Ketchum: Wait for LiDAR? - The Rivian R2 LiDAR story isn’t what you think

[TexasBob]

This seems like such an odd debate. RJ has said:
Late 2026: Supervised Point-to-Point driving (similar to Tesla's FSD) on all Gen 2 R1 and R2 vehicles.
2027: Unsupervised, "eyes-off" Level 3 autonomy.
2028: Full commercial Level 4 robotaxi deployments (via their partnership with Uber).
If he is correct and meets his timeline, the lidar is relevant as an essential part of level 3 next year. There is not exactly room for debate (other than will he make his commitment or not).

https://thenextweb.com/news/rivian-...rom Rivian's current driver-assistance system.

Sponsored

 

[skyguyscott]

While RJ states that supervised point to point driving is coming to all GEN 2 Rivian vehicles, I would not trust any camera – only system with such a task due to the inherent limitations and failure modes inherent in such a system.

RJ also claims that the LiDAR - equipped R2 vehicles will not offer any additional benefits to customers, at least initially.

however, it is also true that a lidar equipped vehicle has the potential to gain much more accurate and reliable sensor awareness of its surroundings and is less likely to be fooled or miss a variety of objects that a camera, only system may fail to register. You can see a number of YouTube videos, where Tesla‘s camera only system has caused accidents, some fatal, while in auto pilot or FSD.

The belief that AI can quickly code a safe self driving algorithm is a misunderstanding of the limitations of AI and how it works. Specifically, AI recognizes patterns and builds only on past experiences. Its reasoning capabilities are limited and its ability to anticipate possible future events is effectively near zero.

I might come to trust a lidar- equipped vehicle with self driving capabilities only when that software has been proven safe and effective after several years of real world testing.

I think it’s also important to remember the many utopian promises made by our tech friends over the preceding decades and temper our expectations and enthusiasm based on the history of how those promises have panned out.

 

[EVtowing]

FSD was approved in europe because the safety was independently verified.

Recommend reading this analysis:

Tesla presented self-published and inflated “Full Self-Driving” safety statistics directly to government regulators in Sweden and the Netherlands as it lobbied for European approval, according to correspondence obtained by Reuters through public records requests.
The data includes a claim that FSD could have “saved 32,000 lives” — a figure that independent researchers say is based on the absurd assumption that every vehicle in the U.S., including freight trucks and motorcycles, would be replaced by an FSD-enabled Tesla.

As we reported last month, a Reuters investigation found that Tesla’s “Full Self-Driving” safety statistics are deeply misleading, relying on invalid data comparisons that exaggerate the system’s safety record. Ten of 11 independent traffic-safety researchers who reviewed Tesla’s methodology called it misleading marketing, not a serious safety study.
Now Reuters has obtained documents showing Tesla used those same inflated statistics when directly lobbying European regulators.

 

[skyguyscott]

Recommend reading this analysis:

well, that pretty much tracks with Tesla‘s modus operandi, and just goes to further validate my decision to go with Rivian rather than Tesla. I can tolerate mistakes made by a company operating in good faith, but I cannot abide a company willfully knowingly, and purposefully,manipulating the truth for profit.

 

[Jeremy3292]

I used to be a believer but am now very skeptical Tesla can achieve anything over L2 or maybe L3 with cameras only. FSD is VERY GOOD today and I like it a lot, but it struggles in anything other than good conditions. I had a critical disengagement on the highway a few weeks ago bc it was raining hard. AI4 M3P so the latest hardware available also. It was stop and go traffic so not even going fast, at no more than 35 mph and it couldn't handle the rain.

Elon said humans can drive with just eyes so why can't a car do just do vision only too? Well humans struggle to see in bad conditions too, so in order to be better than a human you need something better than a human...radar and LIDAR seem necessary at some level to "punch through" rain and still be able to "see". I'm not an expert just a layman with life experiences.

 

[macb00kemdanno]

I used to be a believer but am now very skeptical Tesla can achieve anything over L2 or maybe L3 with cameras only. FSD is VERY GOOD today and I like it a lot, but it struggles in anything other than good conditions. I had a critical disengagement on the highway a few weeks ago bc it was raining hard. AI4 M3P so the latest hardware available also. It was stop and go traffic so not even going fast, at no more than 35 mph and it couldn't handle the rain.

Elon said humans can drive with just eyes so why can't a car do just do vision only too? Well humans struggle to see in bad conditions too, so in order to be better than a human you need something better than a human...radar and LIDAR seem necessary at some level to "punch through" rain and still be able to "see". I'm not an expert just a layman with life experiences.

