Rivian full self-driving

[ATLRivvy]

Anyone plan on purchasing the Autonomy+ when is comes out? I plan too because once it improves, the price will increase

I’m willing to bet the opposite. Self-driving is going to be commoditized. Prices will go down over time. I remember when folks were paying multiple hundred of dollars in fees for just maps updates with connectivity on top of that.

Sponsored

 

[Gigabit]

If they ever get to true full self driving, we won't be able to own one. I think companies will have a fleet of full self driving vehicles that we rent for some duration. I hope im wrong on this but companies are all moving towards consumers not owning anything. /rant

I think this will be first seen in 2030 (complete speculation) but the compute isnt there for any place except fair weather places. It's hard enough to drive when the weather is perfect, can't imagine if every car was driving around in blizzard or monsoon weather. Once cars start to have object detection in road (think the big rig tire that delaminated their tread) and they are as good as a person actively driving that situation then we are close to full self driving.

 

[Weck]

Anyone plan on purchasing the Autonomy+ when is comes out? I plan too because once it improves, the price will increase

I'm planning to just pay the one time price and get it for my Gen 2 for the life of the vehicle. With the existing vision, radar, and tech stack, I expect point to point supervised to be coming at least.

Eyes off unfortunately will be reserved for vehicles with lidar and the next gen processing they announced a few weeks ago, despite saying it was coming to R1T gen 2 earlier last year.

 

[therealcmj]

I do not agree. Try to take a trip across states lines. Until this is managed throughout the country it won’t be fully implemented.
to have it in a city here and there is not fully rolled out.

The laws about vehicles are not federal; they're all at the state level. And always have been.

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

 

[TexasBob]

I’m willing to bet the opposite. Self-driving is going to be commoditized. Prices will go down over time. I remember when folks were paying multiple hundred of dollars in fees for just maps updates with connectivity on top of that.

Agree with you on this. In the far more technically advanced automotive market (China) where there are numerour companies offering systems like this already (with various advancaed in-house and purchased chipsets), your prediction has already come true and prices have fallen to zero. BYD, for example, includes their advanced system for free in their luxury brand (YangWang) with multiple LiDARS and a chipset every bit as advanced as RAP1 hopes to be. In the lower tier models, it is a less advanced system.

 

[Mark_AZR1T]

Agree with you on this. In the far more technically advanced automotive market (China) where there are numerour companies offering systems like this already (with various advancaed in-house and purchased chipsets), your prediction has already come true and prices have fallen to zero. BYD, for example, includes their advanced system for free in their luxury brand (YangWang) with multiple LiDARS and a chipset every bit as advanced as RAP1 hopes to be. In the lower tier models, it is a less advanced system.

Just got back from Shenzhen and spent time riding in several Chinese EVs. What stood out immediately was how cautious China is with self-driving. A DIDI rideshare driver told me self-driving features aren’t allowed at all in rideshare vehicles due to safety concerns.

At an XPENG dealership, the sales manager referenced their CEO benchmarking Tesla FSD 14.2x. He said the SU7 showed fewer interventions overall, while acknowledging Tesla’s stack is still the most comprehensive end-to-end system available.

On the hardware side, the fit and finish of BYD, NIO, and especially the XPENG SU7 was impressive. At around 130 km/h, wind noise was minimal. For longer-term context, my daughter in Australia has been driving a basic BYD Atto 3 for two years, and even that has almost no wind noise at 75 mph.

One note on Rivian. It’s one thing to host an AI Day and announce a custom chip, even if it isn’t fully designed in house. It’s another thing entirely to have billions of real-world driving miles feeding the system. That data advantage is what enables Tesla to move toward true self-driving.

From where I sit, I don’t see Rivian being anywhere near Tesla on self-driving until at least 2028.

 

[TexasBob]

I'm sure that's not the term Rivian is using, bit I'm curious how many think Rivian will get to true point to point self-driving.

Whether or not you are interested personally.

Due to a health issue I'm no longer able to drive and this would enable me to be more independent. Though I'm guessing there will be a requirement that someone in the vehicle be able to take control quickly. And regardless of if they get there, it'll be a wait.

Thanks.

FWIW, I have been tracking this very closely for my own personal reasons. I am not going to buy another vehicle unless it have true eyes-off capability at least on the freeway. Fortunately, there are several sytems from several manufacturers coming in 2027 / 2028 that should actually deliver real eyes-off functionality and 50/50 Rivian will be one of them. Most will be limited to highways, so it is great for commutes and long-distance travel but not a true alternative if you are unable to drive.

