Rivian May Be Much Closer to FSD Than You Think

[ElGuano]

Wasn't it just a couple of years ago that Rivian said it wasn't pursuing FSD?

Did they make a (really late) about-face, or have they been "secretly" focusing on it while publicly saying otherwise all this time?

Sponsored

 

[TexasBob]

Just noting that the XPENG X9 is using 3 of its own chips called the TuringAi chips that run a combined 2,250 Trillion Operations Per Second while the dual Orin in the Gen 2 Rivian run 500Tops. (Still 10X the Tesla HW 4 ). RIvian will need a significant hardware upgrade to catch up to competitors developing their own chips or buying the NVIDIA developer kit.

 

[j.w.s]

I go into way more depth in the video if you want to see my beautiful face:

With Rivian’s AI and Autonomy Day coming up on December 11th, they are making some seriously bold claims. RJ is basically betting the farm on autonomy because he thinks that in 5-10 years, a car that *can't* drive itself will be as appealing as a flip phone in 2024.

But are they actually close, or is this just expensive vaporware?

How are they moving so fast?
* Tesla has a massive head start, but Rivian has the glorious "second mover" advantage. Basically, they got to watch Tesla walk into the glass doors first so they didn't have to.
* Brain Power: They are building on Nvidia Orin chips that are specifically designed for AI. These chips simply didn’t exist when Tesla started their journey 84 years ago (or however long it's been). So they had to engineer their own to get the performance they wanted.
* HD Eyes: Rivian is using high-megapixel cameras and premium radar. It turns out that the "Garbage In, Garbage Out" rule applies to cars, too. Teaching a computer to drive is a lot easier and faster when it’s looking through 4K lenses instead of a potato. To be clear... I'm not saying Tesla uses Idaho spuds for cameras, I'm just saying having multiple kinds of high quality sensors can speed up the training process.

Is it actually working? Sure, the specs look good on paper, but can Rivian actually deliver? There are no guarantees, but I feel that it's looking pretty promising.
* The Gen 2 Glow-Up: If you’ve driven a Gen 2 recently, the switch from the old MobileEye system to Rivian’s internal stack is night and day.
* Hallucinations (The helpful kind): the driver's screen visualizations have started "hallucinating" lane lines on unmarked roads. Usually, hallucinations are bad, but here it seems the AI is actively inventing a safe path to drive on unmarked roads. This is a huge step toward universal Hands-Free.
* On the highway, the system seems to be reading the "body language" of other drivers—especially when making lane changes. It does an excellent job of predicting the movement of other drivers on neighboring lanes. This kind of predicitve behavior is HUGE when it comes to full autonomy

The Elephant in the Room: Subscriptions
We know a subscription is coming. Deep breaths.

While I hate monthly fees as much as anyone, I think this is different from BMW charging a subscription to warm your butt (*insert MASSIVE eye roll here*). This is an ongoing development cost. We aren't just unlocking hardware that's already there; we are funding the army of engineers teaching cars to drive over the next decade(s) of development.

I’m expecting them to drop the pricing details on December 11th.

What do you think? Does Rivian have a shot at the competing? Rivian is betting all their chips on this... Pun very much intended.

The Waymo vehicles do a darn good job of driving themselves in most California weather conditions. Let's see, in the new Gen 6 cars that's:

13 cameras, four lidar sensors including the big 360 lidar on top, six radar sensors, and a bunch of microphones as well. And that's the brand new cars, down from even more cameras and lidar in Gen 5.

That, apparently, is what it takes to do self-driving properly. So unless there is a <insert brand here> vehicle for sale that at least comes close to that sensor suite, I think I'll stick to Waymo for the next few years.

 

[NY_Rob]

The Waymo vehicles do a darn good job of driving themselves in most California weather conditions. Let's see, in the new Gen 6 cars that's:

13 cameras, four lidar sensors including the big 360 lidar on top, six radar sensors, and a bunch of microphones as well. And that's the brand new cars, down from even more cameras and lidar in Gen 5.

Kudos to Waymo for their advanced platform, but even the smallest accident is going to total one of their vehicles because of the sheer number of devices built in and the gigantic number of cables and connections needed to connect everything. Imagine trying to troubleshoot a complex vehicle like that? Rivian has to fly in factory engineers to diagnose a break in a single wire in some cases, and compared to the Waymo vehicle described above the Rivian vehicles are simple devices.

 

[TM1]

Rivian can't even lane keep on city streets, no way they get this going on their own.
They have to be at least 3 yrs away from buying it from someone else.

 

[bigsky]

Heck, Rivian cannot get even much simpler, trivial things like Homelink, MFA, etc. to work well or are half-assed attempts at best. And Rivian now might be closer to FSD than we think? Yeah, sure.

Tesla, the supreme leader, is much farther ahead on this, but I seriously doubt that even Tesla will get there eventually. I will start to believe that true FSD from anybody is here and near when the car can safely navigate around all the newborn winter potholes in the road.

I have come to accept that Rivian still cannot implement correctly simple things. that it is still amateur hour at Rivian in terms of software integration and implementation. I just hope that a Rivian finally equipped with true FSD will not look like a car that is full or herpes, tumors, cists, bumps.

 

[daeHelkcunK]

That’s good ‘cause FSD sucks and is a Tesla gimmick

The latest version of Tesla FSD is incredible! I don't understand how you could say that unless you haven't used it in over a year. I'm happy to pay an extra $3600 over the 3 year lease of my car for the convenience of hands free driving on long trips. I hope Rivian's system is as good.

