What if "full self driving" is impossible?

[AquaMan]

Please forgive if this has already been discussed. I searched but didn't find it.

I love my Tesla, I love the company, and I'm long the stock. But I'm genuinely worried that "full self driving" is a promise Tesla can't deliver. Not just that it will take a long time, but that the hardware literally can't do it.

I'm worried that with "full self driving", paying customer expect their car to eventually FULLY. SELF. DRIVE. To me that means you can send it across town by itself in all traffic situations. But that's not just my interpretation, it's the interpretation that Tesla is encouraging. The name itself implies this. Elon once tweeted that the vision for summon is that the car will find you anywhere in the country. (Although obviously, it would need someone to charge it.) Elon has also talked about sending our cars to work as a ride-sharing service. Last I heard Elon is sticking by the idea that autonomy will be "feature complete" by the end of this year.

I'm sorry, but no way. No. Way.

Yes, I've seen videos of a Tesla driving itself across town. A trip without road construction, crossing animals, narrow roads, pedestrians jaywalking, shovels that fell out of the last pickup truck... With perfectly-painted lanes doesn't count from where I sit.

I've been writing software for over 35 years, and I'm currently a driving school owner. Autopilot is great. (Although its name is misleading.) But we humans deal with an infinite number of complex hazards in ways no computer could ever deal with UNTIL they become capable of fully recognizing the "real world context" of complex situations, which would require a "nearly self-aware" system.

Consider the hand gestures of a cop or road worker directing traffic. Consider the difference between a rag in the road versus a shard of metal. The presence of children throwing a ball on a front lawn. Recognition that when a dome light is on in a parked car, its door is likely to open. Consider that we humans recognize "the edge of the road" by very subtle visual differences between asphalt and gravel.

The way the car perceives its environment is displayed on the screen. Try covering your windshield and driving only from what you see there. (No, don't try that.) I've noticed that the cars lack proximity sensors on the side, which is probably why I've seen videos of summon hitting curbs near the rocker panels.

If you, dear reader, disagree with me and you feel the cars we own will eventually be able to drive themselves almost anywhere on public streets, that's okay. Either I've convinced you or I haven't... But my question is this: What happens if, say 2-5 years from now, those who paid $6,000 or $7,000, many of whom will have sold their car or totaled it, realize that what they got for that money wasn't what they were led to expect? What if they express their dissatisfaction en-masse, in court?

Has Tesla protected itself legally from this scenario? That is my question and the purpose of this thread.

P.S. Tesla cars are by far the best electric cars available, and they should sell them as just that. I just feel they should only take your money for what the car IS, not what they HOPE it will be. If I'm totally wrong, I'd gladly pay $7K to for a car which can reliably get me anywhere in town while I nap in the back seat. But not until such a feature exists.

 

[jerry33]

I guess the misconception here is that neural net is not like writing traditional software. The other issue is that AP is not the FSD software, it's similar to traditional software.

 

[AquaMan]

OKAY... Can we take a step back... My question is in red. Let's say I'm totally wrong, and that our cars really CAN be sent to work as ride-sharing services.

But if I WERE right... Is TSLA vulnerable? Here's a screenshot of my long position on calls by the way.

 

[Doggydogworld]

Please forgive if this has already been discussed. I searched but didn't find it.

I love my Tesla, I love the company, and I'm long the stock. But I'm genuinely worried that "full self driving" is a promise Tesla can't deliver. Not just that it will take a long time, but that the hardware literally can't do it.

I'm worried that with "full self driving", paying customer expect their car to eventually FULLY. SELF. DRIVE. To me that means you can send it across town by itself in all traffic situations. But that's not just my interpretation, it's the interpretation that Tesla is encouraging. The name itself implies this. Elon once tweeted that the vision for summon is that the car will find you anywhere in the country. (Although obviously, it would need someone to charge it.) Elon has also talked about sending our cars to work as a ride-sharing service. Last I heard Elon is sticking by the idea that autonomy will be "feature complete" by the end of this year.

I'm sorry, but no way. No. Way.

Yes, I've seen videos of a Tesla driving itself across town. A trip without road construction, crossing animals, narrow roads, pedestrians jaywalking, shovels that fell out of the last pickup truck... With perfectly-painted lanes doesn't count from where I sit.

