The thing I thought I’d like the most about my Model 3 is the thing I like the least

I know I’m going to catch some flak for this thread whatever I do, so I will try to head off at least some of it by saying up front that I still love my Model 3 overall, think it’s better than other EVs available at the same price point for a variety of reasons, and I have no intention of swapping it for something else. I’m sure I’ll get a flood of “well if you hate it so much why don’t you sell it!!” replies despite having said this

But today I was ‘hypermiling’ down the motorway, sat behind a lorry that was doing a steady 56mph. I don’t normally try to squeeze out every last drop of efficiency like that, but my home charger is broken at the moment and rapid charging costs are ruinous, so needs must! As I did this, I became more and more annoyed that the car would frequently back off a few mph, let some distance build up to the lorry, then surge back up to my set speed (60mph), close right up to the lorry, then back off again, over and over. I find the adaptive cruise control in the Model 3 frequently does this. It doesn’t match the speed of the vehicle in front, it sort of matches the speed of the vehicle in front within a 5mph or so range, backing off and closing up again frequently. It is annoying.

I then thought back to my previous two cars that had adaptive cruise control, which were both VWs. In those cars, the speed always matched exactly to the vehicle in front. No backing off and surging forward, it was extremely precise. Why is the VW system so much better? Presumably because it uses a radar to detect the distance rather than relying on cameras.

It then occurred to me that when I bought the Tesla, I’d believed the hype about its ‘self-driving’ aids, believed it was the leader in this technology, and it was the thing I was most interested in trying out. However, I now think that in the real world, Tesla’s implementation of this is much worse than VW’s, and presumably other traditional manufacturers too. I tried the Enhanced Autopilot as well, but got my money back as, Autopark aside, it was hopeless. The thing I thought I would like most about the Model 3 is the thing I like the least.

Now, that’s ok overall. I’ve discovered over my time owning the car that there are many many things I love about this car, and overall those things make up for the crappy driver aid implementation. But I thought it worth posting my opinion in case belief that Tesla’s driver aids are more advanced than others is a reason anyone here is considering buying a Model 3. If you think that’s the car’s USP, and it is really important to you, then you will be disappointed. You’ll find loads of other amazing things about the car that you love, but you will be disappointed in the self-driving tech on UK roads in 2022.

I also know that data scientists are going to set me on fire for this thread, because I don’t understand the awesome potential of vision based systems, and that the AI models will learn and improve over time. That’s fine, but I bought a car in 2022 to work as a car in 2022. I didn’t buy the car for the joy of being part of a research project, or in the hope that in 5-10 years time it will surpass the abilities of cars that rely on radar. If you’re happy to accept lousy driver aid performance to be at the vanguard of a machine learning revolution then fine, but I just want features on my car to work now, like they did on my VWs.

But again, to repeat one last time - I prefer my Model 3 to my previous cars overall, and have no desire to get rid of it.

(Featured Image Courtesy of Tesla, Inc)

 

[AutomaticMan]

1

I have it set either to 4 or 5. Very occasionally higher. I would be very nervous at 1. I figure if people are going to cut in, they're going to cut in. I leave space and just chill with it.

It’s interesting that other people don’t have issues with setting the follow distance that far back. What I find with that is, let’s say I’m in lane 3 overtaking and the lane is moving at 75mph, while lane 2 is moving at 70mph. If my follow distance is set to 4 or 5 then cars will constantly decide they want to overtake as well and move from lane 2 into lane 3 in front of me and accelerate to 75mph. The Tesla then backs off to 70mph to build up a gap to that new car in front of me. Before the Tesla accelerates back up to 75mph, another car then pulls into the gap in front of me, and so on, and so on.

At which point, I’m not actually doing the 75mph that the lane is flowing in, or overtaking anyone. I’m constantly being forced back down to 70mph by cars pulling in front of me. So the car behind me gets annoyed, thinks I’m hogging lane 3, suddenly I have a BMW or Audi (because it’s always a BMW or Audi) up my backside flashing their lights at me.

That’s why I set it to 1 on motorways, which is still a far bigger gap than most motorists leave to the car in front.

But this thread has made me think that maybe the ACC at maintaining a constant distance to the car in front with a bigger follow distance set, so I should experiment with that when following cars in lane 1, and either set the following distance back to 1 or manually override with the throttle when I move lanes to overtake.

 

[12Pack]

The issue is that according to Musk, since humans manage with two eyes, Teslas should be able to manage with a camera suite. But humans manage with much more that two eyes - our brains have a visual cortex that uses the differences between the images our eyes see to map a 3 dimensional space, but still not to the precision of radar. I experience this very often on the tennis court….

