WIRED / RJ interview: Gen 2.5 Autonomy + What Happens if R2 Flops

[shandering]

It would seem to be common sense that driving on a highway would be the easier part since as you state downtown and urban areas are the most difficult. You don’t get stop signs and traffic lights on interstates. So starting there with decent hands free and eyes free makes sense.

Absolutely fucking not. The freeway is the easiest to fake that you can drive for hours at a time. The freeway is the most challenging driving overall.

The risk is the highest, the chance of death is the highest, and the edge cases are the most dangerous.


We are talking about rivian here, a company who actively does not report to the NHTSA their ADAS collisions even though it is legally required.

They also don't collect videos from any ADAS accidents unless you bring your computer and they manually download it.

This is in contrast to tesla who reports all ADAS accidents, collects accident videos and trains on them.

Sponsored

 

[SANZC02]

Absolutely fucking not. The freeway is the easiest to fake that you can drive for hours at a time. The freeway is the most challenging driving overall.

The risk is the highest, the chance of death is the highest, and the edge cases are the most dangerous.


We are talking about rivian here, a company who actively does not report to the NHTSA their ADAS collisions even though it is legally required.

I have to disagree with you here. On a freeway everyone is going in the same direction, no stop signs or stop lights. No pedestrians, driveways, bicycles, runners, no intersections, no parked cars where kids or pets can run out from behind.

Speeds are higher on the freeway but that is because the risk is lower. For every adverse scenario on the freeway there is probably a dozen on side streets.

 

[shandering]

I have to disagree with you here. On a freeway everyone is going in the same direction, no stop signs or stop lights. No pedestrians, driveways, bicycles, runners, no intersections, no parked cars where kids or pets can run out from behind.

Speeds are higher on the freeway but that is because the risk is lower. For every adverse scenario on the freeway there is probably a dozen on side streets.

One mistake on the freeway and you're dead. One mistake on surface streets and the quick reaction time/accident avoidance of a tesla takes over and prevents something deadly

you have bad problems on the freeway like debris, potholes, people swerving into you, accidents with pedestrians on the freeway, people randomly walking across the freeway, construction zones, bar gates blocking entire entrances to sections of road at high speed, etc.

You stop moving on the freeway even once and you can get rear ended and die. The risk is extremely high

 

[SANZC02]

One mistake on the freeway and you're dead. One mistake on surface streets and the quick reaction time/accident avoidance of a tesla takes over and prevents something deadly

you have bad problems on the freeway like debris, potholes, people swerving into you, accidents with pedestrians on the freeway, people randomly walking across the freeway, construction zones, bar gates blocking entire entrances to sections of road at high speed, etc.

You stop moving on the freeway even once and you can get rear ended and die. The risk is extremely high

I’ll just leave this with you to review:

Driving on freeways is generally safer than on surface streets. Studies indicate that controlled-access highways, or freeways, have a significantly lower fatality rate per vehicle mile traveled (VMT) compared to surface streets. This is attributed to factors such as higher speed limits, fewer intersections, and controlled access points, which reduce the likelihood of accidents.

Crash Rate Comparisons

• Freeways typically show lower crash rates overall, especially for fatal accidents.
• Surface streets often have more variables, such as pedestrians and traffic signals, contributing to higher accident rates.

Recent Findings

• A comprehensive study highlighted that freeway crash rates are lower across various geographic areas and outcomes.
• Additionally, narrower lanes on surface streets have been suggested to improve safety, indicating that road design plays a crucial role.

 

[shandering]

I’ll just leave this with you to review:

Driving on freeways is generally safer than on surface streets. Studies indicate that controlled-access highways, or freeways, have a significantly lower fatality rate per vehicle mile traveled (VMT) compared to surface streets. This is attributed to factors such as higher speed limits, fewer intersections, and controlled access points, which reduce the likelihood of accidents.

Crash Rate Comparisons

• Freeways typically show lower crash rates overall, especially for fatal accidents.
• Surface streets often have more variables, such as pedestrians and traffic signals, contributing to higher accident rates.

Recent Findings

• A comprehensive study highlighted that freeway crash rates are lower across various geographic areas and outcomes.
• Additionally, narrower lanes on surface streets have been suggested to improve safety, indicating that road design plays a crucial role.

