Tesla unlikely to Survive (Vol. 3)
Discussion
I have a hunch that if Tesla were starting from a clean sheet today they’d do things very differently, but they have too much legacy out there they have to deal with. High resolution cameras on top left and right of the windscreen offering better visibility past traffic in front and the ability to triangulate distance more accurately rather than mono vision. Same at the back. Heaters in all cameras to prevent fogging (the image I posted as condensation inside the unit/behind the b pillar glass and I’ve seen the same on plenty of Tesla). More cameras covering the side including downward facing helping both parking and better lane detection rather than relying on forward facing cameras for lane accuracy, I’m sure the list could go on.
It all looks very pretty to see all that traffic around the car, but then AI trying to navigate it is placing far to great an emphasis on how good AI is, AI gets you so far, but look at ChatGPT, deep fakes, cancer detection, etc, they’re not perfect, they might be credible in what they do, but use them enough and you realise you can either tell eventually or need a human in parallel, and these are relatively simple tasks compared to driving and with much lower consequences if they get it wrong (the cancer detection consequence is obviously high but it’s only used as a back up)
It all looks very pretty to see all that traffic around the car, but then AI trying to navigate it is placing far to great an emphasis on how good AI is, AI gets you so far, but look at ChatGPT, deep fakes, cancer detection, etc, they’re not perfect, they might be credible in what they do, but use them enough and you realise you can either tell eventually or need a human in parallel, and these are relatively simple tasks compared to driving and with much lower consequences if they get it wrong (the cancer detection consequence is obviously high but it’s only used as a back up)
TheDeuce said:
durbster said:
How much is it going to cost to cover the car in high-end cameras with sophisticated lenses?
A top end smartphone camera must be £600-800 of the retail price of the phone and you'd need at least, what, eight of them to even begin to achieve anything useable in different conditions.
Are you serious? The bulk buy price of the best smartphone cameras is probably less than $20A top end smartphone camera must be £600-800 of the retail price of the phone and you'd need at least, what, eight of them to even begin to achieve anything useable in different conditions.
Note, I did say retail price so in other words, I would think at least half the price of top end smartphones is the camera.
That's based on there never being a cheap smartphone that has a top-spec camera. You can get pretty much everything else but if you want image quality, you have to spend.
I can't seem to find anything conclusive on manufacturing costs though so maybe you're right.
durbster said:
TheDeuce said:
durbster said:
How much is it going to cost to cover the car in high-end cameras with sophisticated lenses?
A top end smartphone camera must be £600-800 of the retail price of the phone and you'd need at least, what, eight of them to even begin to achieve anything useable in different conditions.
Are you serious? The bulk buy price of the best smartphone cameras is probably less than $20A top end smartphone camera must be £600-800 of the retail price of the phone and you'd need at least, what, eight of them to even begin to achieve anything useable in different conditions.
Note, I did say retail price so in other words, I would think at least half the price of top end smartphones is the camera.
That's based on there never being a cheap smartphone that has a top-spec camera. You can get pretty much everything else but if you want image quality, you have to spend.
I can't seem to find anything conclusive on manufacturing costs though so maybe you're right.
I doubt an S24 ultra has a component price of more than $200.
skwdenyer said:
The only thing I believe I currently know is that both HW3 & HW4 cameras are HDR capable.
It turns out that you need surprisingly little resolution and optical fidelity to do object detection. Most people when starting out playing in this field feed their detection algorithms with images of far too high a quality, resulting in massive spikes in compute requirement for no actual benefit. We humans like to see nice shiny 4K images, but the detection algo really doesn’t care.
The secret sauce is in the optimisation of the image pipeline for the intended purpose, the inference engine, and the general compute algorithms.
So long as the cameras can see enough to get the detail they require, that’s about it.
There are some cool tricks you can use if you do have a higher-resolution sensor, such as foveation. But all the cool stuff you see in many demos I could broadly replicate for you with a simple HD camera, a Raspberry Pi and a (now) pretty cheap GPU or something like a Coral TPU.
I’m talking stuff like this from MobilEye:
(I might need a touch more compute to do that at 40 FPS, to be fair, but not by much)
That colour coding is just a simple segmentation algorithm that needs surprisingly cheap (nowadays) hardware to run. Object detection (putting boxes around stuff and identifying it) is another (today) trivial problem. Figuring out the extents of the road and so on are also straightforward in compute terms.
