Rivian AEB failure and crash

[s4wrxttcs]

The Rivian Gen 1 use Mobileeye EyeQ4 Mid for for the AEB. As far as I can tell it cannot leverage the Cameras around the vehicle for AEB.

This is also why the dashboard only renders things in front of the vehicle.

Your accident is really only relevant to Gen 1 vehicles.

Sponsored

 

[s4wrxttcs]

Rivian sensors and software are slow in my experience with both Gen1 and Gen2 R1. The rear cross traffic alert often misses people and vehicles and when it does warn me, it is almost always after I've already seen the person or vehicle.

RCTA on my previous 10+ year old Toyota vehicles made it almost seem like the vehicle had a sixth sense. With the Rivian it is the exact opposite, despite them being much newer vehicles. I feel like decentralized compute actually has some real advantages when it comes to latency in safety-based applications.

AEB has been a similar story, with frequent false positives and I can only hope I don't run into a false negative like OP.

I can understand Rivian being behind on ADAS vs Tesla, but not getting RCTA and AEB right (compared to 10+ year old ICE vehicles) make me treat it like a vehicle from 20 years ago, when humans controlled almost everything. Not that there's anything wrong with that.

It's strange how some people have entire different experiences.

With my Gen1 Rivian

No AEB braking - This is extremely tense braking that I've only experienced once in my life with a Jeep Wrangler going into a parking garage. The Jeep thought it was going to hit the wall and slammed on the brakes.

Immediate rear cross traffic alert - In fact this is only second to the 360 degree down facing view in terms of my most liked features. I've never seen it not work.

What hasn't worked well.

The times I have experienced phantom braking is with adaptive cruise control. But, that's much more mild. It's been somewhat annoying in my Rivian but nothing like certain updates with my Tesla back in the 2018-2020 years.

Front Ultra Sonic Sensors - Even with the tiniest amount of filth the car will complain about the front sensors being dirty. It's as if the car has gotten OCD about wanting to be clean.

 

[mkhuffman]

Because the cars drive in a selfish way. For example, they will drive at 70 in the middle lane leaving 10 car lengths between them and the next car all while people are trying to get around.

Or, it’s rush hour, and the Tesla is driving at 10mph with 10 car lengths between it and the next car, making it very difficult for people to get around it.

Classic example: someone is trying to pull out of a driveway, but a Tesla is on autopilot and blocks it while the Tesla waits at a light. It doesn’t stop a little early to give the person space to pull out. Meanwhile the Tesla driver just stares at the frustrated person instead of taking over and giving space.

That’s why people in the Bay Area hate teslas. They don’t drive with others in mind. They drive with only themselves in mind, annoying everyone around them.

This is the opposite of my experience in a HW4 MY. Maybe you are stuck in the past, but the latest FSD on HW4 is perfect. And I am not just saying that to be difficult. It is my experience riding in my friend's car for an entire weekend, with multiple complex driving situations.

It does not leave too much space in the front. It does not drive at 10 mph unless that is required because of traffic. It does not block anything. It makes measured, smart and quick decisions that are exactly what I would have done. But it does it better and smoother than I could do it.

Some people might not appreciate how it manages the car in traffic. What I mean is it keeps up and is not a impediment to faster traffic. It keeps with the flow, even if the flow is well over the speed limit. As soon as it is not an impediment, it moves over and slows back to what you set it for. Perfect. Really, it is.

As I posted before, it shocked me how good it is. And I was watching what it did very, very carefully. I was watching for a mistake or it being too cautious. It was the best driver I have ever ridden with. Ever.

My guess is you have never ridden in a HW4 FSD Tesla. Right?

 

[SwampNut]

Weird, I've been able to transfer and post online many clips from my R1S without encountering "hot garbage"

Me too. I don't know what exposes the shitty encoding, which is around 10-20% of the time. It seems to be, that the more important it is, the higher the chance that nothing will want to read the files. It's a well known issue, a few people have written scripts to fix it.




Rivian Gear Guard / camera exports trip up players for a few overlapping reasons:

Codec. Rivian records HEVC (H.265), often Main10/10-bit. Anything without hardware HEVC decode (older Windows boxes, some browsers, default Windows Media Player) either refuses it or stutters. Your M4 MacBooks decode HEVC in hardware, so if those are choking, it's not the codec — it's one of the next two.


Container / moov atom. Automotive recorders frequently write a non-finalized or fragmented MP4. If a clip was cut by power loss or sentry-event boundaries, the moov atom (the index) can be missing, misplaced, or pointing at truncated data. The file's bytes are fine but players can't seek/parse it, so it shows a broken thumbnail or won't open.


Variable frame rate. The cameras record VFR, which makes timestamp-strict players stutter or drift A/V sync, even when the stream itself is valid.


Fix order — try the cheap one first:




bash
# Remux only: rewrites the index/moov, no re-encode, near-instant, lossless
ffmpeg -i input.mp4 -c copy -movflags +faststart output.mp4

That alone resolves most "won't play / won't seek" cases. If the file is genuinely VFR-stuttering or a player still won't decode HEVC, force a constant-frame-rate H.264 transcode:




bash
ffmpeg -i input.mp4 -vf "fps=30" -c:v libx264 -crf 20 -preset fast -c:a aac output.mp4

If even the remux fails with a moov/atom error, the source clip is truncated — recover what's there with:




bash
ffmpeg -err_detect ignore_err -i input.mp4 -c copy output.mp4

 

[shandering]

It would be interesting to see how much less traffic there would be if everyone has a decent driver assist. Not even FSD level but something that can reduce traffic ripple. The biggest issue is you need to execute proper zipper merging and not sure even FSD would get it 100% right

 

[Hereforthesnacks]

This is the opposite of my experience in a HW4 MY. Maybe you are stuck in the past, but the latest FSD on HW4 is perfect. And I am not just saying that to be difficult. It is my experience riding in my friend's car for an entire weekend, with multiple complex driving situations.

