Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Well, my understanding is that Tesla’s CEO argued that since humans can drive without LiDAR, then surely optical cameras and software should be able to handle it. The second part of course being the high cost LiDAR sensors, especially years ago when Tesla decided to double down on cameras.

Maybe that approach would have had more merit if Tesla had reached 100% autonomous driving capabilities long ago (many projected dates come and gone). But given alll the time that has elapsed without accomplishing full autonomy, LiDAR has become much more affordable and smaller scale.

The Volvo EX90 LiDAR costs about $1000 (yet still unfortunately reminds me of the roof lights on a NYC taxi). Now BYD is saying they have a $400 sensor. Such sensors cost tens of thousands not long ago.Further developments and economies of scale will probably bring costs down further.

The optical-only Tesla system is incredibly impressive. But it seems only reasonable to add LiDAR to augment the cameras. The debate has always been presented as an either-or, but considering the costs are progressively dropping, why not take advantage of the strengths of each system by using both? It’s not like errors can be horrendously consequential in self driving cars? Typically redundancy using input types is desired for such important tasks.

Maybe it’s a cost cutting obsession (granted Tesla has included plenty of stupid features over the years that contradict. Maybe it’s an intellectual challenge to figure out. Maybe it’s an ego thing- perhaps he underestimated the trajectory of LiDAR development and so his ego won’t let him admit his prediction / bet was wrong. Who knows. Either Tesla will develop true, mainstream autonomous driving or someone else will.

But I would not be surprised if LiDAR quietly finds its way into Teslas in the future.

But I have to ask, with such a reliance on cameras and knowing how moderns car’s camera-based safety systems are affected by precipitation/inclement weather, how will Tesla Robotaxis operate in heavy rain, moderate fog, a decent amount of snow (esp at night). Tesla FSD doesn’t work. Neither does Subaru’s comparatively simple safety suite. You know, much of the world doesn’t experience SoCal weather year round.

My suspicion: the Tesla Robotaxis will either end up using LiDAR or the Robotaxi remains nothing more but a mythical concept vehicle, much like the Roadster 2. Whether or not Tesla will the leader in future robotransport remains to been.
Many of these use LDAR.

1743919840661.png
 
The optical-only Tesla system is incredibly impressive. But it seems only reasonable to add LiDAR to augment the cameras. The debate has always been presented as an either-or, but considering the costs are progressively dropping, why not take advantage of the strengths of each system by using both? It’s not like errors can be horrendously consequential in self driving cars? Typically redundancy using input types is desired for such important tasks.

Maybe it’s a cost cutting obsession…

I mean, many cars have been using cameras to assist driver aids. Other brands don’t visually illustrate whats in your windshield on an internal screen like the tesla, so i don’t think there’s anything inherently more impressive about Tesla’s implantation. So far, Tesla is the only brand i hear about their driving aids causing their cars to fatally crashed into white trucks on bright days or parked vehicles at night. And that’s before we bring up phantom braking. I think that’s an acceptable error tolerance for a hobby project, but not for something that many trust with their life.


My suspicion: the Tesla Robotaxis will either end up using LiDAR or the Robotaxi remains nothing more but a mythical concept vehicle, much like the Roadster 2. Whether or not Tesla will the leader in future robotransport remains to been.
I suspect the Robotaxi will follow the path of the Roadster as well. I don’t see how that can be profitable and have commercial success.
 
People tend to not trust other driving aids as much as they do Tesla's system. I figure that we would see more of these kinds of incidents if we did.

While novel the new E2E stack does a pretty good job of not killing me (or even looking like it would) in areas no other driving aid operates (aka none high way driving). It still doesn't handle toll booths well, I also don't trust it in school zones (allegedly it will obey school busses). The speed control is annoying (most often it goes slower than I want).
 
People tend to not trust other driving aids as much as they do Tesla's system. I figure that we would see more of these kinds of incidents if we did.

While novel the new E2E stack does a pretty good job of not killing me (or even looking like it would) in areas no other driving aid operates (aka none high way driving). It still doesn't handle toll booths well, I also don't trust it in school zones (allegedly it will obey school busses). The speed control is annoying (most often it goes slower than I want).
Which people? I wouldn't trust the Tesla system more or less than any other on the list tbh. Why would I?
 