I'm with you on that. I use FSD extensively, and it works great under clear conditions. But when it starts raining, the speed drops. It starts raining a little harder, and it slows down even more. Heavy rain, and it starts going slow enough that I just take over myself.

But to be honest, in heavy rain, I'd rather be in full control. Same with snow. But at night or in broad daylight, FSD has been wonderful.

I was driving from Kinston, NC, to my home in Garner, NC, last night on I-42. I was in the slow lane, and the car put its blinker on and got over in the fast lane. At first, I was like, "WTF," because there weren't any cars directly in front of me, but as I got closer, I noticed a disabled car on the shoulder. Once it passed by the car, it got back over. Little touches like that make it feel a bit more "human."

 

[EVtowing]

Elon said humans can drive with just eyes so why can't a car do just do vision only too?

Yes, the full quote (to paraphrase slightly) is “If humans can drive using their eyes and brain then cars should be able to drive with cameras and a computer”.

Elon is a semi-autistic engineer with no empathy, not a neuroscientist and his reasoning is fundamentally flawed. Human brains are fundamentally different from computer AI and eyes are not cameras. The differences are numerous and profound. Humans have driving strengths and weaknesses compared to cars with cameras and software. Humans intuitively understand other human drivers in important ways that software cannot, and since essentially all cars are driven by humans for the foreseeable future that understanding is critical.

Elon has been promising L4/5 autonomous driving multiple times safer than the average human every year for the past decade and Tesla still isn’t even close to that. When I bought my Model X in 2017 I could have paid thousands of dollars for the FSD option (I elected not to) and today I still wouldn’t have it. My X has HW2.5 and even the current Teslas with HW4 have not reached L4/5.

In China, BYD may be closer to truly safe L4/5 than Tesla is. Or maybe not. The point is, Tesla is not in front. Waymo is ahead and providing over half a million driverless rides every week in multiple US states and has an impressive safety record. In Texas, Tesla is providing a miniscule fraction of that number in less than 20 Model Ys without a safety driver.

 

[justinkitswa]

I love the comment that "Elon says humans use eyes only, so why can't cameras only be just as good?" - He's incorrect, humans don't just use eyes, we're motivated by the fear of pain and death if we make a mistake while driving.

We're also infinitely more aware of our surroundings, through things that existing autonomy systems are nowhere near solving for - we hear and feel vibrations, so we know if the road surface is slick, or if there are mechanical problems we need to deal with. We can see emergency vehicles coming, and we know to get off the road every single time (even if there is just barely enough shoulder to do it)

We can see and interpret what other drivers are doing - I'll pull off the road if someone in front of me is driving with a ladder that's sliding around a truck bed, or if stuff is flying out of the back of a truck.

Can autonomy systems be trained to recognize all this? Sure, over time I believe eventually they will. Our problem is that we're all the lab rats for this development. The Tesla "FSD" system has tried to kill me a few times, and I won't touch it unless I know for a fact I am in a situation it can handle (like rolling down a freeway with the sun not blinding me) - The Tesla system is horrible at knowing when to call it quits, and that's what makes it dangerous.

 

[NinjaWrap]

Human brains are fundamentally different from computer AI and eyes are not cameras.

I've argued too much with the hardcore fanboys on this. I love FSD, and it's going to be the hardest adjustment going to Rivian, but there are limitations that cannot be ignored. The placement of the B pillar camera (as well as the viewing angle) along with the viewing angle of the repeater camera, gives it a giant blind spot on each side that in my experience in the mountains with intersections that have weird angles actually causes problems. It just cannot see some things at all. Meanwhile my head can lean forward and to the side to see in these weird situations.

 

[DuoRivian]

While RJ states that supervised point to point driving is coming to all GEN 2 Rivian vehicles, I would not trust any camera – only system with such a task due to the inherent limitations and failure modes inherent in such a system.

RJ also claims that the LiDAR - equipped R2 vehicles will not offer any additional benefits to customers, at least initially.

however, it is also true that a lidar equipped vehicle has the potential to gain much more accurate and reliable sensor awareness of its surroundings and is less likely to be fooled or miss a variety of objects that a camera, only system may fail to register. You can see a number of YouTube videos, where Tesla‘s camera only system has caused accidents, some fatal, while in auto pilot or FSD.

The belief that AI can quickly code a safe self driving algorithm is a misunderstanding of the limitations of AI and how it works. Specifically, AI recognizes patterns and builds only on past experiences. Its reasoning capabilities are limited and its ability to anticipate possible future events is effectively near zero.

I might come to trust a lidar- equipped vehicle with self driving capabilities only when that software has been proven safe and effective after several years of real world testing.

