LiDAR comments by RJ (from 10/6/25)

[ACDC]

In case you didn't listen TheVerge's podcast interview with RJ Scaringe:

Here's what RJ said on LiDAR:

“Yeah, so our view is that there is a real benefit to LiDAR. The view of the entirety of the science community is that having multiple sensors is helpful because you build a more accurate view of the world. The way that we build these neural nets is that you want a broad understanding of the world, and you want the highest accuracy. And if you have more than one camera, you’re going to have multiple cameras that have different signals, which have different signal-to-noise ratios that need to be managed. But ultimately, the way that that information is fused very early, if you have multiple cameras coupled with radar, coupled with potentially lidar, as you said, it gives you a more fulsome and accurate picture. It also allows you to train your model better.​

​

So, it’s analogous to if I had to learn the world with one eye, I would learn a less accurate version of the world than if I had learned the world with two eyes. And if you look at the evolutionary tracks of many species of animals, most animals have multiple modalities of sensing. And the ones that have to operate in maybe the most extreme environments, let’s say extreme darkness, generally combine some optical perception with some wavelength-based perception. Often, like sound waves or sonar, bats are an example of this.​

​

Our view is that it’s definitely beneficial, and our approach to sensors has been that we need to rapidly build our foundation model as fast as possible. Tesla has a lot of vehicles and has made great progress. We have an amazing product. So we have more megapixels in cameras. We have 55 megapixels in cameras in R1, which’ll jump to 65 megapixels in R2. We have a really robust set of corner radars and a really beautiful 3D imaging radar in the front. And that’s rapidly building a robust foundation model, one that we’re going to start to see these features I just described play with.”​

​

The interview asked if Rivian will be ruling out LiDAR for their vehicles, he responded:

​

“No, I wouldn’t rule out LiFAR”​

Sponsored

 

[Dave Cundiff]

RJ's language skills are impressive, but I think in this case he said "fulsome" when he meant "full."

"Fulsome" is a pretty complicated word, with multiple definitions, but when "fulsome" is used to describe abundance, it generally refers only to quantity, not to informativeness. Think "full picture" rather than "fulsome picture."

RJ's understanding of multiple modes of sensing, including radar, is pretty similar to mine. I trust him to design our next car as well as possible.

Thanks, @ACDC, for sharing! Best wishes!

 

[Scottski]

RJ's language skills are impressive, but I think in this case he said "fulsome" when he meant "full."

"Fulsome" is a pretty complicated word, with multiple definitions, but when "fulsome" is used to describe abundance, it generally refers only to quantity, not to informativeness. Think "full picture" rather than "fulsome picture."

RJ's understanding of multiple modes of sensing, including radar, is pretty similar to mine. I trust him to design our next car as well as possible.

Thanks, @ACDC, for sharing! Best wishes!

Dictionary
Definitions from Oxford Languages · Learn more

ful·some
/ˈfo͝olsəm/

adjective

1. 1.
   complimentary or flattering to an excessive degree.
   "they are almost embarrassingly fulsome in their appreciation"

 

[BCondrey]

RJ's language skills are impressive, but I think in this case he said "fulsome" when he meant "full."

"Fulsome" is a pretty complicated word, with multiple definitions, but when "fulsome" is used to describe abundance, it generally refers only to quantity, not to informativeness. Think "full picture" rather than "fulsome picture."

RJ's understanding of multiple modes of sensing, including radar, is pretty similar to mine. I trust him to design our next car as well as possible.

Thanks, @ACDC, for sharing! Best wishes!

I think we knew what he meant.

 

[nsuglerio]

After seeing FSD 14.1 driving down a gravel road and parking in a small garage on top of the other videos I would place bets that it can be done with vision only. Lidar is absolutely useful in ground truth training but not on the final vehicles.

 

[Mark_AZR1T]

LiDar is critical because it offers precise depth perception independent of lighting or weather, which is critical where cameras struggle. Tesla chose vision only at the time (mainly) because of cost and integration challenges with less capable sensors at the time. LiDar is absolutely the future when combined with vision systems.