I've been writing software for over 35 years, and I'm currently a driving school owner. Autopilot is great. (Although its name is misleading.) But we humans deal with an infinite number of complex hazards in ways no computer could ever deal with UNTIL they become capable of fully recognizing the "real world context" of complex situations, which would require a "nearly self-aware" system.

Consider the hand gestures of a cop or road worker directing traffic. Consider the difference between a rag in the road versus a shard of metal. The presence of children throwing a ball on a front lawn. Recognition that when a dome light is on in a parked car, its door is likely to open. Consider that we humans recognize "the edge of the road" by very subtle visual differences between asphalt and gravel.

The way the car perceives its environment is displayed on the screen. Try covering your windshield and driving only from what you see there. (No, don't try that.) I've noticed that the cars lack proximity sensors on the side, which is probably why I've seen videos of summon hitting curbs near the rocker panels.

If you, dear reader, disagree with me and you feel the cars we own will eventually be able to drive themselves almost anywhere on public streets, that's okay. Either I've convinced you or I haven't... But my question is this: What happens if, say 2-5 years from now, those who paid $6,000 or $7,000, many of whom will have sold their car or totaled it, realize that what they got for that money wasn't what they were led to expect? What if they express their dissatisfaction en-masse, in court?

Has Tesla protected itself legally from this scenario? That is my question and the purpose of this thread.

P.S. Tesla cars are by far the best electric cars available, and they should sell them as just that. I just feel they should only take your money for what the car IS, not what they HOPE it will be. If I'm totally wrong, I'd gladly pay $7K to for a car which can reliably get me anywhere in town while I nap in the back seat. But not until such a feature exists.

"Feature complete" simply means the features exist and sometimes work. It doesn't mean they work reliably. Tesla has vigorously walked back all their "summon from across the country" and similar claims from their website and other printed material. Musk still talks about appreciating car values and a million robotaxis, but everyone knows not to take him seriously.

Tesla may have legal liability to people who bought FSD back when it meant something else. But let's be serious, these customers are some of the boyest of fans. I mean, who pays thousands of bucks for something a company can't even demonstrate? Some of these customers have already totaled their cars or had their leases end. Are they marching on Fremont with pitchforks? There was an attempt to get a lawsuit going a while back, but as far as I know it went nowhere. If a FSD customer really wanted their money back, they could probably get it just by asking. Most are simply happy to contribute to the mission.

 

[AquaMan]

Musk still talks about appreciating car values and a million robotaxis, but everyone knows not to take him seriously.

Although I generally agree, the defense: "nobody takes the CEO seriously" seems weak.

I also think the "buy FSD now, because the price will go up later" is an upside-down strategy. It seems they did this at a time when they desperately needed cash coming in. (More so.) For me, autopilot provides about 85% of the convenience that I would want from a car's autonomy. Yes, I said I'd pay $7K if I could safely take a nap en route, which I don't think will exist for 20+ years. But I'd probably pay at most $3K to make my car capable of going through lights on its own while I hold the wheel. Because if I still have to supervise, it's basically a party trick that adds little convenience.

I think they could make a lot more money selling FSD at a more reasonable price to existing owners once it's functional, rather than selling for $7K+ to a smaller audience. After all, for the cars already on the streets, it costs them virtually nothing per-unit to enable it. If a $3K price triples the number of FSD sales (which I think it would) there's more profit to be had that way.

 

[bwong24]

Autopilot is great. (Although its name is misleading.)

Do you really think so? I have always understood AP to be as described here. https://www.faa.gov/regulations_pol...advanced_avionics_handbook/media/aah_ch04.pdf

 

[Puddles]

I consider the payment for FSD, which I did, an investment. I think Tesla was up front about this. Elon, who, like apple juice, is at his best unfiltered, is well known for overpromising. So all of us who invested should have done so a) with a warm fuzzy feeling about being part of an exciting future and b) with reasonable certainty that it's a crappy investment (financially speaking).

If it becomes apparent that FSD will never even be approached, I think FSD investors should be compensated in some way. Selling vaporware is not cool. OTOH, I feel like I'm getting a pretty good value for the investment even now, and it's only going to improve. I tell everyone my Tesla does 2 things very well: Stay in a lane, and not run into stuff. Offloading these tasks frees me up to be much more broadly attentive while driving and I feel like overall I'm a much safer and considerate part of the daily commute as a result.