“Schuyler Cullen, who oversaw a team that explored autonomous-driving possibilities at the South Korean tech giant Samsung, said in an interview that Mr. Musk’s cameras-only approach was fundamentally flawed. “Cameras are not eyes! Pixels are not retinal ganglia! The F.S.D. computer is nothing like the visual cortex!” said Mr. Cullen” ….NY Times

 

[GRiLLA]

I also don't think it's wild speculation to say that a regulator could rule Autopilot to be unsafe and tell Tesla to fix it. It sounds like they're already under investigation so that must be a possible outcome of the investigation.

In the US, not here in the UK or Europe. It's really how the US environment of low regulation works, companies like Tesla can do what they want until the regulator (which is quasi-political) decides they need to do something. There are dozens of investigations happening into different cars. Tesla has been pretty much constantly in a state of investigation for the last decade.

 

[init6]

but still not to the precision of radar.

The current radar sensors (on cars that have them) are very low resolution and infer the positions of object. Since all the computing is done in the sensor, there's not a lot Tesla's AI or computing can do to mitigate any errors. I guess this is why Tesla are creating their own hi-res radar sensors. Vision may only be a stepping stone to Vision + Hi Res Radar.
To quote Musk almost a year ago:

A very high resolution radar would be better than pure vision, but such a radar does not exist. I mean vision with high res radar would be better than pure vision.

 

[MrT3]

it does even on 50 kw charger.

your last 10% drips at ~20 kw, while last 5% are at ~15 kw. it even manages to bottleneck the 50 kw charger tho

I'm not saying the speed doesn't slow down, just that the time to get from 20% to 80% isn't that fast in the first place because it's only a 50 kW charger.

 

[MrT3]

The issue is that according to Musk, since humans manage with two eyes, Teslas should be able to manage with a camera suite.

and I thought they were trying to make autonomous driving better/safer compared to humans. Therefore why limit the system to just vision, which has the same flaws as our eyes - we can't see when the sun is shining directly at us, or in fog and neither can their vision system.

 

[M JR]

Had my MY since May 2022 and have already experienced phantom breaking. Have taken it on two long trips and experienced it on both. Its a bit unnerving. Both occasions had relatives in the car who were interested to experience a Tesla. Needless to say they were not impressed. I still love the car but hope this issue gets fixed soon. Unfortunately I dont see that happening...at least for a while. If it was a simple fix it would have been done a long time ago. Tesla's approach to FSD (vision only using AGI) certainly is the issue. I think the approach will win in the end but its going to be a long and bumpy road. Hopefully the end result will be worth it. Otherwise...they are going to have major issues.

 

[M JR]

and I thought they were trying to make autonomous driving better/safer compared to humans. Therefore why limit the system to just vision, which has the same flaws as our eyes - we can't see when the sun is shining directly at us, or in fog and neither can their vision system.

To be fair...we dont have eyes in the back or sides of our heads nor do we have sonar (and we aren't always attentive). But I get what you are saying. I think in the end you want the least amount of tech as possible and still achieve the level. Will it work? Not sure but I hope it does.

 

[_Redshift_]

The issue is that according to Musk, since humans manage with two eyes, Teslas should be able to manage with a camera suite. But humans manage with much more that two eyes - our brains have a visual cortex that uses the differences between the images our eyes see to map a 3 dimensional space, but still not to the precision of radar. I experience this very often on the tennis court….

“Schuyler Cullen, who oversaw a team that explored autonomous-driving possibilities at the South Korean tech giant Samsung, said in an interview that Mr. Musk’s cameras-only approach was fundamentally flawed. “Cameras are not eyes! Pixels are not retinal ganglia! The F.S.D. computer is nothing like the visual cortex!” said Mr. Cullen” ….NY Times

When traveling to and from Santa Barbara which is around 290 miles from my home last weekend, I realized a couple of things:

My eyes see further out into the distance than Tesla’s cameras. That is, I can see flashing emergency vehicles way out into the distance and prepare to slow down the vehicle accordingly. Tesla will see only a few car lengths ahead and be jerky and brake hard.

I can see the vehicles merging into the right most lane when I am traveling there and start to slow down to accommodate them smoothly. Tesla behaves as if it has not seen the merging vehicle until it’s directly beside it, and proceeds to brake hard to let the vehicle in.

Both of the above scenarios happened to me, the latter one, multiple times.

All this ‘humans do well with eyes, so we will too’ is not really borne out in the real world.

