Musk was wrong... aesthetics aren't everything

Musk is a nazi moron

There's more to ADS safety stats than sensors diversity, so I wouldn't treat it as a foregone conclusion that a camera-only Tesla ADS will be less safe than a Waymo in the next couple years, in terms of involvement in crashes with injuries/fatalities per mile driven. Waymos are still involved in plenty of crashes, with many of them avoidable with different decisions on Waymo's part. And cam-only may be cheaper to produce (unclear because it may also require more expensive processing power or energy storage for similar performance). But I agree with you on aesthetics. The number and location of sensors might be harder to clean and maintain, but they look just fine to me. I'd place no value on having the ADS I'm in look like non-ADS vehicles.

Wouldn't additional sensors also provide data that a camera only system would need additional resources to calculate while providing them faster and more accurately?

Maybe if Waymo adds another dozen Lidar sensors it will suddenly become an intelligent system that can operate in a blackout?

Musk is always wrong

Yes. So provide that extra compute power. Computers have been getting faster and using less power every year for over 50 years by now.

Headline is misleading. Leon is never right.

No. Waymo uses 4000w of compute. Tesla uses 150 Waymo does everything tesla does + what they use the additional sensors for. Additional sensors requires more compute, more memory and more memory bandwidth Waymo has 100 miles of usable city range. Tesla has 350+ Waymo's argument is that they will improve this over time including reducing the costs of the car. It's about delivering the minimum driving to start with and improving ODD, reducing cost, etc. later Tesla is about delivering the maximum ODD to start while at a low cost. It's different problems As far as more sensors it would certainly matter in the future how aesthetic the car looks. It's looking more and more likely that lidar is useless and cars will at most use HD radar in addition to cameras

They got too many false positives that you had to act on and the cost to collect the additional data and program for it meant they would be even longer to get 'something' to market. But ya, if you want better than human level driving, particularly when it is difficult to match human intuition, you want senses that exceed that of humans. Tesla is not years behind multiple other companies.

As an engineer I share your sentiment. What bothers me on top of just camera approach is lack of redundancy. In my view, all tesla does good is (was) marketing.

As another engineer I was amazed and the genius level rationalisation for not having two types of sensors… to paraphrase “when they disagree which one do you trust?” The one which won’t get you killed you halfwit!

Usually because he's lying

Lidar sure isn’t useless so far.  Every company to give public rides without a driver/employee uses LiDAR. Yes, Tesla is close to breaking that trend, but that hardly proves LiDAR useless. You seem to be assuming what the next few years look like in this space while a lot of people assume the opposite .  Maybe you’re right, but I’m not seeing clear evidence yet.

Something all engineers these days need to learn: Form should never take precedence over function!

He can't even keep up with a broken clock which is still right twice a day.

\> marketing Lying! [https://www.youtube.com/watch?v=9h9wStdPkQY](https://www.youtube.com/watch?v=9h9wStdPkQY)

Isn't Musk the guy that said that waymo is going to fail because of all the extra compute overhead they need including compute that needs to manage conflicting sensor data when say the cameras and lidar give different and opposing info which needs to be sorted in realtime? If Musk was so into aesthetics then he wouldn't let the designs be 10+ years old like they are with the Model S. I think you have that backwards. Aesthetics takes a back seat to the ability to mass produce the a viable product cheaply and efficiently. He has also said anyone can make concept car but making the production of it successfully is the real game.

Indeed, some (most) of Musk comments are downright hilarious. The problem of "sensors disagreeing" was solved in avionics decades ago. ;)

Of course, LIDAR literally means "light detection and ranging". It provides exact measurements of the surroundings without (or with much less) need for computer power compared to processing 2D camera data with "neuron networks" (whatever that might be). Not a rocket science, really. ;)

False positives are a wekl documented and significant issues for Tesla's camera-only system, leading to incidents like "phantom braking". This is undoubtedly due to cameras being easily fooled by shadows, reflections, plastic bags, leaves and the what not. Sensor fusion would easily provide cross-validation, not vice versa. Bottom line: Tesla "camera only" is a marketing gimmick not an engineering decision. Fact is that Tesla is behind the competition, basically only having driver assistance features and not true and/or reliable autonomy.