The secret sauce is the training. Not training for every possible case (impossible), but training to know what a road looks like, where the edges are, what the likely behaviour of a moving vehicle is, and so on.
That’s what Tesla is throwing masses of effort at right now with Dojo. And where they differ fundamentally from MobilEye - who instead rely on HD mapping to tell the car where the road is, and then use the onboard sensors to find where reality today differs from the map.
In terms of the obscured camera, you’re right, that’s a problem. For FSD, just asking the human to clean tye cameras is easy. For robotaxis, not so much.
The mistake many make is in assuming that LiDAR is any better. It isn’t. If the sensor is obscured by mud or whatever, it is dead.
Radar might be better, but that has a bunch of other issues to worry about.
Ironically, the optimum solution might be an Optimus robot sitting in the driver’s seat
Without doubt the technology will get there, but not with the layout/hardware Tesla have right now. A simple cleaning mechanism would help with dirty cameras, but Tesla don't have any.It turns out that you need surprisingly little resolution and optical fidelity to do object detection. Most people when starting out playing in this field feed their detection algorithms with images of far too high a quality, resulting in massive spikes in compute requirement for no actual benefit. We humans like to see nice shiny 4K images, but the detection algo really doesn’t care.
The secret sauce is in the optimisation of the image pipeline for the intended purpose, the inference engine, and the general compute algorithms.
So long as the cameras can see enough to get the detail they require, that’s about it.
There are some cool tricks you can use if you do have a higher-resolution sensor, such as foveation. But all the cool stuff you see in many demos I could broadly replicate for you with a simple HD camera, a Raspberry Pi and a (now) pretty cheap GPU or something like a Coral TPU.
I’m talking stuff like this from MobilEye:
(I might need a touch more compute to do that at 40 FPS, to be fair, but not by much)
That colour coding is just a simple segmentation algorithm that needs surprisingly cheap (nowadays) hardware to run. Object detection (putting boxes around stuff and identifying it) is another (today) trivial problem. Figuring out the extents of the road and so on are also straightforward in compute terms.
The secret sauce is the training. Not training for every possible case (impossible), but training to know what a road looks like, where the edges are, what the likely behaviour of a moving vehicle is, and so on.
That’s what Tesla is throwing masses of effort at right now with Dojo. And where they differ fundamentally from MobilEye - who instead rely on HD mapping to tell the car where the road is, and then use the onboard sensors to find where reality today differs from the map.
In terms of the obscured camera, you’re right, that’s a problem. For FSD, just asking the human to clean tye cameras is easy. For robotaxis, not so much.
The mistake many make is in assuming that LiDAR is any better. It isn’t. If the sensor is obscured by mud or whatever, it is dead.
Radar might be better, but that has a bunch of other issues to worry about.
Ironically, the optimum solution might be an Optimus robot sitting in the driver’s seat
Here is another example. Lack of stereoscopic vision to the side. One of the significant issues, with the UK implementation of FSD at least, is that when passing a large vehicle on a motorway, or any multi lane road, is that often when passing and pulling back over to the left from lane 3 to lane 2, the car will suddenly decide that the lorry in lane 1 has jumped to lane 2. This causes a sudden abort of the manoever with the car quickly swerving back into lane 3. The car behind usually sounding their horn or flashing you and thinking you are generally a t
t. With the large slab side of the lorry at a point where there is only one camera, there is no perception of distance. The result is down right dangerous.skwdenyer said:
But the future is, as Musk says, autonomy. That's where Tesla is still in a really interesting position. Mobileye (recall, Tesla's original partner), was bought by Intel for $17bn, amidst promises of their own self-driving revolution. A limited IPO (Intel sold just 5%) valued the business at the same $17bn. Its had a bit of a rollercoaster, and is now valued at about $24bn, but seems to have a pretty tough market.
The recent past is littered with autonomy dead ends. Ford gave up on L4 and wrote off a $2.7bn investment in Argo AI in 2022 (it had partnered with VW). Uber sold its self-driving business to Aurora Innovation in 2020 (whose stock has lost nearly 80% of its value since 2022). Didi spun theirs out to Xpeng, who likewise have lost most of their share value.
