Tesla Talk

listermint

https://www.boards.ie/discussion/comment/121994653#Comment_121994653

It won't and they won't. Stop continually buying into the hype.

Theres zero chance European regulators will allow teslas stripped down and let's face it cheaper 1 sense only version of self driving out and about unbridled on the streets without a human.

Some folks (investors) fall for the hubris though

unkel

https://www.boards.ie/discussion/comment/121994738#Comment_121994738

Maybe you should familiarise yourself a bit more with the UNECE DCAS regulations and procedures before calling me a liar. This is from the UNECE itself - the organisation that has sole responisibility in Ireland, amongst many other countries. Note there is no mention of lidar / radar being required. There never was

"New UN regulation paves the way for the roll-out of additional driver assistance systems"

https://unece.org/media/press/387961

Also honestly, do yourself a favour. Don't stick your head in the sand "lalala it will never happen" and have a look how the latest version of the beta software now drives

https://www.youtube.com/watch?v=5kwn45Ongvo

Conar

https://www.boards.ie/discussion/comment/121994653#Comment_121994653

I'd be delighted if your optimism is warranted but I doubt it to be honest.
I can't go out at night without it telling me cameras are unreliable. My eyes don't stop working at night, neither does lidar / radar.

listermint

https://www.boards.ie/discussion/comment/121994780#Comment_121994780

Just lol,

Sorry Unkel you keep trying to pump your Tesla stocks as an IBM future. Nah.

You won't see a Tesla robot taxi in Europe with their version of self driving.

You know this though as do I. So what's the point of the pantomime.

unkel

https://www.boards.ie/discussion/comment/121994825#Comment_121994825

Quoted this just to prove you wrong. We could have to wait a while for it though 😂

Nothing to do with pumping TSLA stocks, sometimes I have that, sometimes not, would never want to pump it on people. In fact I am extremely careful never to give people financial advice on any investments in general.

I am excited about a changing renewable future and interested in self driving technology. My Tesla does have FSD though (that's why I bought this particular car), so would love to have the beta software myself

listermint

https://www.boards.ie/discussion/comment/121994962#Comment_121994962

I'll be here. We can share a beer.

sk8board

firstly, the Tesla robotaxi is currently just one tweet - so gods knees really which decade something might happen.

He tweeted in 2018/19 that we’d have 1m robotaxis in 2020 and let’s face it we all know what happened.

The real issue with robotaxis is hem having to share the road with non self driving cars - and impatience from paying ride customers as it stumbles around some situation that a human wouldn’t even consider an obstacle.

It will happen - but it’s genuinely years ans years away.

Read up on Waymo and Cruise’s recent retreats and tell us how Tesla can do better.

I don’t believe a word Mysk says, and he tweeted that one tweet on the day of the Reuters story.

It stinks of the same smell as FSD/Semi/Roadster 2/Cybertruck/model 2/Optimus etc etc etc etc.

a half assed presentation on 8/8 and then vanish for years.

cnocbui

https://www.boards.ie/discussion/comment/121994653#Comment_121994653

No Tesla has a processor within a light year of a human brain, it's a facetious argument. As for being safer than a human, that's another of Musks near endless lies. You can't say anything about the safety of a self driving system without knowing the human intervention rate where the guinea pig has to intervene to keep from becoming bacon, and I am not aware of Tesla ever publishing that extremely important statistic.

Here's an experienced Tesla owners take on the subject:

https://www.youtube.com/watch?v=2DOd4RLNeT4

ELM327

I've had 3 teslas and each got worse than the last one on the "full self driving".

The worst, by far, is my current M3 with tesla vision. I barely trust it for adaptive cruise on the motorway let alone anything else. There is no way this is anywhere close to even being in the realm of full self driving. If Tesla let this loose as actual full self driving (where the OEM is liable, not the owner's insurance), the company would go bankrupt.

Full self driving needs radar, lidar and vision. You can make geo limited full self driving (like Bluecruise) work with 2 of the 3. But actual robotaxi full self driving will not work on vision only. To be honest, trying to jam that square peg into a round hole will be more damaging in the medium term as people see it simply doesnt work.

KCross

One argument that is often made about the dropping of lidar/radar is that "sure humans work on vision, so can the car".

The fallacy there is that a static camera(s) in a car is nowhere close to the same as our vision where we have 3d, the ability to move our eyes and eyelids to keep the lens clean…. as plenty have said, Tesla havent even managed to get the wipers to work with vision, so FSD is a long long way off. When I see a video of a Tesla doing the ring of Kerry in bad weather we can have another chat about it! 😀

A few anecdotal videos where it performs well doesnt cut it.