 

[yessuz]

When traveling to and from Santa Barbara which is around 290 miles from my home last weekend, I realized a couple of things:

My eyes see further out into the distance than Tesla’s cameras. That is, I can see flashing emergency vehicles way out into the distance and prepare to slow down the vehicle accordingly. Tesla will see only a few car lengths ahead and be jerky and brake hard.

I can see the vehicles merging into the right most lane when I am traveling there and start to slow down to accommodate them smoothly. Tesla behaves as if it has not seen the merging vehicle until it’s directly beside it, and proceeds to brake hard to let the vehicle in.

Both of the above scenarios happened to me, the latter one, multiple times.

All this ‘humans do well with eyes, so we will too’ is not really borne out in the real world.

Sorry, but couldn't resist:

Did you meet C.C. Capwell??

 

[AutomaticMan]

Well I don’t know what to believe about anything now I was skeptical further up the thread about the driver assistance aids performing differently in two different Model 3s. But, I’m currently sat in a 2019 Standard Range courtesy car* and, well, they do seem to!

First was a negative difference. Using Autopilot in queueing traffic is worse. Braking to a stop is so much harsher in this. Maybe it has to use the friction brake more due to not having motors on the front wheels and that’s why?

But then, a positive difference! This car has Enhanced Autopilot (EAP)** and FSD Beta. I tried out EAP in my Model 3 when I got it in Feb, and the auto lane changes were so unreliably horrible that I got my money back. But I just did a little jaunt down the motorway in this and the car did the lane changes itself reliably every time. I also just reacquainted myself with Autopark which works great (though it did when I tried it with my own Model 3 in Feb as well).

Now I’m not sure what to do. If Auto Lane Change worked as well as this when I tried it before I wouldn’t have asked for a refund. What’s going on?? Is this car just different/better at it? Did Tesla improve it vastly with software updates? Is there something about having EAP and FSD Beta that improves the lane changes? Does the bright, sunny weather make it work better? It was probably gloomier when I tried it in Feb! Was it just luck that it worked well this time, and it would still annoy me if I used it for a longer period of time?

I’m honestly wondering whether I should buy EAP for my Model 3 again, but I can’t get another refund if I do so if it’s pants again on my car then I’m stuck with it this time.

———
*My car has a clicking sound coming from the front right suspension when the steering is turned to full left lock. Searching this forum makes me think it might be the CV joint. So it’s into a service centre for them to investigate.

**Note to any Americans reading, EAP and FSD are still two separate things in the UK.

 

[_Redshift_]

Sorry, but couldn't resist:

Did you meet C.C. Capwell??

He had insisted I visit him, told him my schedule was way too busy, as I had to visit an actual Hollywood A lister.

 

[yessuz]

He had insisted I visit him, told him my schedule was way too busy, as I had to visit an actual Hollywood A lister.

Nah. You had a chance to pay a visit to Kelly though

 

[Ferret]

With regards to the follow distance, I generally use 4 and yes cars move in to the gap, but it doesn't really inconvenience me as the car just carries on doing its thing. It's not like I personally have to brake etc. When I set to 1 I find it harsh on the brakes.

 

[John Nemesh]

Well, I wake up in the morning, jump in my car, and once out of the driveway, engage FSD (beta), and it takes me, with ZERO INTERVENTIONS OR DISENGAGEMENTS 55 miles to work and stops in my work parking lot. When you find another car that can accomplish this, RIGHT NOW, let me know...

 

[ne0x]

I have noticed that when approaching on ramps and off ramps on autopilot the car slows to create a gap for what I can only think would be more visibility for auto exit or merging traffic.
Anyone else notice this?

 

[OliverM3]

I do prefer to drive myself but when I do use auto pilot I find it better in some aspects and worse in others. For me honestly I don't care if I wanted someone else to drive I wouldn't be in my car. I think it handles the lane very well but has to much power for a smooth follow distance. Putting it in chill mode is much better if that's what you actually want to do and not be so aggressive with acceleration and braking. Again I'm very impressed with how it keeps me in a lane.
My Audi felt like there were bumpers and I would bounce bounce bounce off of the bumpers for 100 miles on a trip. It was so annoying I never used it because it couldn't see the lane until it hit the boundary and then bumped you back in.