Mobileye are still in the game, of course, announcing recently a project with VW to deliver L4 autonomous driving on the ID.Buzz platform from 2026. Their Mobileye SuperVision product seems to be promising essentially the same sort of performance as we've seen from Mercedes recently - point-to-point automatic navigation on highways and urban expressways. Their next product is Mobileye Chauffeur (which will be used on the VW ID.Buzz), which seems to put in a bunch of redundant systems to replace the driver's eyes in monitoring what's happening. It still isn't what we'd call "Full Self Driving" however; it is a combination of mapping, other data sources, cameras, sensors, etc. It is reliant upon the quality of mapping and other data available from infrastructure operators.
MobilEye drive is the analogue in the urban "robotaxi" market - very detailed mapping of defined geographical areas.
That mapping is absolutely key. To deliver self-driving, Mobileye literally have to map the world. Now, I've actually run a start-up project aimed at delivering this sort of mapping (for different purposes). It is unbelievably difficult to deliver anything that's actually reliable enough to use for autonomous navigation. Not at all impossible, and I hugely admire them for trying to do it. But it isn't delivering what most would call "real" self-driving.
Now, that 2026 timeline seems agressive. In 2018 they said EyeQ4 was launching then and would allow "eyes off" autonomy, whilst EyeQ5 would launch in 2020 and offer "mind off" driving. That obviously didn't happen.
Ultimately the vision only approach or possibly vision plus basic maps will IMHO be the long term solution to anonymous driving and Telsa are right in going down this route. However what isn't obvious to me is if the now promised timelines are in way realistic, where they haven't been realistic so far. I suspect that timelines for all are still unrealistic and in the medium term heavily augmented and restricted systems are going to be the solution. The recent past is littered with autonomy dead ends. Ford gave up on L4 and wrote off a $2.7bn investment in Argo AI in 2022 (it had partnered with VW). Uber sold its self-driving business to Aurora Innovation in 2020 (whose stock has lost nearly 80% of its value since 2022). Didi spun theirs out to Xpeng, who likewise have lost most of their share value.
Mobileye are still in the game, of course, announcing recently a project with VW to deliver L4 autonomous driving on the ID.Buzz platform from 2026. Their Mobileye SuperVision product seems to be promising essentially the same sort of performance as we've seen from Mercedes recently - point-to-point automatic navigation on highways and urban expressways. Their next product is Mobileye Chauffeur (which will be used on the VW ID.Buzz), which seems to put in a bunch of redundant systems to replace the driver's eyes in monitoring what's happening. It still isn't what we'd call "Full Self Driving" however; it is a combination of mapping, other data sources, cameras, sensors, etc. It is reliant upon the quality of mapping and other data available from infrastructure operators.
MobilEye drive is the analogue in the urban "robotaxi" market - very detailed mapping of defined geographical areas.
That mapping is absolutely key. To deliver self-driving, Mobileye literally have to map the world. Now, I've actually run a start-up project aimed at delivering this sort of mapping (for different purposes). It is unbelievably difficult to deliver anything that's actually reliable enough to use for autonomous navigation. Not at all impossible, and I hugely admire them for trying to do it. But it isn't delivering what most would call "real" self-driving.
Now, that 2026 timeline seems agressive. In 2018 they said EyeQ4 was launching then and would allow "eyes off" autonomy, whilst EyeQ5 would launch in 2020 and offer "mind off" driving. That obviously didn't happen.
The hubris of Telsa around feasible timelines to L4 autonomy does seem to have infected other "players" in this market space with unrealistic promises and resulted in a lot of startups that have since failed.
However MobilEye appear to be adopting a multitrack approach to autonomy probably reflective of where the "tech" really is rather than where everyone hoped the tech would be.
They have two main ADAS/ISA offerings an augmented camera + sensor(s) + maps approach which has been their core business for the last decade and a vision only ISA system.
The published roadmap sees them iterate the augmented system to eyes off, hands off functionality. I'd suspect the plan is to parallel iterate the vision only system towards this using data harvested from the augmented system to train the vision only system.
Interesting thing with MobilEye is they have since the EyeQ4 generation been generating their own "HD" maps from harvested real vehicle data. Simply put driving around in a EyeQ4 onwards equipped vehicle where the OEM has opted into REM and data sharing improves the map data MobilEye provide to OEM's and I suspect is being used as training data to refine the algorithms behind the system.