They are for stupid people, smart people know that form has to take a backseat to function.

To be fair, the waymo cars looks cool and approachable. If it looked like the Bing car from Family Guy, people would probably not want to ride it

But at least he usually has good intentions.

Every company who gives public rides without driver/employee does not use end to end neural networks. End to end neural networks require vision as a backbone. That's how these neural networks work Lidar is simply a red herring. The end to end neural networks are the real story

Tesla doesn't have phantom braking. In fact zoox has more phantom braking than tesla does Zoox has hard braking nearly every ride. Like emergency braking levels. You can't cross validate always because the sensors do not always return correct information. IN end to end neural network based systems I would bet any amount of money that lidar is purely used for emergency braking.

>Waymo uses 4000w of compute. Tesla uses 150 I think you're glossing over a minor detail: TSLA does NOT provide above SAE Level 2 autonomous driving. Using your login, my 2002 Hyundai uses zerow compute, so its better than Waymo too, right?

>Tesla doesn't have phantom braking These are not the droids you're looking for...

...or let the ahem: Safety Driver take over.

I'm still trying to wrap my head around the notion the "Cybercab" looks aesthetically pleasing?

Yes. Like an in-house chip and a flywheel using cameras LIDAR and radar. Like Rivian announced this month.

This has to be one of the craziest statements I've seen in a while. How do you think neural networks actually work? What do you think other companies are doing? Everyone is using neural networks and has sophisticated NN based vision stacks. They're just using additional sensor modalities on top of it to improve accuracy, redundancy and reliability of their systems. Neural networks don't give a shit where data comes from as long it produces a strong and actionable signal for them to use. Waymo itself literally uses NNs for both analyzing sensor input and actually planning out its driving actions. All this crap about "end-to-end" and the idea that no one else is able to do it is a completely fabrication being marketed by Tesla and Musk. Here's a quote from Waymo's own site: > In particular, the model leverages the full expressibility of learned embeddings as a rich interface between model components and supports full end-to-end signal backpropagation during training. Source: https://waymo.com/blog/2025/12/demonstrably-safe-ai-for-autonomous-driving You literally have no idea what you're talking about.

lol

Come on! Vision only is great! As long as its not rainy or sunny.. or snowy. And no weird shadows 🥰🥰

100% with you on this. I don't even entertain a theoretical scenario about this: I do not get into unsafe cars. Cars have ONE purpose - to get you safe from A to B. Tesla cannot do that, with or without Autopilot, and if we are realistic, their cars are not even autonomous and never will be. It is technically IMPOSSIBLE for them to reach a 100% fault free system with cameras only. Any glare, fog, snow, rain, shadow, LEAFS! ... and they freak out and break or try to 'dodge' it and get you killed. Hard pass! You look at a Waymo car and SEE what effort they engineered into that thing. On the other side of the spectrum: TESLA CANT GET DOOR HANDLES RIGHT! Tesla also cannot get WIPERS right... which are (not kidding!!) an older invention than electric light at home!

You don’t want a 100% fault free system before adoption. Better than a sober non distracted human should be the threshold regardless of what technology is used.

You people just wait until next year when all of Elmo's promises come true.

But radar and lidar provide important data not always available with cameras. Elon was afraid of lidar because of the expense. But they’ve since gone from 10s of thousands to hundreds of dollars. And the folks with lidar are winning. I had Elon’s windshield wipers and FSD for six years. Neither one works for shit. Tesla went from years ahead of the competition to years behind in only five years by making dipshit decisions like losing lidar and building that triangle thing no one wants. Maybe it’s the ketamine?

I dont disagree with anything you stated here. I disagree with your evaluation criteria. If cameras can achieve a lower accident and harm rate than people then it would be a win whether or not another option is safer. We need two or three companies to be successful successful in automation of cars to bring price down otherwise a monopoly will mean it will only be cheap enough to eliminate humans but no cheaper.

I generally agree. But in the long run I would be surprised if a significant difference in safety (if there is one) didn’t invite regulators to require whichever technical solutions prove safer and/or drive customers towards safer options. If we even have functional, efficient regulators in the future, definitely not a given.

This will occur well after initial adoption though but I do agree that when it becomes a very low cost adder Europe at a minimum and probably China will mandate it