insaneoctane
Well-Known Member
Hell, the dealerships are collecting $10K markup on vehicles that don't have a new $7,500 rebate....
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable
- Thread starter AudubonB
- Start date
Gigapress
Trying to be less wrong
Chuck Cook's testing on FSD 10.69.1.1 had an uncomfortably close encounter with the red SUV. This particular screenshot is the exact moment Chuck did an emergency disengagement. If you look at the video clip, it's a scary close encounter.
However, I agree with James Douma's argument that FSD probably knew the exact measurements and predictions of ego and the red car and then moved in a spot right behind it. I wonder if FSD did this to maximize the distance between it and the next car coming after the red one.
The Tesla would have avoided hitting the red car as long as the rear end of the red car had cleared the path before the Tesla's right-front corner enters the lane. Chuck's car was about 1 car length from the red car's lane, and the red car would've needed to move about 2 to 2.5 car lengths to be out of the way, as indicated by the white lines. The red car might have suddenly swerved towards Chuck or slammed on the brakes, so let's say FSD needed to go at least 3x slower than the red car to be sure of not colliding.
Chuck's car was moving 7 mph at this moment while gradually accelerating, so without intervention it might've averaged up to ~9 mph over this distance. Instead, by pressing the brake, Chuck actually limited the average speed to 7 mph. In a later frame when the front bumper of Chuck's car begins to cross the line into the lane, the red car has advanced about 4.5 car lengths down the road from where it had been at the moment Chuck disengaged, so it must have been going about 7 * 4.5 = 32 mph.
If Chuck had let FSD do 10 mph average like it seemed to want, then the red car would've been about 32 / 9 = 3.6 car lengths down the road, so the Tesla would've had enough margin.
I wonder if this is what superhuman safety looks like, because I think this was the safety-maximizing move. If the vehicle has extremely high confidence it will not hit the red car, then the main other danger to avoid is the next car coming down the rightmost lane, and the best way to avoid that danger is to clear out of the lane as soon as possible.
The machine has more precision than we do for calculating motion trajectory and it also sees the world in slow motion, so following the movement of a 45 mph vehicle and making control decisions in less than 100 ms is no big deal. The 10.69 release came with the new occupancy network approach to mapping the drivable space, and it's supposedly got much better accuracy for inferring position and velocity of objects in the vector space. If so, this example may suggest that FSD 10.69.1.1 is exploiting this newfound precision to make bolder, more confident moves that occasionally exceed human confidence, although maybe if so they should dial back the confidence a bit to help the humans driving with FSD relax and to avoid scaring other drivers on the road.
However, I agree with James Douma's argument that FSD probably knew the exact measurements and predictions of ego and the red car and then moved in a spot right behind it. I wonder if FSD did this to maximize the distance between it and the next car coming after the red one.
The Tesla would have avoided hitting the red car as long as the rear end of the red car had cleared the path before the Tesla's right-front corner enters the lane. Chuck's car was about 1 car length from the red car's lane, and the red car would've needed to move about 2 to 2.5 car lengths to be out of the way, as indicated by the white lines. The red car might have suddenly swerved towards Chuck or slammed on the brakes, so let's say FSD needed to go at least 3x slower than the red car to be sure of not colliding.
Chuck's car was moving 7 mph at this moment while gradually accelerating, so without intervention it might've averaged up to ~9 mph over this distance. Instead, by pressing the brake, Chuck actually limited the average speed to 7 mph. In a later frame when the front bumper of Chuck's car begins to cross the line into the lane, the red car has advanced about 4.5 car lengths down the road from where it had been at the moment Chuck disengaged, so it must have been going about 7 * 4.5 = 32 mph.
If Chuck had let FSD do 10 mph average like it seemed to want, then the red car would've been about 32 / 9 = 3.6 car lengths down the road, so the Tesla would've had enough margin.