MobilEye have recently started to offer said vision only ISA that doesn't require any maps and is apparently a software only upgrade for EyeQ4 units. Vision only doesn't sound like a particular advancement but in reality to achieve it they will have needed to crack road type identification which means that the problems of lane, edge and turn/side road detection will also have been solved without relying on a pre-calculated database.
RichardM5 said:
Without doubt the technology will get there, but not with the layout/hardware Tesla have right now. A simple cleaning mechanism would help with dirty cameras, but Tesla don't have any.
Here is another example. Lack of stereoscopic vision to the side. One of the significant issues, with the UK implementation of FSD at least, is that when passing a large vehicle on a motorway, or any multi lane road, is that often when passing and pulling back over to the left from lane 3 to lane 2, the car will suddenly decide that the lorry in lane 1 has jumped to lane 2. This causes a sudden abort of the manoever with the car quickly swerving back into lane 3. The car behind usually sounding their horn or flashing you and thinking you are generally a tt. With the large slab side of the lorry at a point where there is only one camera, there is no perception of distance. The result is down right dangerous.
I agree that vision-only is hard. Depth perception in the scenario you describe isn’t impossible. For a start, the side camera should be able to see the road Here is another example. Lack of stereoscopic vision to the side. One of the significant issues, with the UK implementation of FSD at least, is that when passing a large vehicle on a motorway, or any multi lane road, is that often when passing and pulling back over to the left from lane 3 to lane 2, the car will suddenly decide that the lorry in lane 1 has jumped to lane 2. This causes a sudden abort of the manoever with the car quickly swerving back into lane 3. The car behind usually sounding their horn or flashing you and thinking you are generally a t
t. With the large slab side of the lorry at a point where there is only one camera, there is no perception of distance. The result is down right dangerous. But it is difficult to do, and I don’t underestimate the task Tesla have set themselves here.Don’t forget, however, that one-eyed humans can drive quite adequately. Depth perception isn’t all about (or even mostly about) stereoscopic vision, and in fact stereoscopic depth perception in computer vision really isn’t all that great.
The point I’ve been trying to make is that solving the general case *driving* task is IMHO legitimately a vision problem. It may be that a final level 5 system will require some simple lidar or other sensors to provide additional depth measurement to self-calibrate the vision, but that’s not at all the same thing as relying upon those sensors for the driving task. Vision is far more useful in terms of understanding what’s actually going on around the car
But I agree that Tesla’s hardware stack may well be a case of the “road to Dublin problem”
This is odd. Reports that Elon is getting rid of the Supercharger team and the New Model Development team.
Very strange.
https://www.reuters.com/business/autos-transportat...
Very strange.
https://www.reuters.com/business/autos-transportat...
Jader1973 said:
This is odd. Reports that Elon is getting rid of the Supercharger team and the New Model Development team.
Very strange.
https://www.reuters.com/business/autos-transportat...
The new model dev team is odd - they're no longer going to develop new cars...!?Very strange.
https://www.reuters.com/business/autos-transportat...
The supercharger thing isn't entirely weird. There's no reason Tesla should want to own more locations than they already have, nor any reason they should need to maintain activity in the charger sector. They built the chargers to sell the cars,and if course that was extremely welcome and necessary. But now the number of non Tesla rapid+ chargers per non Tesla EV in many countries actually exceeds the ratio of Tesla chargers for Tesla cars, so that need is probably no longer there. I know it's sacrilegious to suggest the Tesla charger supremacy is drawing to a close... But it is. The numbers don't lie.
Is there still a reason Tesla needs to design and build their own chargers and maintain their own network?
I suppose I'm bound to get flamed by Tesla drivers for even suggesting that their own network of chargers might not be a forever thing
TheDeuce said:
The new model dev team is odd - they're no longer going to develop new cars...!?
All I can think of is that they’ve just completed a refresh of everything and the mythical Robotaxi, and won’t be doing anything else for another 5 to 10 years.So in Elon’s mind that means they don’t need the team anymore, and they’ll just hire a new one when they do.
You’re probably right about the Superchargers.