  • Like
Reactions: cyb3rdud3
Which people? I wouldn't trust the Tesla system more or less than any other on the list tbh. Why would I?
That is the issue. It works really well, until it doesn't. You get lulled into thinking it won't make a mistake, then bam it does and you were not prepared. It is easy to get too complacent.

All you guys get right now is the V11 highway stack, which is kinda garbage (comparatively). So FSD isn't actually doing anything for folks outside of US/China (I think Canada may be able to use V12/13 but I'm not positive). So yes it makes sense for all the EU/UK folk to crap on what is available because ya'll are so far behind due to your rules being more restrictive than ours.
 
I mean, many cars have been using cameras to assist driver aids. Other brands don’t visually illustrate whats in your windshield on an internal screen like the tesla, so i don’t think there’s anything inherently more impressive about Tesla’s implantation. So far, Tesla is the only brand i hear about their driving aids causing their cars to fatally crashed into white trucks on bright days or parked vehicles at night. And that’s before we bring up phantom braking. I think that’s an acceptable error tolerance for a hobby project, but not for something that many trust with their life.



I suspect the Robotaxi will follow the path of the Roadster as well. I don’t see how that can be profitable and have commercial success.
Yes… but there’s a difference between cameras for lower level automated aids and using a solely based camera system to operate all autonomous driving features. Whereas other vehicles are also using LiDAR, Radar, Ultrasonic. Maybe thermal/FLIR will find its way into cars automated driving aids, still too $$$ right now (and often export controlled). They all have pros and cons but using a combination of systems give you better reliability, redundancy, and I would assume, to a degree, a way to verify data. To rely solely on CCD camera sensors seems like putting all your eggs in one basket- and we all know the cameras can have a lot of limitations in certain scenarios.

Tesla is the only brand I’m aware of that solely relies on optical cameras for higher level autonomous driving. No radars. No ultrasonic. No Lidar.

If I recall correctly, sometime during COVID (at least probably in part to supply chain disruptions), Tesla dumped their ultrasonic sensors which were used judge proximity.

I think it’s pretty unfair to say Tesla’s FSD is not impressive. The display of vehicles on the LCD has nothing to do with it (I actually find that very distracting). Keep it mind it want that long ago that your new car’s biggest tech advancement was an Aux Jack.

I doubt you’d deny Tesla has not been a leader with advanced automated aids. FSD is the only system I’m aware of currently that can drive off highway, read stoplights, advanced decision making ie confusing traffic situations. Is it a perfect system- definitely not, who knows if we’ll get to fully autonomous driving in our lifetime, but I would argue the past 10-15 years have had far more innovation of vehicles than earlier decades combined. If you look back at my posts, going back years, I have been highly critical of Tesla and Musk- long before it was fashionable. But that doesn’t mean I can’t recognize their impacts on the industry.

GM SuperCruise I actually think is a more practical, safer, and sensible (based on current capabilities of all systems). I also think it’s a very clever solution to the problem at hand. SC of course only works on pre-mapped highways. If hypothetically I was choosing between SC and FSD, it would be SC. I’m a lot more comfortable with GM’s system than the much more unpredictable Tesla FSD.

Mercedes has their “level 3” system but is (currently) so limited in function that most reviews I have seen weren’t that happy with it. Ford BlueCruise I’ve never used. I believe it’s a similar concept to SC. I’m not aware of any other brands with automation at this level of sophistication. Assuming actual full self driving is feasible and practical, ie 10-20 years from now, at some point there probably won’t be much difference in capabilities between automakers.

None of these systems are anywhere ready for full independence from direct human oversight- find me any vehicle with auto emergency braking that hasn’t phantom braked at some
point (usually minor, but even a momentary brake check can be dangerous). In my experience, if you spend enough time in a car, you’ll have it happen. Tesla’s issue I would suggest has to do with their sole use of cameras.

Re: Robotaxis- at best we’ll see some small scale operation in one city so they can say it exists. Like Boring Tunnels. I guess owners sending their vehicles off to Robotaxis while at work isn’t the plan now (the whole “your tesla will earn money” nonsense). Plus, 2-seats is incredibly impractical. If it’s actually stainless steel, Tesla i guess hasn’t learned the lesson about using stainless from the CyberTruck, who should have learned from Delorean 40+ years ago.
 