I wonder if this is what superhuman safety looks like, because I think this was the safety-maximizing move. If the vehicle has extremely high confidence it will not hit the red car, then the main other danger to avoid is the next car coming down the rightmost lane, and the best way to avoid that danger is to clear out of the lane as soon as possible.
The machine has more precision than we do for calculating motion trajectory and it also sees the world in slow motion, so following the movement of a 45 mph vehicle and making control decisions in less than 100 ms is no big deal. The 10.69 release came with the new occupancy network approach to mapping the drivable space, and it's supposedly got much better accuracy for inferring position and velocity of objects in the vector space. If so, this example may suggest that FSD 10.69.1.1 is exploiting this newfound precision to make bolder, more confident moves that occasionally exceed human confidence, although maybe if so they should dial back the confidence a bit to help the humans driving with FSD relax and to avoid scaring other drivers on the road.
Last edited:
This reminds me of the head start that Amazon had in cloud computing (AWS). Amazon prior to AWS was viewed as a ”retailer“, and they came up with cloud computing at least 5 years before the leaders in software like Microsof, Google, IBM etc…Giving Tesla a 15-year head start was a bad idea
Being the first mover in a market has advantages and disadvantages. Advantages include economies of scale, brand positioning, patents, and supply chain relationships. Another underappreciated advantage is gaining knowledge and information in the early stage before others have gotten started, which can be leveraged in later stages of the game to accumulate yet more knowledge and information in a virtuous cycle. Wright’s Law is not just an industry-wide model; it also applies to individual firms. The game is actually a race down the Wright’s law learning curve, as well as a competition over whose curve has the steepest slope—that is, who extracts the most innovation potential out of each doubling of cumulative production. This is why pace of innovation is all that matters in the long run for any market in which there is still a significant cost delta between the best producers and their less successful competitors.
I think for the car market, this first mover advantage is strong for engineering, including supply chain, design, manufacturing, and servicing. Tesla’s head start is a gigantic and probably insurmountable barrier to competition in many ways. Today I want to focus on how it’s given them data and experience.
Apple and Foxconn, for instance, are formidable companies but they haven’t even started an attempt at EV mass production. New companies like Rivian and Lucid have been working longer but are much less capitalized and less known than Apple & Foxconn, and they still haven’t actually been shipping cars to customers for years. Apple coming to the car market would be like when Michael Jordan left basketball to play baseball in 1993. Despite being a world-class elite athlete at the peak of his athletic prime, he did not even make it to the Major Leagues as a baseball player. The obvious reason was that he had not practiced baseball since high school.
When engineers design a machine like a car, they have simulation models and physical test data showing that stuff should work in theory, but there's still significant uncertainty. Engineers also have models and test data for the manufacturing system and associated uncertainty. Automotive engineering has even more uncertainty than most machine designs because the duty cycle is intense, customer expectations are high, and the vehicle spends most of its time outside with all the accompanying stress from vibration, temperature, salt, moisture, even UV radiation. Plus, the expected service lifetime is more than a decade. Accelerated life testing is crucial for planning this but you just never really know until actually putting the cars in service and waiting for them to get old.
Any new car company has to learn all this from scratch. Sure, they can hire people who have worked in the car industry and they can do their best to copy industry best practices, and they can even buy all the latest commercial off-the-shelf software tools, but there's still a limit. Companies have institutional knowledge, policies and procedures, relationships between people, and habits that are hard to transfer over bit by bit to a new company. Tribal knowledge tends to be indigenous to the environment of the tribe. Companies also have critical data that they’re generally unwilling to share.
With greater design uncertainty, engineers need to apply bigger safety margins and sometimes need to add extra layers of redundancy in case of failure. All of this comes at a price: reduced vehicle performance on key design criteria like cost, range, acceleration, handling, safety, etc. Uncertainty also brings the risk of setting margins too thin and having a higher-than-anticipated failure rate in service, like Nissan's battery degradation in the first-generation Leaf, GM's spectacular f-up with the LG pouch cell partnership and Ford's melting high-current electrical contacts. In the fog of misunderstanding, mistakes happen, especially in organizations where decisions are made based on politics, deceit and confrontation instead of logic, honesty and cooperation.