 

[nsuglerio]

LiDar is critical because it offers precise depth perception independent of lighting or weather, which is critical where cameras struggle. Tesla chose vision only at the time (mainly) because of cost and integration challenges with less capable sensors at the time. LiDar is absolutely the future when combined with vision systems.

Well to be honest Stereo cameras do provide depth thanks to some trigonometry and luckily cars have headlights for when its dark. Not to mention rain and weathers effects on a laser beam. Seeing the latest FSD 14 videos and what it can do today in the hands of customers is absolutely impressive and knowing the system works on rural roads makes it even more impressive.

- I wish there was a betting site so we could all have some fun with our predictions

 

[portdirect]

Lidar is all but a done deal at this point, they have been hinting at it for ages, most notably during the Sandy Munro/Vidya Rajagopalan interview, where she stated that they did not include Lidar in the R1 gen2 due to cost. RJ now talking about the cost coming down... I'll say it again as shouting into the internet makes me happy - I've been expecting Lidar to be required for the Eyes off driving in 2026 ever since announced.

 

[Mark_AZR1T]

Well to be honest Stereo cameras do provide depth thanks to some trigonometry and luckily cars have headlights for when its dark. Not to mention rain and weathers effects on a laser beam. Seeing the latest FSD 14 videos and what it can do today in the hands of customers is absolutely impressive and knowing the system works on rural roads makes it even more impressive.

- I wish there was a betting site so we could all have some fun with our predictions

Tesla has defined the vision-based system, no doubt. From what I understand quite a bit of the grey area with vision is overcome mainly with software modeling, simply because they have more data than every manufacturer combined. Modeling has limits. It is limited though mainly in low-light and visibility challenging circumstances (heavy rain, fog, mist).

 

[Dave Cundiff]

I want to believe that, @Mark_AZR1T, but LiDAR is a light-based system. It might be better than camera vision, but apparently not dramatically so.

RADAR is a radio-wave-based system and works much better in rain, fog, or snow than cameras or LiDAR.

I'm sure this is oversimplified, but I hope it's useful.

***

Looks as if RJ is proceeding cautiously and correctly by first implementing camera and radar imaging, then moving to LiDAR when it makes sense.

Best wishes!

 

[Mark_AZR1T]

I want to believe that, @Mark_AZR1T, but LiDAR is a light-based system. It might be better than camera vision, but apparently not dramatically so.

RADAR is a radio-wave-based system and works much better in rain, fog, or snow than cameras or LiDAR.

I'm sure this is oversimplified, but I hope it's useful.

***

Looks as if RJ is proceeding cautiously and correctly by first implementing camera and radar imaging, then moving to LiDAR when it makes sense.

Best wishes!

The main reason most manufacturers are not up to speed with Lidar is now mainly based on the fact it creates bucket loads of data (geometrical not vision based) and the hardware requirements are rigid and finally, it only really can work well and evolve on software defined vehicles.

Rivian has some job posting for LiDar pro's with experience in the integration of LiDar and Radar....I would imagine that the main reason Rivian has gone with it's hardware upgrades is mainly for the 'stock value wealth creation' autonomous driving and the knowledge that surrounds it creates. They aren't going to be successful alone with R1, R2, R3, R4 etc., without this. It's a wise choice for company longevity.

 

[nsuglerio]

I think every brand integrates LIDAR into the process ( ie even Tesla using lidar equipped fleet vehicles ) but once you start having hours of data where you stereo video's point clouds are accurate that's where I think the real work begins.

For the sake of the matter, even with a LIDAR equipped vehicle with every sensor imaginable ( ultrasonics and radar too ) it's down to the software and the training. Not sure if Rivian is using Nvidia Drive AGX boards but Tesla brought in Jim Keller to help them with their hardware side and luckily had Andrej Karpathy for the software side.

In my eyes you need LIDAR just like you need cameras to read street lights. / signs / road lines. It's tough to know for certain what the most cost effective and safe solution will be, but we know the most cost effective one will be the ruler in the end.

With the improvements already seen with the latest Rivian and Comma update not mentioning Teslas 14.1s FSD, I'm really starting to weigh heavily on LIDAR serving its purpose quite well, on the training end.

I would love a reminder to this post when R2 comes out. I'm more confident it too will be a camera based system like the current gen2 R1 vehicles.

Sponsored