Many see the supercharging network as a Tesla USP. Anything that erodes confidence in that going forward, either reducing future investment or opening to others reduces the Tesla appeal. It was taking away the one big concern many had moving from ICE to EV, charging away from home, and whilst it may be an increasingly irrational argument, people’s concerns are concerns whether rational or not. So either charging away from home is a non issue whichever brand of EV you’re looking to buy, or only Tesla have it properly solved and they’ve reduced investment, either way it’s not reinforcing Tesla sales.
Musk seems to be really doubling down on the ‘we’re not a car company’ theme, but even Robotaxis will need to charge somewhere, you’d have thought he’d be redirecting the supercharging team to build automated charging hubs if they were going to be reality any time soon.
Musk seems to be really doubling down on the ‘we’re not a car company’ theme, but even Robotaxis will need to charge somewhere, you’d have thought he’d be redirecting the supercharging team to build automated charging hubs if they were going to be reality any time soon.
Edited by Gone fishing on Wednesday 1st May 08:13
Gone fishing said:
Many see the supercharging network as a Tesla USP. Anything that erodes confidence in that going forward, either reducing future investment or opening to others reduces the Tesla appeal. It was taking away the one big concern many had moving from ICE to EV, charging away from home, and whilst it may be an increasingly irrational argument, people’s concerns are concerns whether rational or not. So either charging away from home is a non issue whichever brand of EV you’re looking to buy, or only Tesla have it properly solved and they’ve reduced investment, either way it’s not reinforcing Tesla sales.
Musk seems to be really doubling down on the ‘we’re not a car company’ theme, but even Robotaxis will need to charge somewhere, you’d have thought he’d be redirecting the supercharging team to build automated charging hubs if they were going to be reality any time soon.
I'll leave the robotaxi thing to one side for the moment - I simply don't believe that's anywhere remotely close to being a commercial reality.Musk seems to be really doubling down on the ‘we’re not a car company’ theme, but even Robotaxis will need to charge somewhere, you’d have thought he’d be redirecting the supercharging team to build automated charging hubs if they were going to be reality any time soon.
Edited by Gone fishing on Wednesday 1st May 08:13
The chargers are seen as a USP for sure, people talk about the network as such on these very forums all the time. But in business terms you have to look to the future of where things will be in a few years time. If they sack off the supercharger dept today, it wouldn't need to affect the network or the chargers they already have designed and can re-produce for years, it would only signify that longer term they don't have plans to remain at the forefront of public charging.
The none Tesla network of chargers has, over the last 18 months or so, grown exponentially in the UK and before that over much of Europe and China. It's not just the number of chargers either, it's the speed, simplicity (no more apps!!) and reliability of the newer chargers that have negated to Tesla advantage. As you say that's not how the situation is perceived just yet, Tesla have a reputation for making en-route charging easier and that reputation will stick for some time whatever the reality - but as the none Tesla networks continue to grow it's reasonable to consider that in several years time, that reputation will fade and simply won't be a factor in most people's decision of which EV to get. At that point, why bother with a Tesla dedicated network?
In four years of Tesla ownership (45,000 miles) I've used the supercharger network maybe 5 or 6 times. But, perhaps unreasonably, I quite like having it there.
I've attempted to use non Tesla charging twice, one broken one (failed) and one fine - but expensive (comparably). Most non Tesla charging appears to be just one or two fairly slow chargers and quite regularly not working - this is only from casual observation. The Tesla sites usually appear to be pretty reliable.
I suspect what Tesla are doing is saying 'we probably have enough locations, so growth of new locations will slow, but we'll continue expanding existing sites'. Who knows?
Tesla have opened up pretty much all of their locations to other EV's now I believe.
I've attempted to use non Tesla charging twice, one broken one (failed) and one fine - but expensive (comparably). Most non Tesla charging appears to be just one or two fairly slow chargers and quite regularly not working - this is only from casual observation. The Tesla sites usually appear to be pretty reliable.
I suspect what Tesla are doing is saying 'we probably have enough locations, so growth of new locations will slow, but we'll continue expanding existing sites'. Who knows?
Tesla have opened up pretty much all of their locations to other EV's now I believe.
NDA said:
In four years of Tesla ownership (45,000 miles) I've used the supercharger network maybe 5 or 6 times. But, perhaps unreasonably, I quite like having it there.
