Real world Tesla camera based safety & self driving tech vs LiDAR

[ChrisO]

Tesla uses monocular depth estimation. There are a lot of tricks but they are more advanced than judging a vehicle's size based on how far the tail lights are apart. Every car is a different size

So when are you going to get your bulletproof vehicle with windows that you could hit with a sledgehammer, boat? You know to go with your auto pilot vehicle that it’s been in beta for over 10 years, but with him claim every year that “this is the year it will be ready”.

I don’t think you really get the point. I absolutely believe in the future of self-driving cars that will be much safer than the crazy people driving their own cars. I just really dislike dishonest people that that misrepresent what they are selling.

Sponsored

 

[tobiasjef]

So when are you gonna get your bulletproof vehicle withwindows that you could hit with a sledgehammer, boat? You know to go with your auto pilot vehicle that it’s been in beta for over 10 years, but with him claim every year that “this is the year it will be ready”.

I don’t think you really get the point. I absolutely believe in the future of self driving cars that will be much safer than the crazy people driving their own cars. I just really dislike dishonest people that that misrepresent what they are selling.

elon never made any claims about self driving being "this year" for the last 4 years. We robot event was the first time they made that claim in years

 

[tobiasjef]

It is amazing that you can not even realize why Tesla can not meet the California and Neveda requirements for level #3 autonomy.

It's amazing you understand zero about the SAE levels and why they are not used by self driving companies

 

[Billyk24]

It's amazing you understand zero about the SAE levels and why they are not used by self driving companies

It is amazing you can't see the trees in the forest.

 

[DevSecOps]

The last thing I'm worried about any car doing is crashing into a fake styrofoam painted wall.

Stationary objects are a problem for almost all non lidar systems, including the MME. Every car I've ever owned would likely crash into that wall or a fake person standing still.

I love Mark's videos and he owns a Tesla. I think people on both sides are looking at this with some level of derangement rather than what Mark is known for, science.

 

[Jimrpa]

Endanger the public?



Drive like this. I dare you. I bet you can't drive this well if you tried.

Ok, I have to say that, assuming this video is real (and I have no reason to believe it’s not), this is pretty impressive. I would still prefer a system that relies on multi-domain sensors (LiDAR, stereoscopic optical, radar, ultrasonics, etc) and data fusion.

 

[Rusky]

My only question Is how many human drivers would that wall have fake out? I know I didn't come equipped with lidar!

 

[Dzrhino]

My daughter has a 2021 Tesla 3 that she paid the extra $10K for at the time to get "Full Self-Driving". Hers has the radar installed, but when it went in for service in 2023, they turned it off and told her that the radar was no longer supported. She noticed after that her car doesn't track as well as it used to in fog or rain, weaves more on cruise, and she is still waiting for FSD to be delivered.

 

[Billyk24]

My daughter has a 2021 Tesla 3 that she paid the extra $10K for at the time to get "Full Self-Driving". Hers has the radar installed, but when it went in for service in 2023, they turned it off and told her that the radar was no longer supported. She noticed after that her car doesn't track as well as it used to in fog or rain, weaves more on cruise, and she is still waiting for FSD to be delivered.

Likely she will never get it as she might have hardware #3. Current vehicles have hardware #4. Is it true that the newest hardware #4 vehicles have radar installed as read in a recent report? Which might explain why the CT stopped before the fake wall which others failed?

 

[ChehRob]

If I remember correctly, Volvo and Mercedes pioneered many of the safety features which are universal now. But more importantly, saw to it that they were quickly shared with the rest of the automobile world.

The original lidar for cars was this Rube Goldburg ugly device sitting on top of the car, and costing more than the car itself. Now lidars are well under $200, and I am hearing maybe half that this year.

My question is why does every car maker have to reinvent all of the ADAS improvements which now MUST include at least one lidar. BYD an apparent leader, or Europe, or the US or any combination of them should have the 'operating system' that incorporates major improvements in ADAS. The goal is to make highway and street transportation safer, save lives, reduce those so very expensive collisions.

I am hoping that Consumer Reports start including in their ratings which car and system most likely will keep me from having an accident. I usually drive a car 4-10 years. But I will replace my MME as soon as something comes along which has significantly better ADAS.

 

[rrobe53]

The last thing I'm worried about any car doing is crashing into a fake styrofoam painted wall.

Stationary objects are a problem for almost all non lidar systems, including the MME. Every car I've ever owned would likely crash into that wall or a fake person standing still.

I love Mark's videos and he owns a Tesla. I think people on both sides are looking at this with some level of derangement rather than what Mark is known for, science.

To me, the point is that cameras aren't as good as LIDAR, which Musk has regularly spouted as true.

This isn't the first time it's been an issue for Tesla either

 

[DevSecOps]

To me, the point is that cameras aren't as good as LIDAR, which Musk has regularly spouted as true.

This isn't the first time it's been an issue for Tesla either

In my comment I suggested that you look at the video for science purposes, which is what Mark Rober is known for. His lead in where he talked about space mountain is a perfect example of how our eyes (or cameras) can't see things that lidar can. I don't disagree that lidar has advantages, but so does vision. It's not a one stop shop. Cameras and vision are known to be better at object recognition and cost effectiveness, which is why all ADAS systems that have lidar also have cameras. Just look at the self driving cars, they have a myriad of components including lidar, cameras and radar. If the goal is to get mass adoption of electric vehicles, lidar is definately not the starting point for entry point cost.

By bringing up Elon, you prove my point. This isn't about the science for you, it's about something else. For every crash video you post there's 10x those videos where Tesla vision saves lives.

 

[Billyk24]

Few mention or remember that Tesla purchased over $2 million in LiDAR in 2024.--see link: Tesla bought over $2 million worth of lidar sensors from Luminar this year | The Verge I am assuming they have install such on their newest models with hardware #4 as a SAE requirement is to have a backup sensory input.

 

[rrobe53]

By bringing up Elon, you prove my point. This isn't about the science for you, it's about something else. For every crash video you post there's 10x those videos where Tesla vision saves lives.

"I DECLARE MYSELF VICTORIOUS" level argument.

Musk is the CEO of Tesla who has directly purported that cameras are better than LIDAR, and Mark conducted an experiment to show that it is not. It's literally scientific inquiry, testing a hypothesis.

I don't think LIDAR is an end all be all either, but that's not the argument being presented or hypothesis being tested.

 

[stoopid]

Here's another video whose conclusions are similar to Rober's.



There's also a few... hundred or thousand, or hundred thousand(?) dashcam videos of self driving cars making egregious errors.

Think what you want about humans, we are very complex creatures and replicating that with silicon is a tremendously far off future.

Sponsored