Drew Baglino discussed this in his Stanford interview earlier this year, saying that a decade ago Tesla had been too pessimistic about Model S battery cell electrochemical degradation, but too optimistic about the other stuff like pack moisture sealing, battery management electronics, mechanical shock and vibration, and thermal cycling. Notably, Drew said that these things "don't show up until you've been in the field for ten years". Yikes. So even the big brains at Tesla were too conservative in some areas and too aggressive in others. It was only after years of vehicles being in the fleet and millions of cars produced that they’ve advanced this far in fixing these problems, making the cars with more reliability, more quality, and less design fat.
Tesla also gets the most data per car per unit time. because they actually had the foresight to design the car for remote data collection and cloud computing. Tesla has been putting electronic sensors on their BEVs since *2003*. One of the very first things they did as a startup was setting up vehicle data collection for trying to reverse engineer the AC Propulsion t-zero prototype. I heard Elon and a few other early Tesla employees talking about this in a panel interview from around the early Model S years (I can’t find it anymore, so no link). I think I recall Elon referring to it as trying to tease out “the ghost in the machine”, because the t-zero used custom analog power electronics and nobody really knew how the hand-crafted mule actually functioned.
All of this means Tesla alone has the luxury of running the tightest tolerances in the industry for their BEV designs, because no one has has produced 3 million BEVs over the last decade. This is like going camping in the wilderness. A novice might be a smart and conscientious planner, but their unawareness of the actual needs of the trip will inevitably result in worse selection of supplies to bring compared to a person going on the same trip who’s done it many times. The novice will bring along some stuff that’s unnecessary and not bring (or not bring enough of) other stuff that they actually do need. The expert also will have a better understanding of which equipment suppliers have the best options. The expert knows what to spend money on and what to go cheap on. The novice needs to spend more time researching and shopping and even then they will probably end up wasting money in some areas and get junky equipment for other items. The expert’s advantage is information and experience.
Novices can surpass experts in the long run. Tesla sucked at making cars 10 years ago, but that was with some prior learning on the original roadster and Tesla-level pace of innovation. This is not normal progress over the first decade of attempting to grow to being a mass manufacturer of cars. I don’t anticipate an iCar or any other competition having a meaningful negative impact on Tesla’s business for at least ten years.
Tesla’s data and expertise lets them get by with less stuff such as:
Tesla also gets performance gains, such as:
- Structural material
- Welds
- Fasteners
- Battery cell depth-of-discharge reserve
- Warranty reserve
- Range per kWh
- Charging speed
- Weight
- More storage space and cabin interior space
- Handling
- More repeat sprints before power needs to be throttled
- NVH (Noise, vibration & harshness)
Tesla's inventions compound each other's gains due to these feedback loops, augmenting Tesla's resultant lead.
Weight reduction and chassis stiffening, for example, reduces the power required to move the vehicle around, reduces NVH, and improves handling which makes the vehicle more efficient, which then enables reduction of the battery size needed for a given set of requirements for range and performance. Weight reduction also in many cases increases cabin storage space by opening up more room, as Tesla has shown with their masterful gigacasting design making for more spacious trunks and frunks. Better understanding of battery degradation and better thermal control means that a more aggressive charging curve can be allowed. And so on.
Example of tech with compound benefit:
I don't think it's physically possible for a competitor to try all of this stuff in their first BEV. They have a long road ahead of them to implement these technologies that are necessary to have a product that can compete with Tesla vehicles on specs, features and cost.