Last edited:
Yes… but there’s a difference between cameras for lower level automated aids and using a solely based camera system to operate all autonomous driving features. Whereas other vehicles are also using LiDAR, Radar, Ultrasonic. Maybe thermal/FLIR will find its way into cars automated driving aids, still too $$$ right now (and often export controlled). They all have pros and cons but using a combination of systems give you better reliability, redundancy, and I would assume, to a degree, a way to verify data. To rely solely on CCD camera sensors seems like putting all your eggs in one basket- and we all know the cameras can have a lot of limitations in certain scenarios.

Tesla is the only brand I’m aware of that solely relies on optical cameras for higher level autonomous driving

If I recall correctly, sometime during COVID (at least probably in part to supply chain disruptions), Tesla dumped their ultrasonic sensors which were used judge proximity.

I think it’s pretty unfair to say Tesla’s FSD is not impressive. The display of vehicles on the LCD has nothing to do with it (I actually find that very distracting). Keep it mind it want that long ago that your new car’s biggest tech advancement was an Aux Jack.

I doubt you’d deny Tesla has not been a leader with advanced automated aids. FSD is the only system I’m aware of currently that can drive off highway, read stoplights, advanced decision making ie confusing traffic situations. Is it a perfect system- definitely not, who knows if we’ll get to fully autonomous driving in our lifetime, but I would argue the past 10-15 years have had far more innovation of vehicles than earlier decades combined. If you look back at my posts, going back years, I have been highly critical of Tesla and Musk- long before it was fashionable. But that doesn’t mean I can’t recognize their impacts on the industry.

GM SuperCruise I actually think is a more practical, safer, and sensible (based on current capabilities of all systems). I also think it’s a very clever solution to the problem at hand. SC of course only works on pre-mapped highways. If hypothetically I was choosing between SC and FSD, it would be SC. I’m a lot more comfortable with GM’s system than the much more unpredictable Tesla FSD.

Ford BlueCruise I’ve never used. I believe it’s a similar concept to SC. I’m not aware of any other brands with automation at this level of sophistication. Assuming actual full self driving is feasible and practical, ie 10-20 years from now, at some point there probably won’t be much difference in capabilities between automakers.
I am excited to see what GM does for their version of Supercruise for City Streets. Especially if they can do an hands off version. I'm not sure if Ford has talked about anything more complex than their current highway stack.
 
For me I’d only interested in autonomous driving on motorways. But I do very little city driving tbh.

No system will work well on the single track country lanes I use when I drive.

Although FSDs is not perfect, there have been drastic improvements in the last year with the rebuild of their code (especially in the last 6 months since the 2024 holiday update). Autopilot is still a separate branch that is using human programmed code from 5+ years ago, with very little changes.

The newer 12.6/13 FSDs code has replaced both City and Highway code. If you haven’t had a look at some of the current abilities and have the time, there are great videos on youtube. It does some impressive things, including single lane country road (even unmarked) driving. It will yield, and go around objects (vehicles/people/objects). The old Autopilot software I’m sure would be TERRIBLE in your use case.

Unfortunately, FSDs isn’t available in the UK/Europe (as I understand it, right now).

I use FSDs every drive, for either part or my full drive in the Long Island (often single lane suburban roads because of cars parked on both sides) and Brooklyn NYC area.

It is not perfect. But the level where it is, is quite impressive.
 
Although FSDs is not perfect, there have been drastic improvements in the last year with the rebuild of their code (especially in the last 6 months since the 2024 holiday update). Autopilot is still a separate branch that is using human programmed code from 5+ years ago, with very little changes.

The newer 12.6/13 FSDs code has replaced both City and Highway code. If you haven’t had a look at some of the current abilities and have the time, there are great videos on youtube. It does some impressive things, including single lane country road (even unmarked) driving. It will yield, and go around objects (vehicles/people/objects). The old Autopilot software I’m sure would be TERRIBLE in your use case.

Unfortunately, FSDs isn’t available in the UK/Europe (as I understand it, right now).

I use FSDs every drive, for either part or my full drive in the Long Island (often single lane suburban roads because of cars parked on both sides) and Brooklyn NYC area.

It is not perfect. But the level where it is, is quite impressive.
I use adaptive cruise control on many of my trips, but never in the country lanes. It doesn’t do well on the bends!

Yes I believe it isn’t available in the UK. Not sure it will ever be tbh.
I don’t think the UK Government wants to help Musk in anyway given some of his comments about our country and how it’s run!
 