 

[Greg Chick]

My cameras get rained on, get mud splashed on, I get notices "Rear sensors not working", Etc.

What will happen when any vehicle using any sensors for driving gets rained or snowed on?

 

[Jeff B.]

I still enjoy the driving experience as part of my vehicle ownership. If that gets removed from the equation or if the cost of ownership is so high, then I'll be one of the many dumb people on the road that your FSD or AI will have to guesstimate to avoid. Flying cars, FSD / autonomous vehicle based AI, it all sounds like a great future for attorneys and insurance companies.

I guess I am glad I still have my old truck which is full self drive, runs on dinosaur juice and farts noxious gas.

 

[portdirect]

Rivian is barking up the wrong tree IMO. Autonomy is coming and it is worth $1,500 lifetime (+/-). I drove 550 miles last week using Toyota's free POJ LKA and Adaptive Cruise and it was, on most dimensions superior to the Rivian Gen 1 or Gen 2 system (I have both).

We have moved to a time when any OEM can buy an NVIDIA Drive system and software and a bunch of them are. Rivian is dependent on older generation (and much, much slower) NVIDIA chips. The best think RJ can do on Autonomy Day is say, "we are a fast adopter of the NVIDIA AGX bundle and offering them on all our vehicles for $1,500 starting next year. The Gen 2 R1 vehicles with the significantly slower hardware can be upgraded for $2,500.

https://blogs.nvidia.com/blog/drive-agx-developer-kit-general-availability/

Much as it saddens me to agree with this - I kinda do. Though unfortunately/fortunately (depending on perspective) - Rivan have shown themselves to be masters of marketing in this space. Eg diverting away from the fact that the stack was until recently *just* mobileye doing the actual business end of the equation.

What we do know is that the Autonomy stack is running on DriveOS, above a reference AGX Orin design (which of the current Nvidia offerings is the low end SKU), and at (at least in the early days) using DriveWorks SDK - it would be interesting if Rivian have actually gone a different way than Nvidia and the horde of other mfrs using the stack (eg: - which name drops, Uber, Stellantis, Lucid, Mercedes-Benz, Aurora, Volvo, Waabi, Avride, May Mobility, Momenta, Nuro, Pony.ai, Wayve and WeRide - while others are also using the stack without being as public about it - GM certainly are), but as they are so in the ecosystem would really require some quite exceptional justification to do so.

 

[1stPlace]

The Waymo vehicles do a darn good job of driving themselves in most California weather conditions. Let's see, in the new Gen 6 cars that's:

13 cameras, four lidar sensors including the big 360 lidar on top, six radar sensors, and a bunch of microphones as well. And that's the brand new cars, down from even more cameras and lidar in Gen 5.

That, apparently, is what it takes to do self-driving properly. So unless there is a <insert brand here> vehicle for sale that at least comes close to that sensor suite, I think I'll stick to Waymo for the next few years.

I believe Rivian's Gen 2 "perception stack" as RJ likes to say, has 11 cameras and 5 radar sensors, pretty close to Waymo, except for the lidar, which I and others might speculate will be in the R2, even if it is only front-facing. So, I think this comes close, if you are only looking at the inputs. The biggest issue other than that is the sheer amount of data need for learning/training the system. Rivian has a huge amount because of how they are uploading and processing their customers' drives.

 

[gerrylum]

Its a great option for long trips. Took our Model 3 on a 4,000 mile trip this summer and it drove 99% of time. So much easier and refreshed once we got to our destination. Then cancel when not taking trips.

But the price is insane. If I can do a drive like Dallas to Houston 98% hands-free and only pay a one-time $999 for a Comma 3X, I can't see why I would ever pay a $99/month subscription for something like FSD.

 

[mpshizzle]

But the price is insane. If I can do a 98% hands-free drive from Dallas to Houston for only a one-time $999 investment in a Comma 3X, I can't see why I would ever pay a $99/month subscription for something like FSD.

What comma lacks (at least for now) is 360 awareness of it's surroundings. If there's a vehicle next to you, but it's not in the range of the blind spot monitor, it will change lanes right into that vehicle. Because of this lack of 360 awareness it can't achieve full autonomy.

That said, at some point in the future, if they achieve their dream of being able to tap into the vehicle's full suite of cameras and sensors to get that 360° perception, and run a full autonomy stack with open source software, that would be excellent.

 

[gerrylum]

What comma lacks (at least for now) is 360 awareness of it's surroundings. If there's a vehicle next to you, but it's not in the range of the blind spot monitor, it will change lanes right into that vehicle. Because of this lack of 360 awareness it can't achieve full autonomy.

That said, at some point in the future, if they achieve their dream of being able to tap into the vehicle's full suite of cameras and sensors to get that 360° perception, and run a full autonomy stack with open source software, that would be excellent.

Full autonomy is a red herring, in my opinion.

Let's take the semi-autonomous cannonball run records as an example. The Comma achieved a 99.125% hands-free run, Tesla FSD achieved a 98.69%. Whatever Comma may lack, when you are talking about what it's able to achieve especially at the price points involved, it really does beg the question how much these other services (FSD, RAP+) are really worth in terms of subscription prices.

To me, I don't think I could justify something like FSD at $99 a month - even if they managed a 100% hands free drive. That extra .875% just isn't worth the extra expenditure.

Sponsored