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

[AtomicInternet]

I don't own a Tesla or plan to, but the idea of independent testing is the tester pays for everything, and nobody gets their name advertised during the results.

If we're just taking face value from everyone, why not just read the Tesla statement from above?

https://www.tesla.com/support/transitioning-tesla-vision

Sponsored

 

[SpaceEVDriver]

As a remote sensing scientist whose work is currently focused on visible light 3D mapping (multi-view "stereo") related to safety for both hardware and human life, I don't trust any safety system based solely on one sensor system, especially visible light cameras. They're just not up to the task. I want LiDAR (preferably in multiple wavelengths), RADAR (multiple wavelengths as well), and multi-view 3D vision.

 

[CarlSagan82]

An article from today which I believe is related to this topic:
https://arstechnica.com/cars/2025/0...r-other-automakers-a-free-pass-advocates-say/
Consumer advocates and the families of Tesla crash victims both called on Secretary of Transportation Sean Duffy this week not to weaken oversight of the automotive industry. As with other government agencies, since the change of administrations in January, much of the messaging from the Department of Transportation has been on overturning regulations on industry and weakening oversight and consumer protection to create a "level... playing field."

On Tuesday, seven family members of people killed or seriously injured as a result of Tesla's Autopilot driver assist wrote to Sec. Duffy with a particular concern: that he not overturn the National Highway Traffic Safety Administration's Standing General Order that requires automakers or operators to report vehicle crashes that occurred while using a partially automated driving system (like Tesla Autopilot or FSD) or an autonomous driving system (like Waymo's robotaxis).

In particular, the families are concerned that Duffy will give special treatment to Tesla's Elon Musk, who these days is busier poking his fingers into all corners of the federal government than he is running Tesla or SpaceX.

 

[Sikkun]

Logically if the car is about to have a collision auto pilot should probably disengage so the vehicle doesn’t attempt to keep driving after the crash?

But also, no surprise Lidar doesn’t fail at the fake wall test. Everyone also knows Tesla failing the fake wall test doesn’t matter. So really the results were ⅗ vs 5/5. Some often the others Teslas newer stuff might have done better but people’s fanboyism took over their emotions and missed the entire point of the video.

Using lidar to create a scale replica of Space Mountain.

(It was also his personally owned Tesla…don’t think he did a hit piece against his own car)

 

[AllTheThings55]

Logically if the car is about to have a collision auto pilot should probably disengage so the vehicle doesn’t attempt to keep driving after the crash?

But also, no surprise Lidar doesn’t fail at the fake wall test. Everyone also knows Tesla failing the fake wall test doesn’t matter. So really the results were ⅗ vs 5/5. Some often the others Teslas newer stuff might have done better but people’s fanboyism took over their emotions and missed the entire point of the video.

Using lidar to create a scale replica of Space Mountain.

(It was also his personally owned Tesla…don’t think he did a hit piece against his own car)

For me at least

Logically if the car is about to have a collision auto pilot should probably disengage so the vehicle doesn’t attempt to keep driving after the crash?

But also, no surprise Lidar doesn’t fail at the fake wall test. Everyone also knows Tesla failing the fake wall test doesn’t matter. So really the results were ⅗ vs 5/5. Some often the others Teslas newer stuff might have done better but people’s fanboyism took over their emotions and missed the entire point of the video.

Using lidar to create a scale replica of Space Mountain.

(It was also his personally owned Tesla…don’t think he did a hit piece against his own car)

I don’t think it was a hit piece, but he sure didn’t help things by not disclosing that the company that provided the lidar test vehicle is owned by his friend who also donated millions of dollars to Marks charity.
This same company also had a direct link to the YouTube video on their website…..really bad optics regardless of intent.

 

[moog]

I don't own a Tesla or plan to, but the idea of independent testing is the tester pays for everything, and nobody gets their name advertised during the results.