If you're talking about deaths, the fatal accidents that would happen on city streets are already being prevented by FSD/waymo

FSD/waymo are extremely safe from a death perspective and modern cars are extremely safe when it comes to crash safety already

It's the accidents on freeways where the risk is. One death and it is a major setback. Also from an injury perspective, one freeway accident can cause pretty serious injury

 

[SANZC02]

If you're talking about deaths, the fatal accidents that would happen on city streets are already being prevented by FSD/waymo

FSD/waymo are extremely safe from a death perspective and modern cars are extremely safe when it comes to crash safety already

It's the accidents on freeways where the risk is. One death and it is a major setback. Also from an injury perspective, one freeway accident can cause pretty serious injury

I’m not talking about anything specific only that statistically freeway driving has less risk than service streets and far fewer adverse events required when coding.

I’ll leave it at that, as the old saying goes horse and water…

 

[shandering]

I’m not talking about anything specific only that statistically freeway driving has less risk than service streets and far fewer adverse events required when coding.

I’ll leave it at that, as the old saying goes horse and water…

The issue though is humans are better than cars at identifying adverse scenarios and taking precaution.

What self driving vehicles are good at is sensing objects and avoiding them. Also having quick reaction times.

The safety of current self driving vehicles only comes from 360 sensing and fast reaction time.

The area where self driving vehicles are significantly less safe than humans is processing all the details on the road and understanding the context.

For example a self driving car may not understand what road flares mean or other types of things that would mean that a particular lane is not safe to drive in. From that a pedestrian could pop out of nowhere and be killed.

A self driving vehicle also may stop for a wall of smoke it thinks is solid and then end up getting rear ended.

Self driving vehicles don't drive with always better abilities than humans.

Waymo is extremely safe but in san antonio it drove into flooding and got washed away for 3 days. They are currently shut down and someone could have died if they were inside the car.

you have to think about all the scenarios that are easy for a human but hard for a self driving car. Tesla has problems with bar gates blocking the entrance to express roads. Waymo even with lidar would have an equally hard time. For a human this is shockingly easy.

 

[shandering]

And if we ignore waymo for a second. Look at all the chinese cars with lidar and look at the accident videos posted on various website. Hitting overturned semis in broad daylight. Crashing in ways that tesla solved 4 years ago.

It's hard enough to get FSD v14 levels of safety on the freeway and rivian ADAS is poor enough as it is, and has never once demostrated a shred of accident avoidance behavior.

it's laughable if you think they are going to go from this to driving at 70mph+ unsupervised

 

[TexasBob]

If you're talking about deaths, the fatal accidents that would happen on city streets are already being prevented by FSD/waymo

FSD/waymo are extremely safe from a death perspective and modern cars are extremely safe when it comes to crash safety already

It's the accidents on freeways where the risk is. One death and it is a major setback. Also from an injury perspective, one freeway accident can cause pretty serious injury

You cannot really compare Waymo and FSD. Waymo Driver is currently delivering 500,000 completely autonomous rides per week in eleven US metro areas. The consumer FSD is delivering zero autonomous rides ever anywhere. The modified robotaxi version is delivering fewer than 500 autonomous rides per week in three places.

 

[shandering]

You cannot really compare Waymo and FSD. Waymo Driver is currently delivering 500,000 completely autonomous rides per week in eleven US metro areas. The consumer FSD is delivering zero autonomous rides ever anywhere. The modified robotaxi version is delivering fewer than 500 autonomous rides per week in three places.

Your numbers are definitely wrong there. Waymo is shut down at least in san antonio (not sure where else) and has been for weeks. They almost killed somebody due to driving in flooded waters. They now shut down in most metros with light rain because of it. Tesla also gives more than 500 rides per week. They operate around 14 vehicles and they do at least 7 ride a day, each. In terms of ride length, many people ride for 1 hr at a time because many people just take them for fun, rather than as a transportation service.

We are talking about reaction time preventing deaths where FSD reacts faster than waymo does.

We are talking about the things the cars are totally ignorant about which is the reason they don't go on the freeway.



