Will single stack bring benefits to standard Autopilot?

[Ozdw]

For those of you using FSD beta, do you have any experiences that might suggest what improvements might flow to non-FSD Autopilot when we have the single stack? I'm thinking of things like improved lane positioning around curves, which today are a real safety issue since the car hugs the centreline far too closely (or even crosses into the other lane). If FSD beta takes curves perfectly, then that might indicate we'll all benefit once the stacks are converged. [Yes, I realise that Autopilot is today intended only for divided highways. It is interesting to carefully experiment with it in other situations, though, to understand how it works.]

 

[diplomat33]

For those of you using FSD beta, do you have any experiences that might suggest what improvements might flow to non-FSD Autopilot when we have the single stack? I'm thinking of things like improved lane positioning around curves, which today are a real safety issue since the car hugs the centreline far too closely (or even crosses into the other lane). If FSD beta takes curves perfectly, then that might indicate we'll all benefit once the stacks are converged. [Yes, I realise that Autopilot is today intended only for divided highways. It is interesting to carefully experiment with it in other situations, though, to understand how it works.]

"Single stack" means using FSD Beta to do highway driving (currently handled by NOA). So it would affect people who are using NOA now. It would not affect standard AP. But if Tesla "ports" the FSD Beta code to standard AP then yes, AP would have the same improvements as FSD Beta.

 

[Ozdw]

Thanks for the clarification. I'm hoping if the beta code has evolved its "how to stay in the lane" methodology, that would become the standard across even base Autopilot. My assumption is all their effort is today in evolving the beta, leaving Autopilot not evolving at all. I guess time will tell whether those advancements will flow outside the premium options into the standard Autopilot.

 

[am_dmd]

"Single stack" means using FSD Beta to do highway driving (currently handled by NOA). So it would affect people who are using NOA now. It would not affect standard AP. But if Tesla "ports" the FSD Beta code to standard AP then yes, AP would have the same improvements as FSD Beta.

What will highway AP do differently or improve with single stack?

 

[nvx1977]

Thanks for the clarification. I'm hoping if the beta code has evolved its "how to stay in the lane" methodology, that would become the standard across even base Autopilot. My assumption is all their effort is today in evolving the beta, leaving Autopilot not evolving at all. I guess time will tell whether those advancements will flow outside the premium options into the standard Autopilot.

I think there's a very good chance that standalone AP will be powered by FSD code (the future of the current FSD-beta code). It will just be nerfed to do only lane keeping and following vehicles. I don't think Tesla has worked on standalone AP for a long time now, so it's likely going to be deprecated at some point and replaced by the new stuff.

I wouldn't expect standard AP to be replaced with the new code when things are on a single code stack. Likely they will keep single stack within the closed beta until most regressions are worked out.

 

[diplomat33]

What will highway AP do differently or improve with single stack?

We don't know yet. We need to see single stack in action to answer that. All we know is that single stack will use the FSD Beta code to do highway driving.

 

[nvx1977]

What will highway AP do differently or improve with single stack?

This is my educated guess:

The visualizations will get a lot better: shows all lanes and vehicles in all lanes; shows brake lights and likely turn signals (although turn signals are currently not visualized yet in FSD-beta).

Car will likely make better evasive maneuvers when necessary. We see this on local roads where the car doesn't treat lane lines as strict boundaries. Reaction time should improve, as the new code will be leveraging HW3.

Car will no longer be fooled when in the rightmost lane by on-ramps where it tries to swerve to the center of the widened lane.

Car no longer brakes for overpasses or signs (more tied to Tesla Vision than FSD). Max speed will be restored to 90mph.

Again, I don't think AP will see these improvements when single-stack FSD-beta is initially released. It will likely stay within closed beta for a while.

 

[Ozdw]

This is my educated guess:

The visualizations will get a lot better: shows all lanes and vehicles in all lanes; shows brake lights and likely turn signals (although turn signals are currently not visualized yet in FSD-beta).

Car will likely make better evasive maneuvers when necessary. We see this on local roads where the car doesn't treat lane lines as strict boundaries. Reaction time should improve, as the new code will be leveraging HW3.

Car will no longer be fooled when in the rightmost lane by on-ramps where it tries to swerve to the center of the widened lane.

Car no longer brakes for overpasses or signs (more tied to Tesla Vision than FSD). Max speed will be restored to 90mph.

Again, I don't think AP will see these improvements when single-stack FSD-beta is initially released. It will likely stay within closed beta for a while.

I can only hope your imagination foretells their intent. These changes would make AP just about perfect for those of us that live far from cities and towns.

 

[clydeiii]

Yes, it should get better. We of course don't know for sure.

 

[Supcom]

For those of you using FSD beta, do you have any experiences that might suggest what improvements might flow to non-FSD Autopilot when we have the single stack? I'm thinking of things like improved lane positioning around curves, which today are a real safety issue since the car hugs the centreline far too closely (or even crosses into the other lane). If FSD beta takes curves perfectly, then that might indicate we'll all benefit once the stacks are converged. [Yes, I realise that Autopilot is today intended only for divided highways. It is interesting to carefully experiment with it in other situations, though, to understand how it works.]

Sadly, FSD beta has the same issue as AP with crossing the centerline of the road on tight righthand curves.

 

[Ozdw]

Sadly, FSD beta has the same issue as AP with crossing the centerline of the road on tight righthand curves.

Wow. This is going to result in tragedy at some point, I fear. Many of the issues with AP are annoyances, but this is a defect in a class by itself.

 

[Undertoad]

Autopilot has logged 3.3 billion miles as of the Q1 report

 

[Ozdw]

Autopilot has logged 3.3 billion miles as of the Q1 report

That's great. How many times has a driver had to disengage it to avoid taking out an on-coming vehicle? Suspect that isn't in the stats.

 

[Dewg]

I wonder how many other L2 cars have similar experiences? Are there BMWs, Audis and Mercedes that have drivers forced to disengage their adaptive cruise control because of phantom braking or lane change problems? Issues with nagging about being attentive? Is it a purely Tesla problem?

 

[damonbrodie]

Elon just tweeted that EAP will be brought back to all markets. I'm hoping that the FSD single stack code will be used for EAP (and normal AP) at some point in the future.