I would be more satisfied if my status as exalted investor was recognized by, say, comping me my connectivity (I mean, come on, $10/mo?), and giving me the sense that I had a bit of priority for the hardware upgrade.

You're asking about a class-action kind of liability brought by disgruntled FSD investors, and I don't think that's going to materialize. We've swallowed the kool-ade and will be perpetually hopeful. I'm a little more worried about manufacturer's liability if it turns out that FSD cars get into expensive crashes. Hopefully that can be managed successfully by the owner's insurance carriers.

 

[defc0n]

Please forgive if this has already been discussed. I searched but didn't find it.

I love my Tesla, I love the company, and I'm long the stock. But I'm genuinely worried that "full self driving" is a promise Tesla can't deliver. Not just that it will take a long time, but that the hardware literally can't do it.

I'm worried that with "full self driving", paying customer expect their car to eventually FULLY. SELF. DRIVE. To me that means you can send it across town by itself in all traffic situations. But that's not just my interpretation, it's the interpretation that Tesla is encouraging. The name itself implies this. Elon once tweeted that the vision for summon is that the car will find you anywhere in the country. (Although obviously, it would need someone to charge it.) Elon has also talked about sending our cars to work as a ride-sharing service. Last I heard Elon is sticking by the idea that autonomy will be "feature complete" by the end of this year.

I'm sorry, but no way. No. Way.

Yes, I've seen videos of a Tesla driving itself across town. A trip without road construction, crossing animals, narrow roads, pedestrians jaywalking, shovels that fell out of the last pickup truck... With perfectly-painted lanes doesn't count from where I sit.

I've been writing software for over 35 years, and I'm currently a driving school owner. Autopilot is great. (Although its name is misleading.) But we humans deal with an infinite number of complex hazards in ways no computer could ever deal with UNTIL they become capable of fully recognizing the "real world context" of complex situations, which would require a "nearly self-aware" system.

Consider the hand gestures of a cop or road worker directing traffic. Consider the difference between a rag in the road versus a shard of metal. The presence of children throwing a ball on a front lawn. Recognition that when a dome light is on in a parked car, its door is likely to open. Consider that we humans recognize "the edge of the road" by very subtle visual differences between asphalt and gravel.

The way the car perceives its environment is displayed on the screen. Try covering your windshield and driving only from what you see there. (No, don't try that.) I've noticed that the cars lack proximity sensors on the side, which is probably why I've seen videos of summon hitting curbs near the rocker panels.

If you, dear reader, disagree with me and you feel the cars we own will eventually be able to drive themselves almost anywhere on public streets, that's okay. Either I've convinced you or I haven't... But my question is this: What happens if, say 2-5 years from now, those who paid $6,000 or $7,000, many of whom will have sold their car or totaled it, realize that what they got for that money wasn't what they were led to expect? What if they express their dissatisfaction en-masse, in court?

Has Tesla protected itself legally from this scenario? That is my question and the purpose of this thread.

P.S. Tesla cars are by far the best electric cars available, and they should sell them as just that. I just feel they should only take your money for what the car IS, not what they HOPE it will be. If I'm totally wrong, I'd gladly pay $7K to for a car which can reliably get me anywhere in town while I nap in the back seat. But not until such a feature exists.

coming from a fellow software developer, I’m going to take a guess that image recognition and machine learning isn’t one of your disciplines, and mine either. Much of the problem will be feeding the neural network more training data, and not typical software development.

that said, I do question the timeline, and can’t fathom a guess as to say whether the hardware is good enough or not. But I’d trust Karparthy and team over my own judgement.

 

[jeewee3000]

What happens if, say 2-5 years from now, those who paid $6,000 or $7,000, many of whom will have sold their car or totaled it, realize that what they got for that money wasn't what they were led to expect? What if they express their dissatisfaction en-masse, in court?

Has Tesla protected itself legally from this scenario? That is my question and the purpose of this thread.