- Octovalve and integrated thermal management across all vehicle subsystems
- Gigacastings with optimized new alloy
- Structural battery with seats directly mounted on top
- Cell-to-pack architecture
- Motors best in the game according to Munro testing and cost accounting (kW/$, kW/kg, kW/cm^3)
- 4680 batteries
- Aerodynamics
- Cybertruck folded stainless steel stressed skin structure
Even Tesla is still learning how to optimize their own inventions. Listen to remarks from the Q2 call:
The Rich Get Richer
All signs point towards acceleration of Tesla's pace of technological innovation. I think Tesla is in a runaway snowball effect situation now.
The EV market has an accumulative advantage dynamic with strong preferential attachment effects. Preferential attachment means a tendency within a competitive system for resources to be biased towards flowing to entities that already have more resources than other entities (i.e. "the rich get richer" / "success breeds success"). Preferential attachment was observed by Italian economist/engineer/sociologist Vilfred Pareto in his famous observation that 80% of the peas in his garden came from 20% of the plants and 80% of the wealth and land in Italy was owned by 20% of the families. The early advantage gained by some pea plants due to genetics or lucky position in the environment made them grow bigger more quickly as sprouts, and they leveraged this small advantage to consume more of the local sunshine, water and root space to grow even bigger, until a minority of plants dominated the garden. This relationship shows up in all kinds of phenomena like formation of stars and planets from dust after a supernova, crater size on the moon, frequency of words used in any language, and much more.
Preferential attachment usually results in a power law distribution, also known as a Pareto distribution. There are theoretical justifications for this and if you want to see the math I recommend reading the link. The stronger the preferential attachment effect, the steeper the Pareto curve is. Whenever there is a Pareto distribution in results of a competition, we can be pretty confident that some kind of preferential attachment effect exists.
View attachment 850038
In some cases, we observe power law rank relationships in which one or two outliers exist at the top, way off the trend line. This is called the king effect. Kings don’t conform to the statistical distribution of the rest, like how China and India have exceptionally large populations while all other nations fit neatly into a Pareto curve.
The EV industry in the US, Tesla's home turf, shows a typical Pareto distribution with one king, Tesla, which alone still holds most of the US BEV market share, and holds all of the profit. Soon enough they'll have more profit than all of the rest of the auto industry combined, including all cars, not just BEVs.
On a linear scale we can see just how far ahead Tesla is. Note that the pink colulmn is the grand total, the red column is Tesla, and the blue columns are the rest. On a logarithmic scale we can see that the power law model is a good fit, because all the data points fall appromixately in a line. All of them except Tesla, whose sales number comes in an order of magnitude higher than the power law rank relationship would predict.
View attachment 850045
View attachment 850044
Source: Inside EVs (link)
Total 193481 Tesla 139338 Ford 11751 Kia 11483 Hyundai 9675 Nissan 5980 Audi 5100 Volkswagen 3527 Mercedes-Benz 2641 General Motors 1648 Rivian 1145 BMW 611 Lucid 582
The dynamics that caused this result are not likely to change any time soon. The rank relationship for 2018 looks almost identical, again with Tesla an order of magnitude ahead of where the Pareto distribution of the rest of the market participants would predict Tesla to be. The numbers have gotten bigger and and the also-rans have shuffled around in the rankings, but in four years nobody has gotten any closer. In fact, if you look closely at the trend lines, Tesla’s deviation from the distribution has almost doubled since 2018, suggesting that indeed they are accumulating relative advantage over time.
View attachment 850054
Source: Inside EVs (link)
Total 239003 Tesla 191627 General Motors 18019 Nissan 14715 BMW 6889 Fiat 2250 Volkswagen 1354 Smart 1219 Kia 1134 Honda 948 Jaguar 393 Hyundai 345 Ford 70 Mercedes 40
Tesla has the lead in data and experience giving better products that cost less
--> Attract customers, investors and employees--> More scale, more capital--> Faster iteration cycles, more fun at work--> More data and experience--> Better products that cost less
Software is relatively easier to catch-up with compared to manufacturing, but even now some 10 years later AWS is still bigger than the combined top companies.
insaneoctane
Well-Known Member
Real question for all owners or those considering powerwalls (or powerpacks, or...), do you consider life expectancy and replacement costs when you buy these? What is the expected replacement schedule of a powerwall these days?
