What If The Cost Of Training Tesla Full Self Driving (FSD) Keeps Expanding, But Progress Stalls Again?

Join daily news updates from CleanTechnica on e mail. Or follow us on Google News!

On this planet of Tesla Full Self Driving (FSD), I hear two essential views. The Tesla bulls (of which I’m a member) appear to fake that it’s a foregone conclusion that Tesla will clear up FSD quickly and the one factor left is particulars like the app to call a robotaxi or designing a car that is self-cleaningleaving the elephant within the room (V12 was an enormous advance, however little progress has been noticed within the final 6 months).

Simply completed a one-week journey to China. I’ve now “survived” all the foremost (~20) L2 self-driving and robotaxi automobiles in each the US and China. Some ideas & observations:

▶️L2 self-driving

I examined main manufacturers like $Huawei, $Li, $NIO, $Xpengand $Xiaomi. Total, they… pic.twitter.com/2iTluT24SU

— Freda Duan (@FredaDuan) July 2, 2024

This latest submit on X by Freda Duan of Altimeter Capital exhibits that Tesla has a number of competitors in China that’s already deploying driverless automobiles. She sees Tesla’s “imitation learning/end-to-end as the only effective approach to self driving,” however I feel that conclusion is untimely, particularly contemplating Waymo and Pony.ai have already solved self driving. Tesla bulls will regularly dismiss those who use LiDAR and mapping as not scalable for 2 essential causes:

1. They state that LiDAR is pricey. That’s true, however electrical automobiles had been costly and the prices got here down significantly. Many firms are engaged on inexpensive sensors they usually would possibly decrease the prices dramatically, particularly as they scale up. As an alternative of researching the fee curves, they simply wave their palms and dismiss it.
2. In addition they state that solely Tesla has the info to coach FSDso nobody else can do it or do it earlier than Tesla. The world of computing is stuffed with firms that had an insurmountable lead and blew it. I prefer to carry up the instance of my youngsters once they had been 15 years previous — they discovered to drive with out absorbing thousands and thousands of miles a day of video from 8 cameras. They appeared to determine it out in a number of weeks of coaching. After all, I understand they’d 15 years of coaching on “the real world” first, which gave them an important foundation to study to drive. I carry that up as a reminder that it is extremely onerous to foretell what the proper algorithm to unravel any drawback is till you succeed.

An financial level they’re lacking is that if 90% of the worth of FSD is within the cities and in 1% of the land mass (cities plus interstate highways), you could possibly map that 1% of the land mass and seize many of the cash. If I can take a robotaxi to all of the locations I am going often and it additionally takes me on my holidays, we shouldn’t dismiss it simply because it will probably’t go the locations I don’t need to go.

Bulls on FSD will say that the take price on FSD will rise because the know-how will get higher, however I discover that FSD has little worth within the metropolis so long as it have to be supervised. I take advantage of FSD on a regular basis and revel in monitoring its progress and it’s a pleasure to make use of on the highways and in stop-and-go visitors. However the place it’s making a number of turns within the metropolis, it’s extra work to fastidiously watch FSD to make sure it doesn’t make a mistake than to only drive myself. I proceed to make use of FSD within the metropolis simply because I’m into know-how, however different members of my household suppose it’s extra problem than its value.

The FSD take price has been horrible during the last 8 years (besides when a brilliant sale is obtainable), exhibiting the shopping for public agrees with me. Tesla simply did a free trial of FSD and you may guess your backside greenback that had it considerably elevated the FSD take price, Elon would have posted about that. The actual fact is that Autopilot is excellent and free with each Tesla, so the one cause to purchase FSD is when you prefer to play with the most recent tech as toy (like I do). After all, there may be Elon’s promise that the automobile might be value a fortune as soon as FSD is solved, however that’s depending on a number of assumptions. It assumes FSD might be solved earlier than I promote the automobile. It assumes will probably be solved for the FSD laptop in my automobile or Tesla will present an improve. It assumes the income cut up between Tesla and the proprietor might be first rate. It assumes will probably be definitely worth the time and problem to cost and clear my automobile between robotaxi rides.

FSD Coaching Prices Explode

My latest fear is that the prices of coaching FSD are quickly increasing whereas we nonetheless aren’t even near the extent of security essential to take away the driving force from the automobile. Some individuals will say that Tesla’s FSD is already safer than a human driver based mostly on the figures Tesla releases. That’s false. The figures that Tesla releases do show that FSD being supervised by a Tesla proprietor is VERY secure and much safer than a median human driver, however utilizing FSD day by day exhibits me that I might have an accident each few days if I didn’t intervene for FSD when it wants some assist.

This study by the American Automotive Affiliation (AAA) agrees with the Tesla data that accidents happen solely each 500,000 miles are so. This doesn’t agree in any respect with my expertise the place I’ve pushed about 500,000 miles over 40 years and been in about 10 accidents (just one my fault). However I’ve a sense that fender benders at 2 mph will not be included in these figures as a result of no police report is filed and most of my accidents had been collisions at 2 or 3 mph. Having stated that, if FSD can go about 10 miles in Florida with out an intervention that may trigger an accident, we have now to enhance the standard of the driving about 50,000 instances earlier than it’s safer than a human driving. Even when every launch improves security 5 instances, that might be 7 releases to get that stage of security. If every launch takes 3 months, that’s virtually 2 years, which may be very cheap, but when there may be something I’ve discovered about FSD during the last 8 years, it’s that progress might be going nice and it will probably stall for a 12 months at any time. Even when Elon posts that he’s VERY assured that FSD might be solved this 12 months, as a result of he has pushed the most recent launch and it’s wonderful, progress might be stalled and nothing might progress till an entire new design is developed by a big group of extraordinarily costly engineers utilizing extraordinarily costly laptop sources.

Conclusion

Of the roughly $10B in AI-related expenditures I stated Tesla would make this 12 months, about half is inside, primarily the Tesla-designed AI inference laptop and sensors current in all of our vehicles, plus Dojo.

For constructing the AI coaching superclusters, NVidia {hardware} is about…

— Elon Musk (@elonmusk) June 4, 2024

The underside line is that with about $30 billion in money and vital optimistic money move including to that stash, Tesla can afford a number of years of spending $10 billion a 12 months on AI-related expenditures, however with expected earnings of about $8 billion a year in 2024 (none of it from AI), how lengthy will the shareholders be keen to take a position greater than their annual earnings on software program that has prices that preserve rising however that doesn’t present any monetary return? The reply is so long as progress is spectacular or till one other firm exhibits FSD has been solved and left Tesla within the mud.

Will Tesla clear up FSD earlier than the competitors? The reply is not any — they’ve already misplaced that race to Waymo and Pony.ai — however will they clear up FSD earlier than others discover a solution to scale stage 4 self driving? I nonetheless suppose they are going to, however it’s actually a race and Tesla engineers shouldn’t be overconfident that they can’t be beat simply because they’ve some distinctive benefits.

Disclosure: I’m a shareholder in Tesla (TSLA), BYD (BYDDY), Nio (NIO), XPeng (XPEV), NextEra Vitality (NEP), and several other ARK ETFs. However I provide no funding recommendation of any type right here.