I also do see the addition of LiDAR in Tesla’s system, at some point. Once it can be integrated in a cost effective way (Cost, Longevity, aerodynamic efficiency, and Programming), I see no current reason they wouldn’t. Now, if there is a breakthrough in camera tech, since cameras are not a static target, I also could see them sticking with vision.

Currently the front cameras all have washers (CT/TMY juniper) and wipers (all models). I rarely have issues where the cameras are obstructed, but it does happen. For example, this week in the rain, I did get an alert that the drivers side B-pillar camera washer obstructed. The redundant overlap with the front wide view and side repeater did continue to allow FSDs to work, but it was something that alerted me to be read for.

I will say that the cameras right now, even on my older HW3 cameras, sees things that I don’t. I was at a light in Brooklyn and the visualization showed someone walking across the street in an unlit area. I was looking out my window and couldn’t seem them, but the visualization showed them walking. They eventually walked closer to a street light, so I actually saw them.
 
Yes… but there’s a difference between cameras for lower level automated aids and using a solely based camera system to operate all autonomous driving features. Whereas other vehicles are also using LiDAR, Radar, Ultrasonic. Maybe thermal/FLIR will find its way into cars automated driving aids, still too $$$ right now (and often export controlled). They all have pros and cons but using a combination of systems give you better reliability, redundancy, and I would assume, to a degree, a way to verify data. To rely solely on CCD camera sensors seems like putting all your eggs in one basket- and we all know the cameras can have a lot of limitations in certain scenarios.

Tesla is the only brand I’m aware of that solely relies on optical cameras for higher level autonomous driving. No radars. No ultrasonic. No Lidar.

This is exactly my issue with Tesla. I’m not going to high-five how good a camera-based system is for getting it “mostly right” given they don’t use the very sensors that can fix most if not all of its shortcomings. This is why I respect other brands - they don’t release driving aids until they work. And they certainly don’t over-promise nearly to the extent Tesla does.

So far Tesla is the ONLY brand I have ever heard of phantom-breaking due to cars approaching in an oncoming lane. I said our business has these as work vehicles and even my coworkers have commented on this. Mercedes has offered adaptive cruise control in 1999 and emergency brake in 2006, and you could get it in a 3 series as early as 2016. Modern BMW’s (their new architecture could have adapted to something new) and I believe even Audi uses 3 radars to create 3D-mapping and better distance validation. In my opinion if that’s what it takes to create driver assistance systems that work as they should, then so be it.


I think it’s pretty unfair to say Tesla’s FSD is not impressive. The display of vehicles on the LCD has nothing to do with it (I actually find that very distracting). Keep it mind it want that long ago that your new car’s biggest tech advancement was an Aux Jack.

I doubt you’d deny Tesla has not been a leader with advanced automated aids. FSD is the only system I’m aware of currently that can drive off highway, read stoplights, advanced decision making ie confusing traffic situations. Is it a perfect system- definitely not,

I am not impressed by Tesla FSD. I would absolutely say Tesla is not a leader in advanced aids. Their aids are, metaphorically speaking, promised to be worth $100 and work effortlessly with no issue. And then they have these inexcusable and common hiccups and feel like a $60 system. If it makes hiccups, I can’t trust the system - the illusion is killed. A system I can’t trust is worthless.


GM SuperCruise I actually think is a more practical, safer, and sensible (based on current capabilities of all systems). I also think it’s a very clever solution to the problem at hand. SC of course only works on pre-mapped highways. If hypothetically I was choosing between SC and FSD, it would be SC. I’m a lot more comfortable with GM’s system than the much more unpredictable Tesla FSD.

LOVE GM supercruise. They promised a system that could do X, and it does X extremely well. They didn’t risk allowing it in areas they knew it couldn’t do X well, and I respect them for it. The system is genuinely really good.


None of these systems are anywhere ready for full independence from direct human oversight- find me any vehicle with auto emergency braking that hasn’t phantom braked at some
point (usually minor, but even a momentary brake check can be dangerous). In my experience, if you spend enough time in a car, you’ll have it happen. Tesla’s issue I would suggest has to do with their sole use of cameras.
Again, that’s my issue with Tesla. Get the sensors to make it better than “good,” and stop letting the CEO parade for nearly (over?) a decade how little humans are needed for their systems to work when they’re not that good.

Tesla’s lack of lidar sensor distance and space validation I think is a key reason why it has phantom braking. When it can’t validate if an oncoming car is in another lane, it has to. Lidar would fix most of these cases.
 