As negative as I am about Tesla, I think there is a real chance they will get there at the end of 2027 when they introduce their new AI5 chipset and improved sensor suite. AI5 is powerful like the RAP1 or NVIDIA Thor to handle the advcanced reasoning models Rivian discussed on autonomy day. The new sensor suite will include some of the upgrades that they have already put into the Austin Model Y "robotaxi" mules (the Austin vehicles run a 3 SOC computer, a phoenix radar, and redundant actuators for brakes and steering). But a word of caution: the current consumer HW4 vehicles are unlikely to ever be capable of autonomous driving. That is why they had to upgrade the piot vehicles.

Waymo (and Baidu if they ever come to the US) are on rocket-fast expansion program. I see Waymo vehicles driving themselves around my neighborhood all the time these days with no passengers or drivers as they prepare for launch here in Houston and a dozen other cities early this year. Not sure where you are located in MN, but they have announced Minneapolis by eoy. The cost of a Rivian buys a whole lot of Waymo Rides! Not as nice as having your own vehicle but next best thing.

Oh, there is one dark horse in the race: the Lucid Earth. Weirdly enough, their robotaxi partnership with NVIDIA, Nuro, and Uber, plus the massive Thor system and their embrace of the full NVIDIA AV drive stack with Nuro Driver. Little Lucid may beat everyone to the street with a real L4 consumer system in the Earth. They say 2028...

 

[therealcmj]

It’s another thing entirely to have billions of real-world driving miles feeding the system. That data advantage is what enables Tesla to move toward true self-driving.

Having a pile of data is part of it. Having that data be of high enough quality to actually be usable, having good algorithms to train on it is another. And having sufficient sensors to then apply the trained model to what's going on in real time around you is another.

Yes, Tesla has one of those. The others are subject to very vocal debate in the industry.

From where I sit, I don’t see Rivian being anywhere near Tesla on self-driving until at least 2028.

I don't disagree with this. Rivian (and everyone else) is being much more cautious than Tesla.

 

[Mark_AZR1T]

Having a pile of data is part of it. Having that data be of high enough quality to actually be usable, having good algorithms to train on it is another. And having sufficient sensors to then apply the trained model to what's going on in real time around you is another.

Yes, Tesla has one of those. The others are subject to very vocal debate in the industry.

I don't disagree with this. Rivian (and everyone else) is being much more cautious than Tesla.

Having a pile of data is part of it. Having that data be of high enough quality to actually be usable, having good algorithms to train on it is another. And having sufficient sensors to then apply the trained model to what's going on in real time around you is another.

Yes, Tesla has one of those. The others are subject to very vocal debate in the industry.

I don't disagree with this. Rivian (and everyone else) is being much more cautious than Tesla.

Granted, all I have is lane assist, and even that’s limited to certain roads on both of our Rivians. That said, I’ve been on three rides in Teslas running FSD, and it was genuinely impressive.

No interventions in city or highway driving. It took us cleanly from point A to point B, handled yellow lights, railroad crossings, and even a cyclist in a bike lane without issue. The whole FSD concept is a bit much for me personally, but it definitely makes you wonder where this is headed.

As it stands today, Rivian just isn’t on the same planet as Tesla when it comes to this.

 

[TexasBob]

One note on Rivian. It’s one thing to host an AI Day and announce a custom chip, even if it isn’t fully designed in house. It’s another thing entirely to have billions of real-world driving miles feeding the system. That data advantage is what enables Tesla to move toward true self-driving.

Tesla's data advantage is wildly overblown by people pumping the stock (not saying that is what you are doing, but that is the origin of the story). Here is why:
1) The data set itself has a fatal flaw - it is a single sensor (vision) data set. If you are trying to use it to train a multi-sensor system like Vision + Lidar + Radar it is not useful. Even Tesla, which is adding radar to its "vision only" system, has a problem here.
2) Simulation software has taken massive leaps forward and you can now get billions of miles of Generative training (with every conceivable edge case) without ever hitting the road. And NVIDIA is giving this Omniverse capability to its customers. Also the simulation systems are far more advanced and now widely available (hence Tesla sidelined its dojo - partly because the AI4 chip was underpowered c.f. previous post)
3) The driving models are no longer what are known as reflexive NN models that rely on repetition and machine learning. They now have a supervisory reasoning layer - like the NVIDIA alpamayo model - that use a world model that understands the physics and intent in the world it is encountering (tech borrowed from the LLM world). These models rely on high quality training data an synthetic logic. The old more is better concept does not apply here.
So I would be more optimistic about Rivian specifically and I am completely confident that Tesla is not even actually ahead of most competitors. It is further along a path that only it chose to take. They are already having to backtrack (adding sensors and a reasoning layer) while trying not to tell people they are going backwards to go forward.