Look at FSD reaction time here. Waymo has shown nothing close to this and they run all their sensing (cameras and lidar) at 10hz where tesla runs at 36hz

What tesla and waymo have in common is they both refuse to go on the freeway. They also drive around with an equal amount of ignorance to things on the road, road signs, and words in general.

If you're driving at night on the freeway and there is a sign that says "accident ahead in 3 miles" and there are a bunch of road flares blocking a lane, both cars are equally ignorant to these things. You can only achieve equivalent safety in this context by having faster than human reaction time and better than human low light perception.

If you increase intellignece that can have negative impacts on reaction time which decreases safety.

The other issue is if you use tricks like following the lead of other cars, you can occasionally be lead to do incorrect things.

 

[TexasBob]

LOL nearly everything you have written here is incorrect
Sensor Refresh Rates and Perception Architecture: Comparing system reaction times by asserting Waymo runs at 10Hz while Tesla runs at 36Hz mischaracterizes the physics of autonomous compute. While the mechanical rotation of a LiDAR unit may operate at 10Hz, a redundant Level 4 architecture relies on continuous, multi-modal sensor fusion incorporating high-framerate cameras and radar. System latency is defined by the compute loop, not the rotational speed of a single localized sensor. Furthermore, Tesla's current unsupervised operations in Texas are not running the standard consumer FSD software; they require Hardware 4 processing and rely heavily on route-specific pre-mapping and localized data labeling.

Waymo experienced an incident on April 20, 2026, when an unoccupied vehicle entered a flooded lane near Salado Creek in San Antonio and was swept away. Waymo subsequently paused passenger service in the city and issued a software recall. However, there were no passengers in the vehicle, and no injuries occurred. The claim that the company "almost killed somebody" is fabricated.

Waymo does not halt operations for light precipitation. The recent operational pauses in San Antonio and Atlanta occurred specifically in response to flash flood events. In Atlanta, an unoccupied vehicle became trapped in standing water during a sudden, intense downpour on May 20, 2026. Waymo vehicles routinely operate in standard rain environments across their active markets. Tesla does not have this problem because it does not operate an autonomous taxi fleet.

But, look, this is a Rivian forum. No one cares about Tesla FSD or its failed attempts at a robotaxi. I think you are looking for Teslarati

 

[shandering]

Sensor Refresh Rates and Perception Architecture: Comparing system reaction times by asserting Waymo runs at 10Hz while Tesla runs at 36Hz mischaracterizes the physics of autonomous compute. While the mechanical rotation of a LiDAR unit may operate at 10Hz, a redundant Level 4 architecture relies on continuous, multi-modal sensor fusion incorporating high-framerate cameras and radar. System latency is defined by the compute loop, not the rotational speed of a single localized sensor. Furthermore, Tesla's current unsupervised operations in Texas are not running the standard consumer FSD software; they require Hardware 4 processing and rely heavily on route-specific pre-mapping and localized data labeling.

Maybe if you didn't rely on AI you wouldn't get such bogus answers.

Waymo runs their cameras at 10hz. It's described in the patent and all of the video they share from the cars (showcasing accident avoidance) is at low framerate. The waymo open dataset they share is also at 10hz. There is no evidence they run their sensors at high framerate. The only sensors that would perform independent classification of objects would be cameras and lidar.

As far as robotaxi it does appear they are running consumer software. Or at least the software and map details are not any different than what is found in consumer cars. There is no HD mapping and the only difference from consumer software is that the routing avoids freeways and sections of frontage roads that could accidentally put you on the freeway.

The hardware difference between robotaxi and consumer cars is camera cleaning. How is the company claiming R2 is going to have L4 driving going to deliver that without camera cleaning? When the company that has 10s of billions of miles running FSD knows that you need it?

Waymo experienced an incident on April 20, 2026, when an unoccupied vehicle entered a flooded lane near Salado Creek in San Antonio and was swept away. Waymo subsequently paused passenger service in the city and issued a software recall. However, there were no passengers in the vehicle, and no injuries occurred. The claim that the company "almost killed somebody" is fabricated.

Waymo issued a recall but it did not fix their issue. Several days after the fix in other cities they drove into floodwaters again. Waymo has been paused in san antonio since april. They are not running now.

