gearcruncher is just...very upset. It was impressive. That was a crazy place and human drivers are just as bad. In fact if you'd take a video of some human drivers and the AI I bet posters like gearcruncher would be equally harsh on both. Which just goes to show driving is hard. Which goes to show that maybe Tesla is right, if they solve FSD they will have solved AI in general...and that...that is a deep discussion. This is why these videos are so fascinating. We're seeing, live, the actual testing and implementation of a generalized approach to true AI, it just happens to be driving a car. Quite simply mind boggling.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Not everyone has a downtown commute which is mainly where I see the issues in the beta videos. Useful is going to be a different definition for everyone.
I already use NOA 20 miles a day, everyday. The other 20 miles of my commute are rural backroads with fairly good visibility at all intersections and no unprotected lefts. In my case I think beta in its current state would absolutely be useful.
That being said I understand everyone's situation is different and will admit that beta still has lots of issues and edge cases to fix going forward.
Don't forget that useful and L5 robotaxi aren't the same thing.
Knightshade
Well-Known Member
Yup... my ratio of NOA roads to rural/suburban is much higher in favor of NoA roads than yours (which is why I'm so interested in seeing the "new" stack applied to highways), but even so like 99% of my "other" driving looks like it could easily be handled by current FSD beta, with that 1% being the rare times I'm driving into an actual major downtown for a specific reason.
As knightshade would say, this is objectively false.
You can literally see other drivers in the same place in these videos. They are not using the bus lane to turn. They are not steering into the pillars. They are not going 10 MPH in a 25. They are not blindly sitting behind a bus. They are not turning into pedestrians.
There is not carnage on the streets like there would be if this FSD beta was left unchecked. The only reason there is not carnage is that a human is constantly monitoring this system and taking over when it makes an error.
As someone that drives this as my commute, this is not a place humans fail often. I have never seen a car crashed into the pillars. In fact, it appears humans are SO good at driving in this area that they can monitor a system which is failing 3 times a minute in this area at the same time, and we don't yell at them for being unsafe for attempting it.
LOL. As @powertold said- "they're fixing the edge cases." You don't fix the edge cases in generalized UI. Generalized UI can just handle all cases. What more "general" rules does Tesla need to teach FSD to get it not to steer into barriers or suddenly turn onto the wrong road at the last moment?
You can't just say "useful" without more description. Why would it be useful? How would it improve your experience as a driver? It's clearly not safety yet, it's far too unreliable. It doesn't allow you to do any tasks you can't currently do. Like I asked before, what assistance does this provide you?
Let's be honest. It's gonna be fun. But it's not useful in the current state.
The current highway stack is pretty good. I'd argue it's even "useful." What are you hoping the highway stack adds that isn't currently there? How will the new stack be more useful? We know it will be less proven, and thus will need more attention in the beginning, which seems like a reduction in the current usefulness of highway autopilot.
Yep, like I said. This is "suburban streets autosteer" not "city streets autosteer." You'd think a generalized system could drive either place like humans do, and I thought this was a mind blowing release, yet even defenders are kind of like... well.... maybe you shouldn't use that on actual city streets.
Knightshade
Well-Known Member
In it's current state it could do 100% of my drive to and from work, easily and safely, based on the types of roads my drive involves.
Not that I'd take a nap or anything, but it would bring the relaxed "I've only got to observe instead of actively make microadjustments all the time" experience I get on highways now to the small portions that are non-highway.
In the short term-
For daily driving- it seems to be better in spots with poor markings... there's one specific spot on my normal drive where if you're in a specific lane the markings are poor- so I know I have to either not be in that lane or be prepared to keep it in lane for about 10 seconds. I know the spot, so it's not big deal- but it'd potentially go away with the new code.
For less-common cases there's things like one highway interchange I do drive on once in a while where the system ends NoA right after taking the first off-ramp, and you have to manually merge onto the next highway- possibly because it's a slightly complex interchange (the off-ramp takes you onto a lane on the new highway that exits quite soon after AND the next lane over ends another 1000 feet past that so you essentially have to merge over 2 lanes quite rapidly). A more advanced stack might handle this better.
In the longer term- solving for things we KNOW the current stack is bad at-- but FSDBeta seems to handle fine- things like a stopped vehicle partly in your lane- would be the gateway to getting L3 or L4 highway only driving.
Which is the exact thing I bought FSD in hopes of eventually getting.
If that's all I ever get, and city driving remains L2, I'm 1000% happy with buying FSD.
So seeing the merged stack will give me some idea how near that might be.
The biggest problem with Navigate on Autopilot is that it misses exits when you are in traffic. It will change lanes into the far left lane, and before the exit it will try to change over multiple lanes but isn't aggressive enough. That and it will cut off faster traffic when trying to pass people. You're stuck behind a car going 65mph, it changes into the left lane but there's a car flying up on your tail at 80mph and it doesn't speed up enough.
And given what you see of city streets AP today, which misses turns due to traffic, you think this will be fixed by changing the highway stack?
I noticed that you answered why highway AP *improvements* would be useful, but not why city streets autosteer would be useful in the current state.
Last edited:
Ahh, good old Tesla marketing and perception bias. Have you actually read in the manual what the aggressiveness setting does? It doesn't do what you think it does.
Aggressiveness setting is only about speed based lane changes, not route based. It only changes the threshold used to determine if it should change lanes for slower traffic.
You can achieve exactly the same effect by just setting your cruise control setting lower.
In fact, because of the way the aggressiveness setting works, it's completely dependent on how much higher than the current traffic speed you have your cruise set to.
What speed above traffic do you find setting AP to gets you the best NoA performance?
It is my experience from talking to other users that they believe the aggressiveness setting is how fast it changes lanes, or how small a gap it will fit into, not only when it will move up a lane to avoid slower traffic. And Tesla's own marketing of this sure doesn't contradict this.
Wrong. Those are 2 different controls and cruising slower will only be the same as a mild lane change at consistent heavier traffic.
Wrong. It also takes into account the speed difference to adjacent lanes.
It depends on the road. My default setting is +10, but I normally try to match other cars. Sometimes it's 0, sometimes it's 15.
Somebody liked my post prior to this and here's yet another example of you trying to twist my words. Please, stop that behavior.
Wouldn't Tesla want to know when the driver is manually spinning the speed up or down so they can determine why? For example is it a mapping error or did FSD incorrectly read a speed limit sign? Immediately dismissing the operator feels a bit harsh don't you think since they are providing feedback to Tesla that could improve FSD?
Last edited:
EVNow
Well-Known Member
Do you use AP in the city now ? Took me a while to start using it - but now I use it all the time. I expect to use FSD beta all the time once it becomes available. I should note that I'm a careful, defensive driver.
I'm as vigilant with AP as I'm without it. AP just makes me safer as a whole. Chances of BOTH I & AP missing a traffic light, stop sign or going well over speed limit are quite a bit less than me alone.
I'm still very careful at points where I know AP needs to make important decisions - like choosing a lane or at traffic control points. I'm extra careful when I ask AP to change lanes (though it has never made a mistake, yet).
My guess is when FSD beta comes, I'll start using it slowly, first on empty roads and slowly moving to moderate traffic roads. Even now - I don't use AP in heavy traffic - so I probably won't do that with FSD too. As with AP, FSD has to earn my trust.
Funny you say that. I love using AP in heavy traffic, in my experience it makes a wonderful job with stop-and-go. Never had an issue, not the tiniest one in this scenario. What makes you not like it in heavy traffic? Or is my interpretation of heavy traffic different? I presume it could also mean "busy", but not "stop-and-go". Is that it?
Similar threads
