Out of Spec roadtrips Gravity and DDP2 has some issues


Kyle takes his long term Gravity loaner on a road trip and had some issues with handsfree DDP2. His experience is very similar to mine from a recent road trip. Off and on zig-zag behavior, sudden disengagements and occasional sh*t the bed on lane keeping. I almost think things have gone backwards from old hands-on DDP2. To the point where I now drive with hand on wheel so I am ready for sudden disengagements or other issues. Can’t wait for NVIDIA system to come on board as I do not have a whole lot of confidence in current system.
This video is a good summary of the Lucid experience
 
And lest we forget Tesla is NOT hands free it’s L2 assisted driving.
Right there front page:

You mean those videos that you linked to that show the car driving hands-free repeatedly?

Imagine my surprise that it isn’t hands free while I was driving over 2k miles this summer without touching the steering wheel, also known as…..hands free.

Or did you mean that I actually still have hands, so technically I am not hands-free?

It’s obvious you will give Tesla zero credit for being light years ahead of Lucid on this. Not sure why you are taking it so personally but there are cybercabs running all over Austin that are autonomous - completely hands-free, regardless of what you think.
 
You mean those videos that you linked to that show the car driving hands-free repeatedly?

Imagine my surprise that it isn’t hands free while I was driving over 2k miles this summer without touching the steering wheel, also known as…..hands free.

Or did you mean that I actually still have hands, so technically I am not hands-free?

It’s obvious you will give Tesla zero credit for being light years ahead of Lucid on this. Not sure why you are taking it so personally but there are cybercabs running all over Austin that are autonomous - completely hands-free, regardless of what you think.
You mean why do I point out it’s LEVEL 2 SUPERVISED so why do Tesla fanboiz keep ignoring the supervised part. I mean I can make a video of me dancing in the Lucid while self driving so what. I’ve repeatedly said fsd works but it’s not autonomous, if it is then get in the back seat with no one in the drivers seat. Show me.
 
I tested the iX3 and it nailed the interior : exterior packaging. But damn these iX BMW's are simply too hard to look at and I never bought a car because it looked good or not bought a car because it was ugly. I've owned the original Honda Ridgeline and a Chevy Bolt.
 
I tested the iX3 and it nailed the interior : exterior packaging. But damn these iX BMW's are simply too hard to look at and I never bought a car because it looked good or not bought a car because it was ugly. I've owned the original Honda Ridgeline and a Chevy Bolt.
I know... I'm with you there. The iX (being discontinued now) was a no go for me because I couldn't get past the looks. The new iX3 and iX5 look much better, but still hard to swallow.
 
I know... I'm with you there. The iX (being discontinued now) was a no go for me because I couldn't get past the looks. The new iX3 and iX5 look much better, but still hard to swallow.
I went to test drive the OG i3 some years ago,. On approach, I saw it, did a u-turn, and went home instead.
Funny though, I saw one again a few days ago, and it wasn't as much of a visual insult as are the current iX/iX3/iX5.
For this reason, and for native Google Maps / Gemini, I'm looking at the new Volvo EX60 instead.
 
I went to test drive the OG i3 some years ago,. On approach, I saw it, did a u-turn, and went home instead.
Funny though, I saw one again a few days ago, and it wasn't as much of a visual insult as are the current iX/iX3/iX5.
For this reason, and for native Google Maps / Gemini, I'm looking at the new Volvo EX60 instead.
True. I do have an OG i3, kept it because my wife loves it. The looks were very polarizing at the time, but now it looks sort of normal. Or maybe I just got used to looking at it 🙂
 
You do realize FSD is being used for 15 billlion miles with about 1 billion miles every month? 100% of these miles are recorded with video. Of course you will see errors. Secondly, not everyone is on HW4. A lot of the bad FSD stuff is HW3, and not every bad thing you see HW4 do is going to lead to an accident.

The nvidia thor chips are weak where it counts. Memory bandwidth. Tesla actually has very high memory bandwidth compared to nvidia, and the more sensors you have, the more memory bandwidth you need. Not to mention AI5 has over 2000 tops for just one chip, and in 2 of them t hat absolutely smokes nvidia thor.

Anyone that is not tesla running a robotaxi, is using effectively server grade chips that consume 1000s of watts. They have terabytes of memory bandwidth and they have ridiculous amounts of memory as well. As you can imagine it's not possible to put these in cars without reducing your range by 50% or more

Tesla is the only company that has demonstrated that it is possible to operate without HD maps and with consumer level compute.


This is false. Tesla safety with robotaxi is very good. The only issue is whether not having lidar or the placement of their cameras is an issue scaling to 100-200K vehicles or a larger ODD. At that point, yes they could add lidar, different camera positions, etc. but that would not set them back as nobody else has a technology that scales to that many vehicles.

Tesla would never end up behind, because that would imply somebody passes them, which is clearly not the case, or that anyone is close to doing it

Even if they added more sensors, they would drive 99.99% like they already do and use the sensors to fill in missing areas, which would not take years and/or set them back at all.


