Tesla has stopped selling vehicles with electronic radar in the US, and Musk has been refusing to use laser radar for years.

[Exscite]

Elon Musk's crusade against LiDAR and radar
"Tesla recorded another milestone in Elon Musk's war on automobile radars [in October 2021]. [Tesla] announced that it will stop equipping the saloon Model S and leisure Model X with electronic radars, a move it already made a year ago with its smaller pair of electric vehicles, Model 3 and Model Y. The move is currently limited to cars that Tesla sells in the U.S., and comes after years in which Musk claimed that laser radar, LiDAR, is unnecessary too, even though other car manufacturers are actually adopting the technology for models with partially autonomous capabilities.
"Humans drive through the eyes that see, and a brain made up of biological neural networks that analyze the information. There's no reason an autonomous vehicle would not work the same way, with silicone neural cameras and networks processing the information," Musk wrote in October 2021. "
(Source: Calcalist 23 Mar, 2022)
This article is for the technology geeks among us.
autonomous vehicle will need backup for its cameras. A set of cameras and electronic radars and separate laser radars, with each providing its own roadmap that the autonomous driving system will analyze, and in case of conflict will decide which of them to give priority, depending on a priority set in advance.
The problem with additional sensors is their high costs, especially of the LiDARs.
Read more

 

[Brando]

When a camera and a radar conflict -- which one do you trust? This is another reason having different sensors might not help and only confuse.

 

[jeremymc7]

No reason that the cars can’t have better optical (and/or digital) vision than a human.

Sure humans drive with sight. So can a car. But why not provide better than human capabilities.

The neural network seems to be the biggest problem at the moment. Not so much the much the source of visual information.

That said the car needs more (and better) cameras.

 

[israndy]

Would love to know what the original poster was thinking when he posted this article. The headline makes it seem like another Elon is an Asshat post.

The point is Elon has a plan and if he is right, he will have the most successful autonomous driving company, cause the other guys will have to write software for their LiDAR and can't just license Tesla's vision only AP.

I can see that Tesla MIGHT learn that it would have been smarter to add side cameras AHEAD of the driver for those situations where a car or truck is blocking the view of oncoming traffic from the B pillar, but it may be that the A.I. stack figures out a better way to drive with the cameras it has than we do with our two eyes and ears only, even if it seems weird to us. Time will tell.

 

[jabloomf1230]

The neural network seems to be the biggest problem at the moment. Not so much the much the source of visual information.

That said the car needs more (and better) cameras.

^
This.

Cameras, radar and LIDAR all have benefits and drawbacks. The factors that presently limit the capabilities of automotive driving systems are the quality, quantity and placement of sensors combined with the abilities of the NN system.

 

[Brando]

I think digital cameras can sense IR light. We shall see if Tesla can/needs/wants to use that data.

 

[Tha_Ape]

When a camera and a radar conflict -- which one do you trust? This is another reason having different sensors might not help and only confuse.

you assign a confidence factor to each measurement and go off the higher one.

foggy: camera says nothing there, low confidence. radar says object in road, high confidence. apply brakes.

busy city street clear day: radar picks up a lot of scatter and inconsistent readings, low confidence. Vision system clearly id's objects, high confidence.

This isn't hard. I guarantee you SpaceX doesn't only use a single sensor when approaching the ISS for docking. likely uses LIDAR, IR, and visual sensors. Which get cycled depending on range.

 

[a.deodatus]

I'm hoping that you will be able to use autopilot at speeds greater than 80 MPH. That issue alone was enough for me to sell my 2022 Model Y Performance, as the speed limit where I drive a lot of my work day is 75 MPH, and going 80 MPH will leave you in the slow lane with people tailgating you!

In comparison, my 2019 Model S LR Raven can do 90 MPH with autopilot. Far more reasonable!

 

[jeremymc7]

Is that because they haven’t removed / shutoff radar on S/X yet ?

 

[israndy]

This isn't hard. I guarantee you SpaceX doesn't only use a single sensor when approaching the ISS for docking. likely uses LIDAR, IR, and visual sensors.

Yeah, but they don't produce a million rockets a year that dock in space, they don't care about the aerodynamics of the module as they dock, and you can't get out and walk if you have trouble docking to the space station. This is a poor example.

 

[Tha_Ape]

Yeah, but they don't produce a million rockets a year that dock in space, they don't care about the aerodynamics of the module as they dock, and you can't get out and walk if you have trouble docking to the space station. This is a poor example.

this is an example that sensor fusion is common in autonomous systems and that they don't rely on a single sensor.

Radar costs $100? removing it is because of supply constraints, not because of sensor fusion difficulties or that they can get "better" results from a single system.

I used SpaceX as an example to show Musk is well aware of this

 

[Brando]

this is an example that sensor fusion is common in autonomous systems and that they don't rely on a single sensor.

Radar costs $100? removing it is because of supply constraints, not because of sensor fusion difficulties or that they can get "better" results from a single system.

I used SpaceX as an example to show Musk is well aware of this

Elon Musk claims his use of LiDAR at SpaceX for docking showed him why it wouldn't help a self driving car.

Aviation has redundancy and some triple redundant systems.
Consider the 737 MAX with two crashes (2018), killing about 300. Redundancy didn't help.

 

[israndy]

Well, if all your systems are programed the same way, they will be in agreement that it has to do whatever it's been instructed to do. It's when one system FAILS to do what it's been instructed to do that redundancy come into play.

 

[Earl]

I suspect that Tesla is using/has used both and compared the results over millions of miles driven so they're likely to know more than pretty much anyone, including other collision avoidance system developers, just do to their potential sample size.
Personally, I'm skeptical that any active collision avoidance system (one that counts on receiving Radio, light, or sound that it emits) will scale well when there are literally thousands of them in close proximity as one will find on urban freeway choke points.
Passive systems such as cameras that can use self-generated (headlights), 3rd party (sunlight, street lights, other headlights, etc) emissions with their own sensors have huge advantages when it comes to high capacity systems.
I haven't done any specific analysis or studies, however.