I've attempted to use non Tesla charging twice, one broken one (failed) and one fine - but expensive (comparably). Most non Tesla charging appears to be just one or two fairly slow chargers and quite regularly not working - this is only from casual observation. The Tesla sites usually appear to be pretty reliable.
I suspect what Tesla are doing is saying 'we probably have enough locations, so growth of new locations will slow, but we'll continue expanding existing sites'. Who knows?
Tesla have opened up pretty much all of their locations to other EV's now I believe.
What you're describing about none Tesla chargers is what I would have said up to about a year ago, but things really have improved a lot recently. The newer banks of rapid+ chargers seem to be 12 chargers or more in a row, installed or being installed at most major services and increasingly in supermarkets, which is very useful.I've attempted to use non Tesla charging twice, one broken one (failed) and one fine - but expensive (comparably). Most non Tesla charging appears to be just one or two fairly slow chargers and quite regularly not working - this is only from casual observation. The Tesla sites usually appear to be pretty reliable.
I suspect what Tesla are doing is saying 'we probably have enough locations, so growth of new locations will slow, but we'll continue expanding existing sites'. Who knows?
Tesla have opened up pretty much all of their locations to other EV's now I believe.
Like I said, it'll take years for people to forget the old charger woes and realise the none Tesla network now 'just works', and that has to devalue to the need for Tesla to bother pushing that aspect of their brand in the future - it's entirely possible the move to reduce R&D of the superchargers is just in preparation for a relatively near future where they wont matter anymore.
As you say, they have enough sites and chargers for anyone that does still see it as a USP or comfort blanket.
TheDeuce said:
What you're describing about none Tesla chargers is what I would have said up to about a year ago, but things really have improved a lot recently. The newer banks of rapid+ chargers seem to be 12 chargers or more in a row, installed or being installed at most major services and increasingly in supermarkets, which is very useful.
Like I said, it'll take years for people to forget the old charger woes and realise the none Tesla network now 'just works', and that has to devalue to the need for Tesla to bother pushing that aspect of their brand in the future - it's entirely possible the move to reduce R&D of the superchargers is just in preparation for a relatively near future where they wont matter anymore.
As you say, they have enough sites and chargers for anyone that does still see it as a USP or comfort blanket.
Good points.... I must say I'm a bit out of touch as I so rarely have need to charge on a journey. Encouraging news about things improving.Like I said, it'll take years for people to forget the old charger woes and realise the none Tesla network now 'just works', and that has to devalue to the need for Tesla to bother pushing that aspect of their brand in the future - it's entirely possible the move to reduce R&D of the superchargers is just in preparation for a relatively near future where they wont matter anymore.
As you say, they have enough sites and chargers for anyone that does still see it as a USP or comfort blanket.
Puzzles said:
A lot of Tesla owners ime value the supercharger network and overlook other issues because of the charging.
Would seem a massive error to sack all. They still have on going projects in the UK don’t they?
They don't need to sack it off, just stop actively developing new chargers once their current plans (whatever they are in the UK) are complete.Would seem a massive error to sack all. They still have on going projects in the UK don’t they?
I think it could make complete sense to save the future development monies today, but keep what they have until some point in the future where it's no longer really needed as a USP (even in people's minds) and at that point they can continue to won the network, sell it to a third party, whatever really - it wouldn't matter anymore.
Puzzles said:
Tesla are between about 25p and 45p iirc which is significantly cheaper than other providers?
That's definitely true, but as above, it's an commercial decision they make to be at that sort of price point, it doesn't require a nuts and bolts R&D effort or the procurement of new sites to continue to use what they have and price it generously if that's a factor in selling Tesla cars.TheDeuce said:
They don't need to sack it off, just stop actively developing new chargers once their current plans (whatever they are in the UK) are complete.
I think it could make complete sense to save the future development monies today, but keep what they have until some point in the future where it's no longer really needed as a USP (even in people's minds) and at that point they can continue to won the network, sell it to a third party, whatever really - it wouldn't matter anymore.
I thought there was a pipeline of new sites, along with upgrades and work in progress. That would need to be managed by someone.I think it could make complete sense to save the future development monies today, but keep what they have until some point in the future where it's no longer really needed as a USP (even in people's minds) and at that point they can continue to won the network, sell it to a third party, whatever really - it wouldn't matter anymore.
More details will come out I guess.
Gassing Station | Tesla | Top of Page | What's New | My Stuff