 

[DrChaos]

I know I’m going to catch some flak for this thread whatever I do, so I will try to head off at least some of it by saying up front that I still love my Model 3 overall, think it’s better than other EVs available at the same price point for a variety of reasons, and I have no intention of swapping it for something else. I’m sure I’ll get a flood of “well if you hate it so much why don’t you sell it!!” replies despite having said this

But today I was ‘hypermiling’ down the motorway, sat behind a lorry that was doing a steady 56mph. I don’t normally try to squeeze out every last drop of efficiency like that, but my home charger is broken at the moment and rapid charging costs are ruinous, so needs must! As I did this, I became more and more annoyed that the car would frequently back off a few mph, let some distance build up to the lorry, then surge back up to my set speed (60mph), close right up to the lorry, then back off again, over and over. I find the adaptive cruise control in the Model 3 frequently does this. It doesn’t match the speed of the vehicle in front, it sort of matches the speed of the vehicle in front within a 5mph or so range, backing off and closing up again frequently. It is annoying.

I then thought back to my previous two cars that had adaptive cruise control, which were both VWs. In those cars, the speed always matched exactly to the vehicle in front. No backing off and surging forward, it was extremely precise. Why is the VW system so much better? Presumably because it uses a radar to detect the distance rather than relying on cameras.

It then occurred to me that when I bought the Tesla, I’d believed the hype about its ‘self-driving’ aids, believed it was the leader in this technology, and it was the thing I was most interested in trying out. However, I now think that in the real world, Tesla’s implementation of this is much worse than VW’s, and presumably other traditional manufacturers too. I tried the Enhanced Autopilot as well, but got my money back as, Autopark aside, it was hopeless. The thing I thought I would like most about the Model 3 is the thing I like the least.

Now, that’s ok overall. I’ve discovered over my time owning the car that there are many many things I love about this car, and overall those things make up for the crappy driver aid implementation. But I thought it worth posting my opinion in case belief that Tesla’s driver aids are more advanced than others is a reason anyone here is considering buying a Model 3. If you think that’s the car’s USP, and it is really important to you, then you will be disappointed. You’ll find loads of other amazing things about the car that you love, but you will be disappointed in the self-driving tech on UK roads in 2022.

I also know that data scientists are going to set me on fire for this thread, because I don’t understand the awesome potential of vision based systems, and that the AI models will learn and improve over time. That’s fine, but I bought a car in 2022 to work as a car in 2022. I didn’t buy the car for the joy of being part of a research project, or in the hope that in 5-10 years time it will surpass the abilities of cars that rely on radar. If you’re happy to accept lousy driver aid performance to be at the vanguard of a machine learning revolution then fine, but I just want features on my car to work now, like they did on my VWs.

But again, to repeat one last time - I prefer my Model 3 to my previous cars overall, and have no desire to get rid of it.

View attachment 819366
(Featured Image Courtesy of Tesla, Inc)

I have the feeling the ADAS code performs much worse in EU than it is in USA. They will be training different neural networks.

I'm using AP in its native environment (California freeways) and it works great and drives idiomatically and intuitively for me, substantially better than the TACC in my previous BMW i3.

 

[DrChaos]

We don't know if they will use the extra resolution or not. Potentially the camera itself could downsample to reduce noise, or the camera spec is better in some other way such as low light performance or higher dynamic range so it isn't affected by sunlight quite so much etc.

It could be tough and require some clever programming and approximations without needing a bigger processor. One thing they could do is to use the high resolution only in a virtual "fovea", in the predicted direction ahead on the road and downsample elsewhere.

But yes I think to get quality autonomy they will need multiple higher resolution cameras, physically separated with overlapping fields of vision to obtain direct hardware parallax, plus high-resolution imaging radar.

Contrary to what Elon says there are many radar startups with good resolution 77 GHz 'imaging' radar. Not quite lidar but close enough.

 

[AndrewZ]

^
These accounts bare virtually no resemblance to how we are using and finding AP in our 2017 X.

We did 200 miles of M1/M40 this weekend, AP handled 90% of M-way driving, 90%+ of lane changes, and to be 100% honest I recon if the EU regulations allowed the car to change lanes without human confirmation it I would have been 'happy' to be 100% hand off. I've no experienced any aborted lane changes for about 3 months now, and equally cannot remember the last episode of PB, though they do still occur more commonly than lane change aborts.

We can use the 2017 Model 3 for comparison since it will have much larger numbers out there than the X, but just compare the NHTSA pages for the model years

2017 Model 3 - 7 owner complaints

2017 TESLA MODEL 3 | NHTSA

www.nhtsa.gov

2021 Model 3 - 328 owner complaints

2021 TESLA MODEL 3 4 DR RWD | NHTSA

www.nhtsa.gov

Almost all of the 2021 complaints are attributable to Autopilot and the auto wipers/lights

Of course many more 2021s were produced than 2017s, but it‘s a massive disparity that I don’t think can be explained by production numbers alone.