AP slow to cotton on.....

[Jez_GB]

I've noticed (in the last few weeks) that AP is slow to accelerate once the car in front of you moves out of your lane or pulls away/increases speed. Both the delay and rate of acceleration once the car does respond are both noticeably and annoyingly slow - to the point where I now override and accelerate up to the set speed to stop impatient drivers behind me from tailgating.

Running 2021.24.5

Anyone?

 

[kelvin 660]

Snap!

 

[Dilly]

Marginal improvement in 32.10 but not exciting by any means

 

[Jez_GB]

Found a similar thread on the main forum…. Not just us then

Slow Acceleration on highway

I recently went on my first extended road trip in some time and found that autopilot was accelerating very slowly during traffic speed ebbs and flows. When traffic slowed, autopilot slowed appropriately behind the lead car. Once the lead car accelerated, autopilot very slowly increased speed...

teslamotorsclub.com

 

[Simavon]

I’ve noticed this too. I think also that the following distance is all over the place. The car will be slow to react to a deceleration, and then slow to react when things speed up again. I reckon they took the guy who did the A/C temperature regulation and asked him to see what he could do with AP.

 

[Mr_Dave]

The AP stystem is by far the worst in any car I’ve had. Alfa Romeo Giulia had Adaptive Cruise Control that was very good and only phantom braked in extreme conditions (cattle grid effect). Peugeot 3008 had Adaptive Cruise with Autosteer and that worked perfectly every time.
I’ve said before that if Tesla can’t get TACC right, FSD simply will not work. I just don’t trust it.

Edit: I still love the car though…

 

[MrBadger]

I’ve said before that if Tesla can’t get TACC right, FSD simply will not work. I just don’t trust it.

There is minimal overlap with the current radar based Autopilot Highways and vision only FSD City Streets beta - you cannot compare the performance of the functionality we currently have with that of the vision only stack and additional FSD city streets functionality. The vision based system is designed to eliminate the most significant cause of phantom braking and is a significant focus of their current effort as they have all but ditched radar.

The next few months (even days if you believe Elon time for early adopters) will be interesting with the Autopilot highways functionality what we all have here adopting the vision stack from FSD City Streets (but not the City Streets module). Tesla will then have no excuses for poor highways performance but until we get that we will continue to use a system that Tesla have (finally) admitted an issue with. They have a solution that in limited scope has rolled out in newer vision only vehicles (currently US market only) and is actively being worked upon - whether parity has currently been achieved with radar solution remains moot, but that is the basket that Tesla have currently put all their eggs in.

I am personally of the opinion that should Tesla put some additional effort into the highways functionality (over and above the consequential changes it receives from autopilot tweaks - such as the significant NN changes in 2021.4.15.x/.4.18 that were are still witnessing the fallout from, eg OP) it will be able to deliver highways performance that will fully (and exceed if it were allowed) the Level 3 motorway proposals.

 

[Fursty Ferret]

I can't see why Tesla would want to invest time and money in highways autopilot improvements or FSD in the UK given the restrictive UN/ECE 79 regulations.

While there's a consultation taking place the idea that anything might change (either quickly or sensibly, or both) is laughable.

 

[MrBadger]

I can't see why Tesla would want to invest time and money in highways autopilot improvements or FSD in the UK given the restrictive UN/ECE 79 regulations.

While there's a consultation taking place the idea that anything might change (either quickly or sensibly, or both) is laughable.

The UNECE r79 regulations are not as restrictive as many make out. In 2 years, I've probably been affected by a regulation limitation no more than a handful of times - the rest is down to Tesla's poor implementation, in particularly not adequately slowing at intersections. On most of the occasions I have seen, its been down to users using Autopilot autosteer outside the documented operational domain, exasperated by these incorrect use cases being normalised in social media. FSD/EAP is very usable right now for the appropriate UK roads and conditions but does, like driving itself, take a little time to gain the experience on how and when to use it properly.

I worked in the transport industry for a company involved in autonomous vehicle testing. Whilst things do not change very fast, for good reason, autonomous vehicle testing is nothing new. We ran (I think still run) a fully autonomous vehicle (what people would call Level 4 as it had a very well defined operational domain but no steering wheel and a very hands off safety monitor) and also other more traditional vehicles running autonomous shadow mode for various transport sectors. I can tell you that there is a lot of data out there and when the go ahead is given for limited operational domain Level 3 use, it will be possible to very quickly introduce.

 

[Fursty Ferret]

It is odd that the system fitted to Volvos and VAG cars is able to slow perfectly for pretty much any turn but the Tesla implementation is either way too much or nothing at all.

 

[Mr_Dave]

There is minimal overlap with the current radar based Autopilot Highways and vision only FSD City Streets beta - you cannot compare the performance of the functionality we currently have with that of the vision only stack and additional FSD city streets functionality. The vision based system is designed to eliminate the most significant cause of phantom braking and is a significant focus of their current effort as they have all but ditched radar.

The next few months (even days if you believe Elon time for early adopters) will be interesting with the Autopilot highways functionality what we all have here adopting the vision stack from FSD City Streets (but not the City Streets module). Tesla will then have no excuses for poor highways performance but until we get that we will continue to use a system that Tesla have (finally) admitted an issue with. They have a solution that in limited scope has rolled out in newer vision only vehicles (currently US market only) and is actively being worked upon - whether parity has currently been achieved with radar solution remains moot, but that is the basket that Tesla have currently put all their eggs in.

I am personally of the opinion that should Tesla put some additional effort into the highways functionality (over and above the consequential changes it receives from autopilot tweaks - such as the significant NN changes in 2021.4.15.x/.4.18 that were are still witnessing the fallout from, eg OP) it will be able to deliver highways performance that will fully (and exceed if it were allowed) the Level 3 motorway proposals.

I don’t dispute the above and understand the difference between the radar and visual systems and their intended applications. But, what I’m saying is that, as someone that only has basic AP that is so unreliable it’s close to unusable, why would I (literally) buy into the idea that FSD is even vaguely reliable or that the vision tech it’s based on is going to fix everything? It’s a question of trust. I’ve watched the YouTube vids and quite frankly, if I was a passenger in a car being driven by a person the way the FSD drives in the vids, I’d jump out and get the bus.

A simple solution for me would be to just add a driver controlled adjustable speed limiter that could be used around town and limit FSD to motorways/highways. But I know that won’t happen.