If we're just taking face value from everyone, why not just read the Tesla statement from above?

https://www.tesla.com/support/transitioning-tesla-vision

The first sentence alone lets you know the rest of the statement is going to be a lot of BS.

Always a "brilliant" idea to rely on one type of sensor system versus multiple different types of sensor systems.

Every see the complaints about the camera based auto wipers for Tesla?
Yea, Tesla is too smart for a rain sensor that 99.9% of automakers use...and is proven to work.

I'm all for pushing technology forward, but when it becomes dangerous, and it's clearly done for cost cutting.......you know.....don't pee on my leg and tell me it's raining.

Because we have all seen the cybertruck, and we know Tesla's recent MO is cost cutting galore.

 

[Jimrpa]

"Respected YouTuber with years of science based content makes video of car's safety features failing in unusual circumstances and discusses the science behind it. This YouTuber, having never had any political content on his channel, is obviously part of a large political conspiracy to undermine Elon Musk. More RealNews at 11"

Well, that sounds like my Newsmax (fox is too left-wing) watching friend ?

 

[DeltaDelta]

For me at least

I don’t think it was a hit piece, but he sure didn’t help things by not disclosing that the company that provided the lidar test vehicle is owned by his friend who also donated millions of dollars to Marks charity.
This same company also had a direct link to the YouTube video on their website…..really bad optics regardless of intent.

Yeah, I agree with that.

It's also a very contrived scenario. (Though one the car absolutely should not fail).

 

[Jimrpa]

I’m reminded of videos I saw, many years ago, when Ford was first developing/testing their adaptive cruise control and ADAS. The test cars were Lincolns. Obviously, Ford didn’t want to spend a fortune on test vehicles that were going to be run into a lot, so they developed “inflatable Lincolns” to use as “targets” for both following and emergency braking. The early emergency braking tests were really funny as the car with the test system would plow into the target “vehicle” at speed and, instead of crumpled metal and broken glass everywhere, the target would just sort of bounce down the track.

it WOULD be interesting to try a Mustang Mach E on BlueCruise against the wall (that has styrofoam blocks behind it). Would the styrofoam blocks be dense enough for the ADAS radar to detect them and brake the vehicle?

 

[tobiasjef]

There's a lot of troubling things about mark rober's video.

he claims he did not use full self driving because he believed an address was needed for full self driving to work. That is incorrect and you can just turn on full self driving anywhere and it drives straight. If you want to make the car drive somewhere specific, no address is needed as you can click on a spot on the map and drive there.

Secondly, tesla is driving over the centerline to hit a child in the rain. How do you engage autopilot while driving over the centerline? You can't

Autopilot or FSD does not just disengage without beeps or warnings. The reason autopilot disengaged is because mark rober tugged on the steering wheel. Conveniently both times he did the wall test, autopilot disengaged exactly when he tugs on the steering wheel.

Tesla Full Self driving absolutely sees through heavy rain and fog. And when it can't see, it goes the speed it needs to safely drive with the visibility it has. No question that child is seen in the rain test.

 

[ChrisO]

Here is my take on it. The first point is "what are you telling the public". This is as much part of "safety" than the actual technology. If you give everyone the impression that the car "can drive itself", but then turn around and play down the fact that they need to be able to be ready to take over at any moment, and it the car does something bad and hit someone, it the driver's fault not the company that makes the car, then that isn't "safe".

The second part is about what I would envision the future to be. I don't want a self-driving car that is "as good as humans", I want one that is better. The "wall" didn't impress me at all, it is not going to happen in the real world. What did, is the difference in the tests in low visibility.

Yes, the Mach-E doesn't have LIDAR, but like any expensive feature they usually start in the most premium of cars and as they get adopted and cost less the features migrate down to the cheaper cars. That's how we get such things to be common.

Some of the Tesla are over $100,000 but somehow, he can't afford to put in such technology?
When you listen to him you quickly realize that he just has a chip on his shoulder and says, "if a human can do it with just vision, then AI and cameras should be able to do it." ignoring the fact that you can't compare apples and oranges. Humans and cameras have their own set of limitation and strengths. Remember that he has even removed cheap radar sensors.

