Big brother Tesla

[icowden]

Any Robotaxi will need Lidar thus raising more questions about fatalistic FSD marketing.

Why?

 

[albion]

It is interesting that it is being reported Musk is struggling to win the $$$$$$ billion vote.
However it is also reported that Scottish Mortgage Investment Trust will be voting in his favour.

They are a high growth stock investor meaning they doing a high roller gambit that Eon Musk cones up with the goods.
Gets them publicity I guess. The problem for them is that risk is their nature, making it hard to vote otherwise.

 

[the snail]

A low cost EV (around $25k - Tesla 2) has been discussed for a long time by Tesla. At the beginning of April, there were reports it had been scrapped but Musk said that Tesla were planning launch of a low cost vehicle by beginning of 2025 during the shareholder call. The plan was to move down the car cost families with the previous (more expensive) family funding the next development. A good plan as more expensive cars in the luxury market tend to give more profit per unit than the cheaper families which rely upon volume and production efficiency to be profitable.

The Cybertruck seems to be a diversion from this solid plan, it adds nothing to the EV market aside from a flawed, huge pickup which won't be able to be sold in key markets.



The new Polestar 2 LR has a better WLTP than the Tesla 3 LR, this is the recent upgrade car rather the original one. I agree about the rear seats, it felt more like a BMW 3 series type space rather than some similar sized EVs where rear space has been maximised. The Tesla supercharger infrastructure is a great reason to choose Tesla, other networks are rolling out fast, but still not as developed as superchargers.



Polestar ADAS is much simpler but has no pretensions to be anything else other than a driver aid. It doesn't add too much to what I have on my aging BMW with front radar and adaptive cruise control.

My fundamental problem with the Tesla approach to marketing FSD is that it deliberately muddies the water between true autonomy and driver aids. FSD is a driver aid and should be considered as such. I mentioned at the beginning of this thread that the automatics on commercial aircraft has been shown to cause human issues (either through complexity, user complacency or unusual automatic reactions) and these are highly trained pilots. Car manufacturers need to take human factors much more seriously in car design and automation. Single large IPAD type screens outside the driver eyeline with no HUD is not great IMO. Automatics need to be designed and marketed in an honest way with care taken around driver involvement - it is no good to have automatics installed which require driver oversight and then not have any consideration to the human interaction component.

Musk and Tesla are in a bit of a bind with autonomy. The Tesla share price (and P/E) are priced as you might expect a high growth tech company rather than an automotive company. P/E on Tesla is an order of magnitude larger than most large auto makers. If Tesla are making cars with autonomy that is only likely to reach level 2 in the medium term, then investors might start to wonder why it is valued so differently to other companies doing similar things. So Musk needs to keep the show on the road, announcing Robotaxi and the low cost car in the last investor call. Do I think Robotaxi will be making their way onto our streets anytime soon? Absolutely not - it is another shiny thing waved at investors to keep them thinking about Tesla as a technology company rather than a car company.

Well there were reports that the M2 had been scrapped, but they seem to be nonsense. The cybertruck probably won't sell outside the US, but the market for ridulous trucks in the US is huge, and very profitable for the likes of Ford and GM. There is some interesting technology in it too, 48v electrical system and a networking system which are a step ahead of their competitors. IMO they are way ahead on autonomy - some of their competition has basically thrown in the towel, and Waymo won't be able to compete. If their machine-learning approach works they have a massive advantage over the competition, with a vast dataset collected from tesla vehicles worldwide - nobody else has anything like that. The machine learning means they can train their system without having to program every detail of behaviour, and they can produce a system that works anywhere in the world - even if Waymo can get their system to work in San Francisco, it's going to be hard to scale it to the rest of the world. Tesla may be valued like a tech company, but that's what they are - a data/AI/robotics/computing company who happen to make cars. Perhaps in the future they won't manufacture cars at all?