I think it’s also important to remember the many utopian promises made by our tech friends over the preceding decades and temper our expectations and enthusiasm based on the history of how those promises have panned out.

Good job Rivian is not camera only but have radar for gen 2.

 

[shandering]

however, it is also true that a lidar equipped vehicle has the potential to gain much more accurate and reliable sensor awareness of its surroundings and is less likely to be fooled or miss a variety of objects that a camera, only system may fail to register. You can see a number of YouTube videos, where Tesla‘s camera only system has caused accidents, some fatal, while in auto pilot or FSD.

Autopilot and FSD are not the same thing. Nobody has died from FSD which has crossed 12 billion miles.

in fact, FSD improves safety which is why the RDW independently approved FSD in the netherlands

It has been years since FSD has had a showstopping issue that was related to camera only

I might come to trust a lidar- equipped vehicle with self driving capabilities only when that software has been proven safe and effective after several years of real world testing.

There is currently zero evidence of a vehicle in china that uses lidar/radar that has a safety level even close to FSD. There is zero evidence that in an end to end transformer neural network system (as RJ calls it AV 2.0) that lidar even increases safety

There is also no evidence that anyone with E2E driving other than tesla can get to the level of safety they have now with FSD v14. If it's not a data collection issue there is some bottleneck for others not having a similar safety level.

 

[StinkyPete]

RJ also claims that the LiDAR - equipped R2 vehicles will not offer any additional benefits to customers, at least initially.

IMHO, there's no way that Rivian will "hold back" functionality - particularly safety related - if it can be realized early on with LiDAR - or better yet at launch. The idea that they would wait to roll something out until they can figure out how to apply it to non-LiDAR versions is a lawsuit waiting to happen. The first person injured in a LiDAR Rivian is going to be represented by a lawyer who is going look into all these statements and ask... "So you just gathered data to make the overall fleet safer, but didn't immediately make my client's car safer - even though the technology existed in the version my client paid for?"

I just can't see that the first LiDAR equipped R2 doesn't use LiDAR to SOME extent - even if it is not autonomy-related.

 

[shandering]

IMHO, there's no way that Rivian will "hold back" functionality - particularly safety related - if it can be realized early on with LiDAR - or better yet at launch. The idea that they would wait to roll something out until they can figure out how to apply it to non-LiDAR versions is a lawsuit waiting to happen. The first person injured in a LiDAR Rivian is going to be represented by a lawyer who is going look into all these statements and ask... "So you just gathered data to make the overall fleet safer, but didn't immediately make my client's car safer - even though the technology existed in the version my client paid for?"

I just can't see that the first LiDAR equipped R2 doesn't use LiDAR to SOME extent - even if it is not autonomy-related.

There is zero evidence of a car in china with lidar preventing an accident that tesla would not prevent. In fact you see deadly accidents with chinese ADAS that have lidar.

ones that tesla solved 4 years ago

There is no evidence that with E2E neural network driving that lidar does anything meaningful. In some cars it is purely there for marketing. It costs $200 for lidar and they can claim that it future proofs the car.

 

[skyguyscott]

...Tesla had to conduct years of testing with the RDW to get FSD approved

RDW did independent testing that found that FSD improved safety

The RDW independently drove millions of miles in europe. Tesla had to make hundreds of modifications to FSD, and they conducted 4500 tests on a test track for various scenarios...

I appreciate the passion you bring to the defense of Tesla's FSD, but you might want to link to sources in support of your position. Using hyperbole undercuts your argument; for instance, to achieve even 1,000,000 miles of testing would take one test vehicle 2.85 years if driven at an average speed of 40 mph, 24/7 non-stop, even for charging, or swapping out tires/drivers/etc. If we assume a fleet of 10 test vehicles, and given battery capacity, a 4/hr day driving, 5 days a week, non-stop at an average speed of 40 mph, then it would take about two years to achieve one million miles.

But in Europe, they measure distance in kilometers.

 

[skyguyscott]

Autopilot and FSD are not the same thing. Nobody has died from FSD which has crossed 12 billion miles.

in fact, FSD improves safety which is why the RDW independently approved FSD in the netherlands...

https://en.wikipedia.org/wiki/List_of_Tesla_Autopilot_crashes

From AI web summary of "Tesla FSD fatalities"

As of October 2025, there have been a total of 65 reported fatalities involving Tesla vehicles. Among these, 2 fatalities occurred while the Full Self-Driving (FSD) system was engaged.

Tesla reports a fatal crash rate of 5.6 deaths per billion miles driven for its vehicles, which is higher than the national U.S. average of 2.8 deaths per billion miles across all vehicles

Sponsored