Oops, one more for the Tesla-owning King:
His coronation plans have been set a long, long time. And now it’s time to pull them out.
What’s this? A floppy disk? And…..punch cards?
I put in order for two power walls right before they started bundling with panels (we already have solar panels).
At this point, I am happy to wait until the next gen of the PW is announced/released. I expect it to be made up of LFP batteries, which are of course a better solution for home storage than Lithium. I don't know what the life expectancy is expected to be, but they should be greatly improved over the current PW2+.
insaneoctane
Well-Known Member
Not sure how we'll ever know. As long as FSD is L2, it's the driver's responsibility to ensure it's safe. Disengaging IS the right thing to do here with FSD Beta. If FSD has super human skills, I'm not sure a responsible driver will ever let the mothership know these skills are working.
Just saw a tweet of a dealer selling a bunch of Model Ys...around 100k sticker price on them.
Thorough analysis (as always); thanks.
If I were the driver in the red SUV, that would have been very frightening. I imagine Tesla will have to program in some amount of a safety buffer for moves like this.
You may be completely right. But after watching all of Chuck's 10.69.* videos, this was the least successful. So it could also be that there was a bit of regression in Chuck's famous left turn for 10.69.1.1. But all of it could also be explained by luck of the traffic.
That's OK. It's definitely, positively far better than 10.12. And progress is the name of the game.
That post was referring to the "$30,000" 2024 Chevy Equinox EV that was revealed today, not a Tesla.
I am sure @Thekiwi is referring to when Tesla was going through production hell with the Model 3 and about to go bankrupt. Elon reached out to Tim Cook to discuss the possibility of Apple buying out Tesla. But Tim Cook refused the call. The scuttlebutt is there has been some friction between the two ever since.
Words of HABIT
Active Member
Did China's BYD Sell More BEVs Than Tesla In H1 2022? Not Really
BYD sold 641,350 electrified vehicles in H1 2022, split evenly into plug-in hybrid vehicles and battery electric vehicles; Tesla sold 564,743 BEVs.
insideevs.com
Kudos to BYD Auto for manufacturing and selling electric vehicles. However take note:
InsideEVs:
"Many media outlets from around the world are reporting that Chinese carmaker BYD Auto has overtaken Tesla as the world's biggest electric vehicle maker by sales in the first half of this year.
Is that accurate? Well, it depends on what the definition of an electric vehicle is. BYD, which is part-owned by Warren Buffett's Berkshire Hathaway, has sold 641,350 new energy vehicles in the first six months of 2022, a massive 314 percent jump from the same period the year before.
There's a (big) catch, though: not all these vehicles were fully electric. Of the total, 323,519 units or just over 50 percent were battery electric vehicles; the rest were plug-in hybrids. Under China's sales rules, PHEVs and BEVs are counted together as new energy vehicles (NEVs).
During the same period, Tesla has sold 564,743 vehicles, all of which were obviously fully electric. This means that the US electric vehicle maker is still the world's top battery electric vehicle manufacturer by sales..."
Also, according to South China Morning Post: "Most of the BYD’s models are priced between 100,000 yuan (US$14,371) and 200,000 yuan, about half the price of Tesla’s Model 3s and Model Ys."
Last edited:
ifish
Member
Thank you for that reply, now I know I have company
The Australasian New Car Assessment Program (ANCAP) youtube channel published clips of their test on the Model Y, including crash tests and collision avoidance from other cars and pedestrians.
Unfortunately it has zero video commentary and no description, and no comments allowed.
[Edit: Euro NCAP youtube channel has the exact same video clips on their channel]
Unfortunately it has zero video commentary and no description, and no comments allowed.
[Edit: Euro NCAP youtube channel has the exact same video clips on their channel]
Last edited:
Similar threads
- Locked
- Locked
- Locked