A good few years ago someone on here asked whether it was worth paying for FSD as they were about to buy a new Tesla…. I think it was about €5k at the time. I said it wont be working in the lifetime of your car so no point in giving the money to Elon. Keep your money. That was true then, the same is true today.

unkel

https://www.boards.ie/discussion/comment/121995815#Comment_121995815

You never had FSD beta, you only had autopilot which is a completely different beast. Like comparing the capabilities of a monkey to that of a slug 😂

ELM327

https://www.boards.ie/discussion/comment/121995851#Comment_121995851

A-f00king-men to that. We have vision, as humans, but we have other senses. I'm looking here typing this at my screen but I can hear the wind to the right of me and a ww2 documentary in the background. I can feel that my orientation is correct even if I close my eyes.

I spent 2 hours on sunday on the motorway with the wipers on, no rain to be seen. Couldnt turn it off as I can't turn off wipers when AP is active.

https://www.boards.ie/discussion/comment/121995884#Comment_121995884

I had FSD in my X (the good one, with NOAP that took you off the motorway exit without prompting), and AP1 highway autopilot in my S. In addition, I drove several loaners with FSD at various stages of development. Regardless of what the enabled software is, that's not the issue. The issue is hardware. You can have the best horses in the world in your cavalry unit, but if the enemy has panzer tanks you'll still lose. And probably get killed.

slave1

We live in a damp country and no system can work when it's cameras are blocked, during rain Saturday evening on motorway my M3 even disengaged cruise control as all cameras became blocked at once.

AP1 in my 2015 Tesla remains superior to my 2025 Tesla

unkel

https://www.boards.ie/discussion/comment/121995942#Comment_121995942

You had the hardware, but never the software. Which has hugely evolved even in the last few months as it is now based on machine learning and not code. You really should have a look at what this software can now do. Of course it is still in beta and only level 2. And for sure it is different in sunny weather in California compared to the brutal weather we get here. But it will be general release and level 4/5 at some stage.

wassie

I would of agreed in the past with the sceptics here about FSD and unrealistic time frames. However with the advancements in machine learning and AI that I have witnessed in just the last 12 months alone, the technology will be here much faster than people expect and Tesla are in the mix. The biggest hurdle will be regulatory approval.

ELM327

https://www.boards.ie/discussion/comment/121996169#Comment_121996169

I have never had the hardware and software at the same time.

The hardware was best in 2020-2021. The software is probably best now, I watch a lot of the YT channels showing FSD and it's impressive but it looks like what I'd call a pareto release. It does 80% very well, but the other 20% will take 4X the effort (and cost) to develop than the current 80%

sh81722

https://www.boards.ie/discussion/comment/121996246#Comment_121996246

I think the development cost is going to be less than expected due to machine learning. The AI will just need to figure out how to deal with the issues when flagged. And I suspect every time a driver override happens the car will flag this and the issue is simulated until a solution is found.

sk8board

I saw an interesting tweet earlier today where some Musk bot was a passenger in a Waymo driverless iPace and remarked how great it’ll be when that’s a Tesla on “8/8” and a Waymo guy replied to list the hundreds of thousands of dollars of cameras (29), LiDAR and server-grade hardware that was in the car. He described the trial vehicles as data centers on wheels - and Musk just tweeting “robotaxi 8/8” changes nothing really

KCross

https://www.boards.ie/discussion/comment/121996211#Comment_121996211

the thing with AI is that it’s only as good as the data it’s fed. If the cameras are obscured in any way there is nothing AI can do for you then, other than go into a fail safe mode and stop.

I can’t see how Tesla can deliver true FSD with current hardware. They may deliver something that’s streets ahead of everyone else but it’s those edge cases are the problem.

listermint

https://www.boards.ie/discussion/comment/121998589#Comment_121998589

Exactly this. Working with AI analysed remote video it's incredibly difficult to get accuracy. Machine learning or not the camera being the sole sensory input is probably the biggest and most relied on achilles heal. You can't have a vehicle carrying passengers with other vehicles in the road with such an open goal for risk.

In a world where aircraft require multiple sensory inputs for decisions people thing we will role back decades of safety regulation for 1 company need to really take a step back and have a good think about it.

unkel

https://www.boards.ie/discussion/comment/121998510#Comment_121998510

You do realise this is the polar opposite of the machine learning approach that Tesla uses? Waymo and the like are trying to hard program everything into the car, geo fencing and all. This was always doomed to fail if it ever had to compete with deep learning AI

sk8board

https://www.boards.ie/discussion/comment/121999428#Comment_121999428

yes, but my point was about the hardware, and the costs - Tesla would have you believe the hardware and chips can be taken from your PlayStation.