I use adaptive cruise control on many of my trips, but never in the country lanes. It doesn’t do well on the bends!

Yes I believe it isn’t available in the UK. Not sure it will ever be tbh.
I don’t think the UK Government wants to help Musk in anyway given some of his comments about our country and how it’s run!

Yeah, adaptive cruise control really limits one to highways. Makes sense to not use it in the country.

It is ashamed, his mouth/wallet really are getting in the way of very useful features getting to people who have spend a TON of money and never had access yet. There are FSDs buyers in the UK that are still waiting.

I would like to see the FSDs highway/city code that exists today replace Autopilot. It would simply replace the AP code, still requiring FSDs purchase and regulatory approval to use the FSDs based features, but at least regular owners would be able to use a much safer and WAY better “AP.”

I look at my 87 year old parents who live alone in the country (in Florida), their quality of life would be greatly improved if they had true autonomous driving. Things like going to the grocery store, doctor’s offices, and even just to get out and go to the park to get some exercise would be great.

With the improvements in FSDs in the 3 year’s I’ve had my Teslas, I definitely see a light at the end of the tunnel now. I know by the time I am their age, it will be the norm.
 
It seems he may have since deleted the tweet since I originally mentioned it in this thread, but Elon still firmly stands that Tesla does not need lidar:


Should have taken a screenshot. But as long as he’s at the helm, it appears he carries this position for Tesla.

I did see the post in the past.

I think from a cost perspective, which includes hardware, efficiency, programming complexity at the time, that was an appropriate stance. Even now I don’t think we are there yet, but in the future as hardware gets smaller and cheaper it would be appropriate to change.

But, just like the tech behind LiDAR is improving, so are cameras. So if cameras improve at the same relative speed as LiDAR, I also can see an argument for sticking with camera only.
 
I did see the post in the past.

I think from a cost perspective, which includes hardware, efficiency, programming complexity at the time, that was an appropriate stance. Even now I don’t think we are there yet, but in the future as hardware gets smaller and cheaper it would be appropriate to change.

But, just like the tech behind LiDAR is improving, so are cameras. So if cameras improve at the same relative speed as LiDAR, I also can see an argument for sticking with camera only.

I can’t see videos like this:

And go “Tesla should’nt invest a cost into lidar, they can improve the camera.” The output of a camera has to be processed to interpret objects. Lidar is essentially a distance map. Although this is one case, this is one of many instances a Tesla autopilot has either fatally killed the driver or more due to an ignorant stance, almost certainly monetarily driven, to not use lidar. Situations like this is where lidar excels.

So what, these situations where the AI can’t make a determination on its own, that kills someone, is just the necessary cost of doing business? Not for me.

I found about a dozen other situations where lidar would have protected a fatal crash, before finding this specific example. Because this clip is a great example where cameras will almost certainly not be good enough. Tesla seems to be the only “forward thinking company” that has such a resistance to the proven technology.
 
I can’t see videos like this:

And go “Tesla should’nt invest a cost into lidar, they can improve the camera.” The output of a camera has to be processed to interpret objects. Lidar is essentially a distance map. Although this is one case, this is one of many instances a Tesla autopilot has either fatally killed the driver or more due to an ignorant stance, almost certainly monetarily driven, to not use lidar. Situations like this is where lidar excels.

So what, these situations where the AI can’t make a determination on its own, that kills someone, is just the necessary cost of doing business? Not for me.

I found about a dozen other situations where lidar would have protected a fatal crash, before finding this specific example. Because this clip is a great example where cameras will almost certainly not be good enough. Tesla seems to be the only “forward thinking company” that has such a resistance to the proven technology.
When Tesla's had radar folks complained about phantom braking, and cars decapitating their drivers due to cross traffic being ignored (do other adapter cruise control systems react to cross traffic on surface streets?). Maybe lidar would help, but at the time their CEO thought it was too expensive to be added into every vehicle they make. As far as I can tell most auto manufactures don't include lidar on every model usually just the really expensive ones. And at this point I assume they are being difficult about it as a point of pride. Interestingly as far as I can tell only the CEO talks about how vision is as good as other sensors, so my assumption is if he ever dies or steps down more sensors would probably be added back in. Especially now that the code being ran is based on NN and not hand/hard coded like the earlier iterations.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.