 

[albion]

This is what Teslas trumped up Robotaxi has to compete with.
https://fortune.com/2024/05/29/waymo-self-driving-robo-taxi-uber-tesla-alphabet/

Obviously its 'deemed unsafe' autopilot thingy will need lots of new hardware, though at least he has his manufacturing arm for actual cars, unlike Google who buy and adapt.
I sense Google has still been very lucky, even if caution has helped. All but Tesla have no driver to blame when disasters strike.

 

[Bazzer]

"The heat may soon be on again, as Tesla CEO Elon Musk has promised that his company will reveal its own robo-taxi in early August."
Well let's see if that comes to fruition for the greedy vapourware salesman.

 

[the snail]

This is what Teslas trumped up Robotaxi has to compete with.
https://fortune.com/2024/05/29/waymo-self-driving-robo-taxi-uber-tesla-alphabet/

Obviously its 'deemed unsafe' autopilot thingy will need lots of new hardware, though at least he has his manufacturing arm for actual cars, unlike Google who buy and adapt.
I sense Google has still been very lucky, even if caution has helped. All but Tesla have no driver to blame when disasters strike.

You don't half post some nonsense. When was the tesla system deemed unsafe? What makes you think it will need extra hardware, given that IIRC they removed sensors from the system and moved to training on video data?
Tesla reckon their system is 5-10x safer than non-assisted driving:

• Miles per accident with active driver assistance: 7.63 million
• Miles per accident with driver assistance deactivated: 955,000
• Average mileage per accident in the United States: 670,000

https://uk.motor1.com/news/720987/tesla-fsd-safety-study/

 

[albion]

Either that or you were simply too lazy to read the link.

Google initially deemed the technology 'unsafe' thus inferring the Tesla path as 'unsafe'. Google went with safer Lidar yet even that has it's dangers.
Tesla of course need Trump in power so money wins over safety and Robotaxi can exist bigly killingly without Lidar.

 

[Deleted member 159]

As an automotive engineer described autonomous driving vehicles.

"They don't need to be perfect, just better than humans."

The driver safety aids on many vehicles, certainly helps with that.

 

[albion]

https://www.nextbigfuture.com/2024/05/trillions-from-tesla-robotaxi-will-come-from-china-first.html
A quick search on that biggest market, China shows those top 5 autonomous taxi companies all use Lidar meaning Musk will be forced to use Lidar. FSD, as is, will almost certainly be deemed far too unsafe.

 

[stowie]

Well there were reports that the M2 had been scrapped, but they seem to be nonsense. The cybertruck probably won't sell outside the US, but the market for ridulous trucks in the US is huge, and very profitable for the likes of Ford and GM.

The cheap Tesla rumours did seem to be wrong, although we need to wait and see what is released. Tesla are capable of creating a great small car, but I don't trust Musk's investor call commitments!

Truck market is large, I guess the question is whether a radically different electric truck will appeal to a large part of that market. I think the design is deeply flawed and the execution has been worse. But then, I probably am not the target market for a 'merican poor pastiche bladerunner wannabe 4x4.

here is some interesting technology in it too, 48v electrical system and a networking system which are a step ahead of their competitors.

48V has been around for years, but car manufacturers have been reluctant to change out all the equipment to accommodate. There are a number of hybrids / mild hybrids with 48V systems on board, as it makes a lot of sense for this type of vehicle - although many of these will be mixed systems with 12V stepdown for legacy systems - cheaper but still retains the advantages of smaller wiring looms etc.

IMO they are way ahead on autonomy - some of their competition has basically thrown in the towel, and Waymo won't be able to compete. If their machine-learning approach works they have a massive advantage over the competition, with a vast dataset collected from tesla vehicles worldwide - nobody else has anything like that. The machine learning means they can train their system without having to program every detail of behaviour, and they can produce a system that works anywhere in the world - even if Waymo can get their system to work in San Francisco, it's going to be hard to scale it to the rest of the world. Tesla may be valued like a tech company, but that's what they are - a data/AI/robotics/computing company who happen to make cars. Perhaps in the future they won't manufacture cars at all?