There’s such a massive massive question too about how robotaxis will exist alongside the many decades of normal cars that will still be sold.
again to the point though - “robotaxis 8/8” is just another fabrication. They might do a presentation of something, but like all the other non-car manufacturer projects , itll will vanish for many years - indeed in the case of robotaxis this is an absolute certainty due to the basic complexity.

If he’s so far away from figuring out level 3 FSD (and genuinely might never achieve insurance-liability levels of safety), then a driverless car with paying customers is pipe dream stuff for Tesla if they continue without lidar.

listermint

https://www.boards.ie/discussion/comment/121999428#Comment_121999428

I get the impression you don't know a huge amount about machine learning and are using the term as being a catch all as if it's near sentient and can drive the car by itself given enough processor power.

Top tip. Any AI is only as good as its source data the inputs.

Tesla stripped that all out a few years ago. I'm sure their AI is advanced and the team are doing sterling work. But it requires a vastly improved multiple sensor array to be capable on the road for the multitude of situations driving in real world conditions requires.

ELM327

https://www.boards.ie/discussion/comment/122000334#Comment_122000334

+1

As someone who worked in what we would now term primitive generative AI like 10 years ago, AI can appear sentient but the problem is it has no idea what it is doing and why. If you think of generative AI as like a super fast script to do a ctrl+f in a text file a trillion lines long, the most it can do is learn from prior searches and adapt for the known environment and parameters. But it doesn't know that "there's a car approaching from my right", for instance, it knows that parameter 37 has increased from 0.35 to 0.78, and there's a corresponding logic built to handle that. It's like I say to anyone relying on a tableau or looker dashboard for "the reports" - it's only as good as the coder who built it.

The more "AI" scripting I work with, the less optimistic I am that we'll see actual FSD within 10 years. Actual FSD being full hands off, no driver required, no steering wheel, no geo fencing, and no insurance (as the liability lies with the OEM).

AI is the current buzz word. In all things tech, these pass and in 6-18 months the next fad will take off.

sk8board

Great explanation.

I think the business case and insurance liability is where Cruise and Waymo and many other trials are now starting to be shelved -

does it work: yes.
Does it work well enough for companies like Tesla to assume insurance liability without having to charge $15k for it?: hell no

even if my some miracle, Tesla got their camera only FSD approved as level 3, what would you be prepared to pay for it?

take up on FSD internationally is zero, and in the US it was 6% a few years ago, before people realised it wasn’t ready

liamog

https://www.boards.ie/discussion/comment/122000717#Comment_122000717

It's fun to see Waymo's trials being shelved by expanding into LA, doesn't seem like a scale back to me

https://www.independent.co.uk/news/waymo-ap-los-angeles-google-san-francisco-b2512223.html

Although Waymo isn't charging for rides in its robotaxis in Los Angeles to start, the company said in a blog post announcing the expansion that it will eventually collect fares from passengers there too. Waymo also hopes to begin commercial operations in Austin, Texas, later this year, a goal that makes its robotaxi service available in four major U.S. cities 15 years after it began as a secret project within Google. Waymo's robotaxis have been charging for rides in Phoenix since 2020.

ELM327

https://www.boards.ie/discussion/comment/122000717#Comment_122000717

I would only pay for full self driving, actual full self driving, that can drive me home from the pub capably and legally.

zg3409

https://www.boards.ie/discussion/comment/122000717#Comment_122000717

$15k for taxi insurance where no driver is required would be a bargain.

I would expect a driverless taxi to retail for 200k but in reality it would be some form of licence, rental or branding. At least at the start individuals are not going to be purchasing full driverless cars and as said elsewhere it's likely city to city trucks will be the first realistic use of this technology where the route is straight and simple and a real driver can take over for last mile. An extra 200k or 400k on a truck that needs no driver makes financial sense quickly. If it can do 10 runs per 24 hours it may be the cheapest option to transport between city depot's.

liamog

https://www.boards.ie/discussion/comment/122000954#Comment_122000954

I don't expect a driverless taxi to ever be sold to a third party by its manufacturer. If you can sell it for $200k, that means someone else can make at least that much from running it. In that case so can the manufacturer. I think it's more likely to be car is "sold" to separate mobility services operating company under the control of the maker.

Whether they achieve it or not, I think that's a big reason why it makes sense for Tesla to cancel an affordable consumer car, from a share price point of view the company is gambling on big profits from selling many people journeys instead of selling a car once.

It reminds of the pivot that happens as companies move from on-prem software to SaaS. The revenue per customer is valued much higher for a SaaS customer. You see companies end their on-prem development before the demand is gone to move customers to the more valuable platform.

sk8board

https://www.boards.ie/discussion/comment/122000954#Comment_122000954

what you’re suggesting is all correct and valid - but it’s the dream scenario, many decades into the future