1) In a nutshell: yes, Tesla is legally protected since the purchase agreement of the FSD clearly states "'The currently enabled features require active driver supervision and do not make the vehicle autonomous. The activation and use of these features are dependent on achieving reliability far in excess of human drivers as demonstrated by billions of miles of experience, as well as regulatory approval, which may take longer in some jurisdictions. As these self-driving features evolve, your car will be continuously upgraded through over-the-air software updates".

Therefore, they limit the delivery of a truly fully self-driving-feature in two ways: it must be scientifically proven to be far more reliable than human drivers AND your local jurisdiction must allow it.

Therefore the legal obligation of Tesla is limited by an external factor: the government. The responsability (for not providing FSD) can therefore always be blamed on legislators not co-operating with FSD rollout.

In Europe for example, the smart summon feature is illegal (as far as I know) and therefore not released by Tesla, but the European Tesla customers have paid the same amounts (actually more given the USD/EUR exchange rates). Will they sue Tesla? They might but they'd lose. Tesla can't help it that the EU countries do not currently allow it.


2) We have discussed this a lot. As pointed out, it is crucial to differentiate between "actual Full Self Driving (level 5)" and "feature-complete FSD".

Feature complete is currently promised on Tesla's website by year's end. Even IF they release this in the USA, I'm quite certain they won't release feature complete in EU this year. Could people sue? They might. But then again you have to read the fineprint and it will again limit Tesla's responsability with future promises.

I agree the name "FSD" is kinda misleading. But the fineprint limits its weight in the contract between Tesla and consumer. But hey, if people buy a Tesla online and click agree without reading the fineprint, it's their loss.

Going back to the difference between level 5 and feature complete, it is known that level 5 will take at least some years, maybe a lot more. We will only know after the fact. But Tesla's FSD feature does not promise level 5 at the moment, only feature complete.

 

[WarpedOne]

FSD is possible because you and I can drive a car.
FSD may end up looking much different from you or me driving the car though.

Neural nets are not like conventional programming in a way that they are able to 'cover' much more complexity than human brains are able to.
We are good at solving problems in 'divide et impera' way, we split one big hard problem into small less-hard isolated subproblems.

Driving in a real world is not really dividable into isolated less-hard subproblems. For it to work, everything has to be taken into consideration at every single point in time.
Not something our brains are big enough to solve in a IF THEN ELSE way.
NeuralNets do 'statistical' assessment... show them enough examples and they will generalize a rule the fits them.
The fit may not be optimal for some new example, therefore this process is never really complete, the rule just gets better and better with more examples it sees.

To simulate this aproach (evolution) by modifying IF THEN ELSE set of rules is just ... un-possible...

 

[Not Sure]

I'm still not sure.

 

[DurandalAI]

I would wager money that it will be Level 2 feature complete in 2020... In fact, I did, I bet $5,000 by buying FSD.
I think they will hit Level 3 in 2021, and level 4 in late 2021.

 

[Ludalicious]

OKAY... Can we take a step back... My question is in red. Let's say I'm totally wrong, and that our cars really CAN be sent to work as ride-sharing services.

But if I WERE right... Is TSLA vulnerable? Here's a screenshot of my long position on calls by the way.

I think they are vulnerable in the sense, they advertised FSD on city streets would be available by the end of this year. They charged large amounts of money with this claim. It wasn't a tweet but an official timeline on their site. That clearly hasn't or is going to happen within the next few weeks. They may be liable for a massive class action lawsuit IMO. With how vulnerable the market is on Tesla I havent bought in again for this very point. Holding out until the new year to see how this plays out.

 

[North75]

I think it was a good move on Tesla's part to reorgainze AP and FSD the way they currently have them.
FSD has features that work now, that you do not get with the included AP. This is what you are paying for with your $7000. There is a promise of additional features to come, but there are tangible features available now.

You can decide if the current features are worth $7000, or if you think the future features will be worth $7000 within the expected lifetime of you owning the car. If you don't think they are worth $7000 you can wait until they are (with the risk of Tesla raising the price before you buy in)

I think this setup puts Tesla in way less risk then when they sold it as EAP + FSD. (Not sure if they might have liability from people who bought both during that time)

 

[Matias]

If a FSD customer really wanted their money back, they could probably get it just by asking..

No. People have tried this without luck.

 

[thaian207]

Please forgive if this has already been discussed. I searched but didn't find it.