This is 100% false and you need to stop believing the nonsense you read on the internet. Cybercab was literally 99% FMVSS compliant. The only "missing" feature was is the foot activated brake control. Turns out the internet was wrong and cybercab does have a foot activated brake control.

There's a button underneath the carpet that you can very clearly hear in one spot of the car if you press on it

The zoox story is different. Zoox self-certified their car as compliant like tesla did. They used a BS excuse that they could design alternative test procedures for missing components of the car (because they did not apply). Therefore zoox self-certified a non compliant car. Zoox was penalized because they were told that they could not use alternative test procedures.

Tesla did not do this. they built a 100% FMVSS compliant car by including every function required under the FMVSS. Turns out FMVSS does not require a steering wheel or accelerator pedla, and only a foot activated brake control is required. They complied with this by putting a hidden button underneath the carpet.

There are other required functions like wiper control, turn signals, lights, shift indicator, speedometer etc. but that stuff is all right there on the center screen

I'll address everything you've said here:

(1) Yes, Teslas put a lot of miles on FSD every day. Daily safety issues are still not ok in that context.

(2) There is more nuance to the story then outright memory bandwidth. For one, Nvidia's chips compress their data allowing them to achieve similar/higher throughput to AI4. Second, as everyone has moved towards VLA models, memory capacity has become a critical factor and Tesla's newest AI4's 32GB is significantly less than Thor's 128GB. As you've stated yourself, Thor is much more powerful in compute than AI4. AI5 hasn't been released yet, so not sure how relevant that is. Nvidia is no doubt working on Thor's successor, but that is also not relevant.

(3) Not much to say here other than it's laughable that you think using Nvidia's chips cuts range down by 50%.

(4) Adding additional sensors is not like flipping a switch. It takes years to develop the hardware and would take years for their end-to-end model to learn how to make decisions using the additional sensor data. The models performance would decline materially upon launch.

(5) You're not correct here. Their strategy has always been to release a non-compliant car in hopes that it'll push the NHTSA to update it's dated regulations. Their non-compliance is not just the foot activated brake control. It's also that the brakes are strictly software-based with no mechanical fallback, it's lack of mirrors (rearview and exterior), lack of driver controls (which are not on the screen as you say), etc. Zoox actually applied for an exemption, so everything they've done is by the book. Tesla chose not to do that.
 
I just want to add that the purpose of my original comment was not to say that FSD isn't good. I clearly said I thought it was good. I also wasn't trying to compare FSD to its competition.

My point was just that FSD makes enough critical mistakes to indicate that it is not ready for unsupervised driving, and that we DO have enough compute to have more sensors than just cameras in consumer vehicles.
 
Please, for the love all that is good, can we keep this thread on topic, and not have yet another pissing contest about FSD?

Seriously, it’s extremely annoying that we keep having to stop folks from hijacking every thread on a Lucid forum to make it about Tesla.

Enough already. We get it. One guy loves Tesla. Another hates Tesla. It’s so boring at this point. Nothing ever learned. No minds ever changed.
 
To me, I'm not that interested in autonomous driving tech until it's so good that it doesn't need supervision. And I think that autonomous driving that works 99.99% of the time is worse than a system that works 99% of the time. The 99% system fails frequently enough that the driver is likely to stay alert and intervene early. The 99.99% system risks lulling drivers into complacency and not supervising at all. But even if you try and stay alert, you're fighting the way human attention works when you try to supervise a system that rarely fails.

Until full autonomy becomes available, I pretty much just want the features that are easy to make highly reliable, such as adaptive cruise control. And features that might save your life when you screw up while driving, like lane departure warnings, antilock brakes, automatic emergency braking, etc.

My mother died in a car crash, and I could be more concerned about the safety issues around vehicles than most people. To me, Lucid is making the same error that Tesla did. They're starting with unreliably autonomy, and planning to steadily improve it as time goes by. I believe Waymo's approach is much better, in that they decided early that they needed to be building a system that didn't expect a human to intervene, and work on the hardest part of the problem right from the start.
I've been saying this for 5 years. When it's perfect enough to not need a steering wheel, then fine. I'll happily ride in those robotaxis. But supervised means it's not good enough to trust it to steer itself. And "supervision" is usually a joke. To properly supervise self-steering, you have to be able to react just as fast as you would if you were actively steering. That means hands on the wheel and watching the road every second. We all know most people aren't doing that. Which means delayed reaction when something goes wrong. Maybe 1 second, maybe 2, maybe 5. A lot of bad things can happen at high speed in a second or two.

It's the worst of both worlds. Not good enough to not need human intervention combined with delayed human intervention.
 