 

[Mark_AZR1T]

Tesla's data advantage is wildly overblown by people pumping the stock (not saying that is what you are doing, but that is the origin of the story).

....I would be more optimistic about Rivian specifically and I am completely confident that Tesla is not even actually ahead of most competitors. It is further along a path that only it chose to take. They are already having to backtrack (adding sensors and a reasoning layer) while trying not to tell people they are going backwards to go forward.

I understand the underlying physics, it was my major, and I’m very familiar with lithography and chipset design from years in that industry. I also care about real-world results and have firsthand experience with Chinese self-driving vehicles, which are advancing quickly.

That said, three rides in a Tesla running FSD 14.2 were genuinely impressive. Rivian’s lane assist isn’t even as robust as what the Big Three are offering today.

We’re clearly looking at this from two different lanes in terms of where Tesla actually is and where Rivian is headed, and that’s fine. What we can certainly agree on, is the final approach for hardware and software is yet to be determined. Both are accelerating at ridiculous rates and by the time they are in vehicles, they are partially obsolete. I imagine it will be that way for the next decade.

 

[SadHill]

The laws about vehicles are not federal; they're all at the state level. And always have been.

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

That is correct but the states need to get together and figure it all out. Right now they have not done it. Some cities have the approval to do self driving but the state still does not. It is all a mess

 

[ATLRivvy]

FWIW, I have been tracking this very closely for my own personal reasons. I am not going to buy another vehicle unless it have true eyes-off capability at least on the freeway. Fortunately, there are several sytems from several manufacturers coming in 2027 / 2028 that should actually deliver real eyes-off functionality and 50/50 Rivian will be one of them. Most will be limited to highways, so it is great for commutes and long-distance travel but not a true alternative if you are unable to drive.

As negative as I am about Tesla, I think there is a real chance they will get there at the end of 2027 when they introduce their new AI5 chipset and improved sensor suite. AI5 is powerful like the RAP1 or NVIDIA Thor to handle the advcanced reasoning models Rivian discussed on autonomy day. The new sensor suite will include some of the upgrades that they have already put into the Austin Model Y "robotaxi" mules (the Austin vehicles run a 3 SOC computer, a phoenix radar, and redundant actuators for brakes and steering). But a word of caution: the current consumer HW4 vehicles are unlikely to ever be capable of autonomous driving. That is why they had to upgrade the piot vehicles.

Waymo (and Baidu if they ever come to the US) are on rocket-fast expansion program. I see Waymo vehicles driving themselves around my neighborhood all the time these days with no passengers or drivers as they prepare for launch here in Houston and a dozen other cities early this year. Not sure where you are located in MN, but they have announced Minneapolis by eoy. The cost of a Rivian buys a whole lot of Waymo Rides! Not as nice as having your own vehicle but next best thing.

Oh, there is one dark horse in the race: the Lucid Earth. Weirdly enough, their robotaxi partnership with NVIDIA, Nuro, and Uber, plus the massive Thor system and their embrace of the full NVIDIA AV drive stack with Nuro Driver. Little Lucid may beat everyone to the street with a real L4 consumer system in the Earth. They say 2028...

There is 0% chance either Lucid or Rivian is in a leadership position by 2028. Lucid's ADAS system today is like 6-years behind current tech and Rivian is just now getting to the level of performance Ford/GM achieved in late-2022. They aren't getting there just by brute forcing with better silicon.

 

[TexasBob]

I understand the underlying physics, it was my major, and I’m very familiar with lithography and chipset design from years in that industry. I also care about real-world results and have firsthand experience with Chinese self-driving vehicles, which are advancing quickly.

That said, three rides in a Tesla running FSD 14.2 were genuinely impressive. Rivian’s lane assist isn’t even as robust as what the Big Three are offering today.

We’re clearly looking at this from two different lanes in terms of where Tesla actually is and where Rivian is headed, and that’s fine. What we can certainly agree on, is the final approach for hardware and software is yet to be determined. Both are accelerating at ridiculous rates and by the time they are in vehicles, they are partially obsolete. I imagine it will be that way for the next decade.

I think that the FSD as it currently stands is a very useful system that does what it does very well. It is a very impressive, but highly unreliable, self-driving system that will crash at a rate 100X more often than a typical human so you need to watch it like a hawk and never be complacent.

It won't ever get beyond that with the current hardware which is why (I told you so) Tesla has started shipping revised hardware - news out today. But even the new hardware won't get it across the threshold to L3 or L4, IMO. It needs AI5 to get there and at least the new radar to give it ground truth depth information.

Oh and on one thing we are in violent agreement: the Rivian Driver + and Gen 2 whatever-they-call-it absolutely blows. It is awful. I rented a poj Toyota and the free system was much better.

Sponsored