I love my Tesla, I love the company, and I'm long the stock. But I'm genuinely worried that "full self driving" is a promise Tesla can't deliver. Not just that it will take a long time, but that the hardware literally can't do it.

I'm worried that with "full self driving", paying customer expect their car to eventually FULLY. SELF. DRIVE. To me that means you can send it across town by itself in all traffic situations. But that's not just my interpretation, it's the interpretation that Tesla is encouraging. The name itself implies this. Elon once tweeted that the vision for summon is that the car will find you anywhere in the country. (Although obviously, it would need someone to charge it.) Elon has also talked about sending our cars to work as a ride-sharing service. Last I heard Elon is sticking by the idea that autonomy will be "feature complete" by the end of this year.

I'm sorry, but no way. No. Way.

Yes, I've seen videos of a Tesla driving itself across town. A trip without road construction, crossing animals, narrow roads, pedestrians jaywalking, shovels that fell out of the last pickup truck... With perfectly-painted lanes doesn't count from where I sit.

I've been writing software for over 35 years, and I'm currently a driving school owner. Autopilot is great. (Although its name is misleading.) But we humans deal with an infinite number of complex hazards in ways no computer could ever deal with UNTIL they become capable of fully recognizing the "real world context" of complex situations, which would require a "nearly self-aware" system.

Consider the hand gestures of a cop or road worker directing traffic. Consider the difference between a rag in the road versus a shard of metal. The presence of children throwing a ball on a front lawn. Recognition that when a dome light is on in a parked car, its door is likely to open. Consider that we humans recognize "the edge of the road" by very subtle visual differences between asphalt and gravel.

The way the car perceives its environment is displayed on the screen. Try covering your windshield and driving only from what you see there. (No, don't try that.) I've noticed that the cars lack proximity sensors on the side, which is probably why I've seen videos of summon hitting curbs near the rocker panels.

If you, dear reader, disagree with me and you feel the cars we own will eventually be able to drive themselves almost anywhere on public streets, that's okay. Either I've convinced you or I haven't... But my question is this: What happens if, say 2-5 years from now, those who paid $6,000 or $7,000, many of whom will have sold their car or totaled it, realize that what they got for that money wasn't what they were led to expect? What if they express their dissatisfaction en-masse, in court?

Has Tesla protected itself legally from this scenario? That is my question and the purpose of this thread.

P.S. Tesla cars are by far the best electric cars available, and they should sell them as just that. I just feel they should only take your money for what the car IS, not what they HOPE it will be. If I'm totally wrong, I'd gladly pay $7K to for a car which can reliably get me anywhere in town while I nap in the back seat. But not until such a feature exists.

This is one of my biggest concerns as well. I have been around computers for a very long time, machine learning is not yet proven at this level and probably not for a long time. I do understand the capabilities and limitations of these cameras we're using as well. Even high-end Canon L series lenses are not perfect in dark conditions and required a very low frame rate to capture an image clearly.

 

[RFernatt]

Personally, I think Tesla will get to FSD and may achieve a general purpose, scalable solution before others do. The timeframe is the issue. I think FSD, as most people would understand it, is 5-10 years out with regulatory approvals, etc. AP1-2.5 weren't adequate and I doubt AP3 will be adequate. So, I won't purchase FSD for my Model 3 as I don't see it arriving in the timeframe that I own the car. I hope I'm wrong, but I wouldn't put my own money toward FSD. The few incremental features included under FSD vs my EAP are minimal and not very valuable to me at this time. Certainly not worth thousands of dollars.

There are lots of reasons to be excited about the company and it's prospects, but having a million cars capable of being robotaxis around the country next year is not one of them. No one else will have that either. My 2 cents.

 

[aronth5]

First, to the OP this topic has been discussed with thousands of posts in a 100+ threads but that is beside the point. Regardless of your view point moving to FSD will be incremental.
So that begs the question of how your define FSD which is a controversy itself. So its a journey and some will declare victory and others will claim failure. Only time will tell.
One interesting data point for me is whether Tesla changes any of the camera/radar/ultrasonic hardware in the Model Y. If they do it will be a sign they are reassessing how they are going to achieve FSD.
If the hardware stays they same it increases the chances Tesla will get close to FSD however you define that. For me I'm enjoying the journey even though I don't know what the end game will be with the current technology.