And here is an illustration of how he thinks. Early on, he was using auto pilot to drive into work. There was one stretch of the road with a curve where the markings were missing, and the car would always mess up in that area. He would tell the engineers that it needed to be fixed. And they would tell him they needed better sensors. His take was "a human can do it, make it work". So, how did it get "fixed"? The manager went to the city and got the line properly painted. Of course, the car now "did the right thing". And as far as he was concerned "problem fixed".

 

[superdave80]

Logically if the car is about to have a collision auto pilot should probably disengage so the vehicle doesn’t attempt to keep driving after the crash?

Logically auto pilot should hit the brakes if it is about to have a collision. I mean, how concerned are you about auto pilot function AFTER you crash?

 

[Mach-Lee]

Tesla Full Self driving absolutely sees through heavy rain and fog. And when it can't see, it goes the speed it needs to safely drive with the visibility it has. No question that child is seen in the rain test.

I’m not so sure about that. Automatic emergency braking is supposed to work regardless of the Autopilot state. If you drive a Mach-E into a person it will try to stop even if cruise control isn’t engaged. That should have been the test, no Autopilot or FSD, just hand driving to see if it would stop. But he tried to be more fair by using Autopilot in hopes it would be more sensitive to objects.

 

[Billyk24]

There's a lot of troubling things about mark rober's video.

he claims he did not use full self driving because he believed an address was needed for full self driving to work. That is incorrect and you can just turn on full self driving anywhere and it drives straight. If you want to make the car drive somewhere specific, no address is needed as you can click on a spot on the map and drive there.

Secondly, tesla is driving over the centerline to hit a child in the rain. How do you engage autopilot while driving over the centerline? You can't

Autopilot or FSD does not just disengage without beeps or warnings. The reason autopilot disengaged is because mark rober tugged on the steering wheel. Conveniently both times he did the wall test, autopilot disengaged exactly when he tugs on the steering wheel.

Tesla Full Self driving absolutely sees through heavy rain and fog. And when it can't see, it goes the speed it needs to safely drive with the visibility it has. No question that child is seen in the rain test.

Tesla FSD Fails in Heavy Rain: Stormy Weather Test At Night a You tube video of FSD in heavy rain.

 

[SpaceEVDriver]

The "wall" didn't impress me at all, it is not going to happen in the real world.

I mostly agree with you, but this is inaccurate. Many ASAD systems have (in)famously failed to recognize such things as "walls" whether they're other vehicles like tractor trailers or "actual" walls. People are accepting at face-value the "(full) self driving" claims and letting the vehicle take over when they shouldn't. It's a problem caused by both manufacturer and driver.

A system with "stereo vision" uses automated feature matching to determine the depth/distance to objects within the scene. Automated feature matching fails dramatically when it's presented with a flat, monotone, or otherwise featureless object. A brick wall that fills most of the field of view will look like nothing--like a blank scene; there will be no depth/distance information in that scene. Any 3D features that are in a wall are too small to match at the resolutions and speeds required for safety while driving.

Automated matching also fails when there are confusing signals such as flashing lights or repeated similar or identical objects. This is especially true when those flashing lights alternate colors or positions (safety vehicles). A flashing blue/red light in one scene that's either off or in a different place in the next scene can cause very weird distance/movement results even though the lights are not moving. Confusing results can also happen with a series of identical traffic control devices.

A poorly-designed safety system will not respond well to such scenes. An overly-permissive system might decide it's a glitch and continue driving. An overly-responsive system might slam on the brakes on a freeway when the vehicle is heading into the sun b/c the scene is washed out and the system decides it might be seeing a wall. A cautious system might yell at the driver to take over, but if they weren't paying attention, it may be too late. Figuring all of this out is not an easy task to program.

A system with two or more sensor types and a diversity of data would be able to cross reference the data and respond more appropriately.

Sponsored