It does not leave too much space in the front. It does not drive at 10 mph unless that is required because of traffic. It does not block anything. It makes measured, smart and quick decisions that are exactly what I would have done. But it does it better and smoother than I could do it.

Some people might not appreciate how it manages the car in traffic. What I mean is it keeps up and is not a impediment to faster traffic. It keeps with the flow, even if the flow is well over the speed limit. As soon as it is not an impediment, it moves over and slows back to what you set it for. Perfect. Really, it is.

As I posted before, it shocked me how good it is. And I was watching what it did very, very carefully. I was watching for a mistake or it being too cautious. It was the best driver I have ever ridden with. Ever.

My guess is you have never ridden in a HW4 FSD Tesla. Right?

I have ridden in 4 different 2026 MYs - two performance and two premium. With FSD. (I have ridden in probably 50 M3s and MY over the past couple of years. A lot through uber and, even though drivers are not really supposed to use FSD, some will just to test it out after telling me. It kills me that uber always gets me a Tesla because the ride is not good, though it’s better on the new models). I have never driven a H4 car though.

I can say the HW4 is not aware of those around it in such a situation: someone is signaling that they need get across lanes. Even though it’s not a hardship for the Tesla, it will not let them by. This frustrates the other diver and slows traffic as the other driver needs to maneuver around the Tesla. It’s just now how FSD is built. FSD is designed to drive the car around what’s in and around the road. It does not drive the car with what others may need - out of curtesy - in mind.

 

[2kwik4u]

This should do it...

Crash sucks. Sorry man. Glad everyone was safe.

FYI, gotta plug that buffer in before it'll do anything.

 

[mkhuffman]

...
I can say the HW4 is not aware of those around it in such a situation: someone is signaling that they need get across lanes. Even though it’s not a hardship for the Tesla, it will not let them by. This frustrates the other diver and slows traffic as the other driver needs to maneuver around the Tesla. It’s just now how FSD is built. FSD is designed to drive the car around what’s in and around the road. It does not drive the car with what others may need - out of curtesy - in mind.

I didn't see that situation in my weekend trip, but the Tesla does move out of the way when it detects faster traffic approaching from the rear. So, it seems like it would also detect the situation you described, but I didn't experience that. Maybe because it was always moving over for faster traffic to pass.

It will accelerate to get past a vehicle if faster traffic is approaching, even if that means exceeding the speed limitation you have specified. As I mentioned before, some people might not like that. I think it is exactly how people should drive. Blocking the left lane is a huge pet peeve of mine, and this Tesla never blocked the left lane. It was perfect.

Edit: one thing it does do is wait before moving over to the right lane to exit if there is traffic it is passing in that lane. We talked about that in the car (there were four of us). The drive monitor commented that he would get over earlier, and his wife said she would not. It was a funny discussion. I can see some drivers getting irritated because a Telsa is trying to move over in front of them into the right lane to exit. But it never did it dangerously and it only changed lanes when it was safe. Exactly like it should. Was that perfect? I prefer to wait before changing lanes also. My wife always gives me a hard time about it. "Our exit is coming up, get over!" So yes, IMO it was perfect.

 

[Indy avocado]

Me too. I don't know what exposes the shitty encoding, which is around 10-20% of the time. It seems to be, that the more important it is, the higher the chance that nothing will want to read the files. It's a well known issue, a few people have written scripts to fix it.




Rivian Gear Guard / camera exports trip up players for a few overlapping reasons:

Codec. Rivian records HEVC (H.265), often Main10/10-bit. Anything without hardware HEVC decode (older Windows boxes, some browsers, default Windows Media Player) either refuses it or stutters. Your M4 MacBooks decode HEVC in hardware, so if those are choking, it's not the codec — it's one of the next two.


Container / moov atom. Automotive recorders frequently write a non-finalized or fragmented MP4. If a clip was cut by power loss or sentry-event boundaries, the moov atom (the index) can be missing, misplaced, or pointing at truncated data. The file's bytes are fine but players can't seek/parse it, so it shows a broken thumbnail or won't open.


Variable frame rate. The cameras record VFR, which makes timestamp-strict players stutter or drift A/V sync, even when the stream itself is valid.


Fix order — try the cheap one first:




bash
# Remux only: rewrites the index/moov, no re-encode, near-instant, lossless
ffmpeg -i input.mp4 -c copy -movflags +faststart output.mp4

That alone resolves most "won't play / won't seek" cases. If the file is genuinely VFR-stuttering or a player still won't decode HEVC, force a constant-frame-rate H.264 transcode:




bash
ffmpeg -i input.mp4 -vf "fps=30" -c:v libx264 -crf 20 -preset fast -c:a aac output.mp4

If even the remux fails with a moov/atom error, the source clip is truncated — recover what's there with:




bash
ffmpeg -err_detect ignore_err -i input.mp4 -c copy output.mp4

Can you get the video to playback at all? If so, consider the stupid option of a screen recording of the video playing back.

Sponsored