I don't think Waymo and Tesla are really on the same path. Tesla makes good (or very good) driver aid systems at a price that is feasible for their cars which they hope they will be able to evolve to full general autonomy. Waymo set out for autonomy from the first instance with a no limit budget using another manufacturer's base vehicle. Waymo aren 't looking to sell vehicles, but they definitely are in the business of AI and advanced algorithms. The geographic limit is useful to Waymo for a couple of reasons. Firstly, it bounds a very complex problem. Secondly, it allows full autonomy with backup - something goes wrong and the Waymo stops to await a local Waymo employee to come and sort out the issue.

Corner cases kill autonomy. In pareto style, 80% of the functionality takes 20% of the time and the remaining 20% is a nightmare.

Currently, Tesla manage corner cases by having the driver as authority. Waymo manage it by having the vehicle stop and someone run out to fix it. Autonomous general driving has Tesla closer but still a long way away, but Waymo is far closer technically and commercially with the concept of autonomous taxi. Bear in mind, taxis are generally short trips in city - geographic bounding is fine, and the initial cost of a taxi vehicle can be much higher than a private car and still make economic sense.

Data is definitely an interesting one. Data is key. Waymo has a very large amount of real data, but Tesla has truly vast amounts of real data. Waymo has made extensive use of simulated data, and this is becoming a big thing in AI - just look at NVIDIA Omniverse to see how much money is being spent on simulated data generation tools.

 

[icowden]

Currently, Tesla manage corner cases by having the driver as authority. Waymo manage it by having the vehicle stop and someone run out to fix it.

Actually most of the time a human driver starts driving the car remotely. It's not as robot controlled as it pretends. If the computer is unsure, someone in a call centre takes control of the car and pilots it remotely, or tells the robot what decision to make.

 

[icowden]

A quick search on that biggest market, China shows those top 5 autonomous taxi companies all use Lidar meaning Musk will be forced to use Lidar. FSD, as is, will almost certainly be deemed far too unsafe.

That's nonsense again. Just because one company chooses to use LIdar it doesn't follow that it's the best route. Musk and Tesla are going a different route. Their argument is that if a human driver can use vision, then a robot driver should be able to do the same.

 

[stowie]

That's nonsense again. Just because one company chooses to use LIdar it doesn't follow that it's the best route. Musk and Tesla are going a different route. Their argument is that if a human driver can use vision, then a robot driver should be able to do the same.

I have heard Musk say that Humans have eyes and a brain and the Tesla has eyes and a brain. Which is an utterly facile statement. Firstly, a human brain has somewhere around 84bn neurons. Any semiconductor "brain" even one from Tesla is so far off this complexity that it is difficult to make an analogy on the difference. And yes, we use eyes but they might not be the best sensor combination for a task which evolution never had "in mind" and certainly not when the compute power available to stitch together and overcome the deficiencies is limited.

My pet guinea pigs had eyes and brains but I find it unlikely they would have managed to drive a car no matter how much time and data we gave them.

As I said above, the corner cases kill autonomy. Tesla has only vision sensors and both vision and LIDAR are terrible in heavy rain or fog. Tesla gets past this currently by handing control to the driver, making their systems Level 2. Waymo have radar which helps to a certain extent (and is getting more granular very rapidly), but ultimately falls back on local support available quickly, so Level 4/5 but with an extensive support structure when it falls down. The difference between high autonomy for a qualified driver and total autonomy that doesn't need immediate human override is vast. I believe Musk and Tesla have made it very difficult for them to reach the latter through their sensor choices, although I understand why they made those choices (commercial considerations being the top one). For autonomous taxis, the commercial dynamics are different and I think the vision sensor only route makes a lot less sense than for a private vehicle.

 

[albion]

That's nonsense again. Just because one company chooses to use LIdar it doesn't follow that it's the best route. Musk and Tesla are going a different route. Their argument is that if a human driver can use vision, then a robot driver should be able to do the same.

Can you also not read?
I said the main 5 in China all use Lidar. Musk will go with Lidar too, it totally insane not to.

 

[albion]

Anyway, it is highly possible that Robota taxi onky surfaced in order to persuade shareholders to give away that $56 billion. The threat to 'go elsewhere' is a co-part of that persuasion.