 

[TexasBob]

Maybe if you didn't rely on AI you wouldn't get such bogus answers.

Waymo runs their cameras at 10hz. It's described in the patent and all of the video they share from the cars (showcasing accident avoidance) is at low framerate. The waymo open dataset they share is also at 10hz. There is no evidence they run their sensors at high framerate. The only sensors that would perform independent classification of objects would be cameras and lidar.

As far as robotaxi it does appear they are running consumer software. Or at least the software and map details are not any different than what is found in consumer cars. There is no HD mapping and the only difference from consumer software is that the routing avoids freeways and sections of frontage roads that could accidentally put you on the freeway.

The hardware difference between robotaxi and consumer cars is camera cleaning. How is the company claiming R2 is going to have L4 driving going to deliver that without camera cleaning? When the company that has 10s of billions of miles running FSD knows that you need it?

Waymo issued a recall but it did not fix their issue. Several days after the fix in other cities they drove into floodwaters again. Waymo has been paused in san antonio since april. They are not running now.

In the two and a half hours since you typed that, Waymo has delivered another 7,500 actually autonomous paid rides.

"As far as robotaxi it does appear they are running consumer software. Or at least the software and map details are not any different than what is found in consumer cars. "
https://electrek.co/2026/05/28/tesla-fsd-safety-stats-misleading-reuters-investigation/

Rivian may find that it must add automated camera washers like BMW's iX3 has. It is straightforward enough. Though as I am sure you know from reading SAE J3016 202104, being able to operate in all conditions is not a requirement of Level 4 autonomy.

"While Level 3 and 4 ADS features/vehicles are designed to operate exclusively within their respective ODDs, some ODD conditions are subject to rapid change during on-road operation (e.g., inclement weather, obscured lane lines)."

 

[shandering]

In the two and a half hours since you typed that, Waymo has delivered another 7,500 actually autonomous paid rides.

maybe if you stopped using AI you wouldn't embarrass yourself. They are down in a lot of markets because of flash flood warnings. As I said. They don't run when there is remotely a chance of anything. They are not running in houston and haven't run in san antonio for months.

Austin, miami, orlando, houston, dallas, san antonio, atlanta etc. all have high risk of flooding. They don't run in any of these markets with anything more than light rain anymore. This can be several days at a time.

Rivian may find that it must add automated camera washers like BMW's iX3 has. It is straightforward enough. Though as I am sure you know from reading SAE J3016 202104, being able to operate in all conditions is not a requirement of Level 4 autonomy.

Tesla does not just have camera washers. They have high pressure washers for all cameras (much stronger than what is in the usual cars) and they also have an air jet that blows water and debris from every camera. Even the side cameras, repeater cameras, pillar cameras, etc.

Being able to operate in all conditions is a requirement of level 4 in the s ense that the car can never leave the ODD.

As you typed yourself. ODD conditions are subject to rapid change and you have to handle these rapid changes. Level 4 does not allow the handoff to someone in the driver's seat.


If there is a fallback ready user that is level 3, and that opens another can of worms where level 3 systems require users to pay attention to your surroundings in order to identify failures of the car not covered by the DDT. It's not a system where you can go without supervision.

https://electrek.co/2026/05/28/tesla-fsd-safety-stats-misleading-reuters-investigation/

Did you even read or understand what is happening there? Clearly not.

Tesla does all their AI data labeling in the U.S.

All these people who they talked to are data labelers. They are low skilled workers making $20 an hr in Draper, Utah and Buffalo, NY.

Tesla needed to add a lot of functions in order to get robotaxi/FSD build to something that could be unsupervised. Pickup and dropoff behaviors, emergency vehicle handling, etc.

These people reported on training on failures which doesn't exactly tell you anything about how FSD performs. Failure exist across billions of miles.

Waymo had a failure where they would blow past school buses in austin. They pushed a recall. The next month it happened 22 more times. They mentioned that you "can't expect perfection."

Imagine if you only train on clips of waymo blowing past school buses. You would think they drive like shit, too.

 

[Dave Cundiff]

Can one of you educate me on what "ODD" means in this context? Thanks!

Sponsored