momo3605]

I've seen FSD beta collides with obstacles on youtube, but I've seen none on Xpeng City NGP.

you're comparing tens of thousands of miles of video on FSD Beta vs <100 on City NGP on video. Also as I mentioned above, i'd HARDLY call that video city streets driving.

I'm not saying you're wrong, but it's kind of an apples to orange comparison. We don't have the full picture of Xpeng. It's still just pre-approved press videos

 

[Tam]

I'm all for competition and seeing others release L2/L4 software, but this video doesn't instill a ton of confidence. I know it's using HD maps, but it looks very obviously like a pre-mapped route for the video.

Other systems require more hardware and more preparation and HD map is a must. Even GM Super Cruise requires HD mapping in advance before it can work.

That's the difference between the pre-planning method and Tesla's generalized method.

Should your car be crash-free in one city only in the pre-planning method or should it be capable of crashing in all cities as in Tesla's generalized method?

A couple of really basic noticeable omissions:
- It couldn't handle a cone and stopped car in the street, this means in any type of downtown situation with a stopped car in a lane, this will likely fail

That's why it's an L2.

Even an L4 got stuck and required humans:

Watch A Waymo AV Get Freaked Out By Traffic Cones, Block Traffic, And Evade Support Vehicles

Making cars drive themselves is very, very hard. Sure, we’ve come an awful long way since that Stanford-modified Volkswagen Touareg won the first DARPA Grand Challenge back in 2005, but we’ve still got a long way to go before full Level 5 autonomy. No matter what anyone says. In case you don’t...

jalopnik.com

Thus, instead of expecting an L4 that doesn't need human interventions, I would expect to get an "uncrashable" L2.

That's because because the current L4 doesn't need a driver at all, but human interventions are still required to get it out of trouble/bugs.

Thus, why not change it into an "uncrashable" L2 that already has a human at the wheel when it got into troubles/bugs?

 

[Tam]

you're comparing tens of thousands of miles of video on FSD Beta vs <100 on City NGP on video. Also as I mentioned above, i'd HARDLY call that video city streets driving.

I'm not saying you're wrong, but it's kind of an apples to orange comparison. We don't have the full picture of Xpeng. It's still just pre-approved press videos

Agreed.

Both @Knightshade and you are correct in my biased assumption due to limited or no value of youtube.

The Tesla 50% discount FSD price might be its way of taking the China L2 competition seriously even though there's no data just yet.

 

[Dewg]

We have to be careful using terms like "uncrashable". Uncrashable is fairly easy to achieve with an L2 system. Telsa could make a few tweaks to their FSD Beta code and become uncrashable. The problem is that uncrashable has to be in conjunction with performance and driver expectations.

10.69, when it was first released, was incredibly timid around pedestrians. If a ped even thought about moving towards the street from a sidewalk, the Tesla would brake. People got annoyed with it, even though we were told it would be very cautious around peds. As the new code was validated, and updates came out, the car got more confident around peds, though it's still more cautious than it used to be.

If FSD Beta was timid around everything, and braked for a falling maple leaf, it could easily be "uncrashable". I'm sure others would crash into the Tesla, but as you pointed out, those don't count. However, the "uncrashable" FSD Beta would be useless to drivers.

Getting a car to drive confidently, smoothly, safely, and never crash into anything is a hard task. As an L2 system, it's up to us drivers to disengage and take over when things aren't working well. Just like in the XPeng video where he needed to take over a few times.

I'd even argue that L2, by its very nature, can't be uncrashable. L2 means that the system can handle driving in some conditions, and a driver has to be able to intervene for the system when it cannot handle it correctly. L3 is more appropriate for uncrashable, as the system can handle driving in some conditions, and the driver is still there to intervene, but the system can handle situations better, and when it detects that it cannot handle something it warns the driver in advance that they will need to take over. L2 can alarm and the driver has to engage immediately. L3 can alarm and the driver has several seconds to engage.

As for Russell's comments:

Unlike Musk, Russell has said he believes a future with fully autonomous vehicles is decades away, calling himself "the chief autonomous vehicle skeptic" and saying there's always been an "assumption" that the issue of self-driving would be solved within the next few years.

"Even today, it's still not within grasp," Russell told Time. "There's no commercially viable business that's centered around all that."

He's chosen his words VERY carefully there. Waymo is operating full L4 vehicles in a few cities and expanding, so we're already seeing full AVs right now. However, those businesses are not turning a profit today, so when Russell says "no commercially viable business", he's hoping the reader thinks he means that autonomous driving is decades away, which it isn't.

 

[Oil Freedom]

All we can see now is the Youtube in both Tesla FSD beta and Xpeng City NGP.

The data is limited and biased because City NGP videos are done not by owners but by the press or influencers.

However, my prediction is the price competition is not between FSD beta vs. L4 like Waymo but FSD beta vs. less crashing L2 and toward Austine Russell's "uncrashable" L2.

That's why Tesla is giving a 50% discount on FSD for existing Tesla owners in China because China is starting to release FSD-beta-equivalent not as advertised L5 but as an L2 with the price of L2.

Putting it in another way, will consumers settle for City NGP style of L2 ADAS or Tesla Full Self Diving?

If so, why would someone settle for a less functional L2 when Tesla Full Self Diving someday summons your car from the Fremont factory and greet you in Detroit?

Maybe then the $1,200 delivery can be eliminated?! Would also need the snake automatic Superchargers.

 

[Tam]

Maybe then the $1,200 delivery can be eliminated?! Would also need the snake automatic Superchargers.

Tesla wants to increase revenue so I doubt the destination charge will be waived.

On the hand, self-charging was announced in Tesla 2016 blog:

"During this Beta stage of Summon, we would like customers to become familiar with it on private property. Eventually, your Tesla will be able to drive anywhere across the country to meet you, charging itself along the way. It will sync with your calendar to know exactly when to arrive."

 

[Knightshade]

Destination charge can not legally be waived based on how/where it is delivered- thus why even folks who take delivery from the factory in CA pay the same fee as everyone else.

I suppose Tesla could set the fee to 0, if they believed every single car in the US could be delivered via FSD (if it's 0 for anyone it has to be 0 for everyone)--but I'm pretty dubious they're going to be doing that anytime soon (or maybe ever- given the need to charge for far-away destinations, and the potential damage it might arrive with, and the miles it'd arrive with, etc)