There is more nuance to the story then outright memory bandwidth. For one, Nvidia's chips compress their data allowing them to achieve similar/higher throughput to AI4. Second, as everyone has moved towards VLA models, memory capacity has become a critical factor and Tesla's newest AI4's 32GB is significantly less than Thor's 128GB. As you've stated yourself, Thor is much more powerful in compute than AI4. AI5 hasn't been released yet, so not sure how relevant that is. Nvidia is no doubt working on Thor's successor, but that is also not relevant.
VLA models are not safe. VLA models are slow which gives them poor reaction time. That's the opposite of safe. Basically it is unknown whether you can really drive using a VLA model because, yes you can have decent looking driving, but safety is unknown. Look at the comparisons of reaction time between chinese systems and FSD. It's night and day. Not even close.

Secondly, tesla does not have 32gb of memory. In the cars they are using HW4.5 which is at least 48gb. They also have 50%+ higher memory bandwidth (the real limitation of real time AI) because they are using GDDR6x memory.

The new AI5 computer has 72gb of memory per chip and there are 12 modules per chip


look how fast this reaction and braking is here.

Yes, Teslas put a lot of miles on FSD every day. Daily safety issues are still not ok in that context.
Of course it is. Tesla can even at-fault kill people with FSD (they haven't), and to some extent it is unavoidable with 15 billion miles.

Not much to say here other than it's laughable that you think using Nvidia's chips cuts range down by 50%.
I'm not talking ab out nvidia thor. I'm talking about the chips used in waymo and the like. These are not using LPDDR5 memory. They are using high bandwidth server memory, 100s of thousands of tops, 4000w, etc.

These are not like what you have with nivida thor. At one point it was rumored waymo was using 4 H100s. These have memory bandwidth and compute well beyond what tesla/nvidia were using.

Adding additional sensors is not like flipping a switch. It takes years to develop the hardware and would take years for their end-to-end model to learn how to make decisions using the additional sensor data. The models performance would decline materially upon launch.

Not for tesla. You assume that they will drive materially any different with lidar. They won't. They'll drive exactly the same as they do for 99.99% of scenarios and maybe use lidar to see cross traffic or something like that. Keep in mind FSD has the best accident avoidance in the world with camera only. It's only going to get better. If this is the worst robotaxi will ever perform, that is saying something.

Their strategy has always been to release a non-compliant car in hopes that it'll push the NHTSA to update it's dated regulations. Their non-compliance is not just the foot activated brake control. It's also that the brakes are strictly software-based with no mechanical fallback, it's lack of mirrors (rearview and exterior), lack of driver controls (which are not on the screen as you say), etc.

False. FMVSS does not require a steering wheel. it literally says so in the FMVSS that if no steering wheel is present you need passenger-seat crash protection. They have that. FMVSS does not require an accelerator pedal. Just foot activated brake control.

There is a foot activated brake control and therefore the brakes are not strictly software based as you claim. There is also an emergency stop button on the ceiling.

Mirrors are not required or vehicles like the polestar 4, it is purely digital

All of the functions of the car are on the screen. Manually cancellable turn signals, light controls, shift display, speedometer, etc.

you can see it right there while the car is driving

Zoox actually applied for an exemption, so everything they've done is by the book. Tesla chose not to do that.
This is false. Zoox actually claimed their vehicles was FMVSS complaint. They used a ruling from waymo to assume that they could design alternative test procedures for non-applicable functions of their car. The NHTSA told them t hey couldn't do that and then they needed to apply for an exemption. Zoox was illegally operating the car for a while.

Tesla, on the other hand, just built a car that satisfies NHTSA requirements. Zoox needs too many exemptions. Tesla does not, because the way their self driving software works pretty much requires it to be a normal looking car with windshield wipers, defogging, etc.
 
Last edited:
Please, for the love all that is good, can we keep this thread on topic, and not have yet another pissing contest about FSD?

Seriously, it’s extremely annoying that we keep having to stop folks from hijacking every thread on a Lucid forum to make it about Tesla.

Enough already. We get it. One guy loves Tesla. Another hates Tesla. It’s so boring at this point. Nothing ever learned. No minds ever changed.
A lot of people with too much free time on their hands. On the other hands; if people are criticizing you a lot; it most likely means you are doing something right. Lucid must be doing something right despite all the doomers claiming otherwise.
 
True. I do have an OG i3, kept it because my wife loves it. The looks were very polarizing at the time, but now it looks sort of normal. Or maybe I just got used to looking at it 🙂
I think you maybe became used to ugly.
 
We
A lot of people with too much free time on their hands. On the other hands; if people are criticizing you a lot; it most likely means you are doing something right. Lucid must be doing something right despite all the doomers claiming otherwise.
to be honest if lots of people criticize you it is possible that they are on to something
 
With all respect for those with DD Pro I have to share my experience. I am about to order a Gravity GT with DD2 Premium and Dynamic Handling. DD2 Premium has what i need. I have four years of experience with DD Pro in my AGTs. I have every confidence that Lucid will continue to improve DD Pro's capabilities, but until it perfects the system I will pass on it. It was good on open stretches of interstates but not confidence inspiring with tractor trailer trucks a few feet away. Its red warnings were annoying, especially because I always had my hands on the wheel at 9 and 3 in hands free mode. Anything less than 100% is not good enough for me.
 
Back
Top