I've had my MY for about the same amount of time as @ryanjeffords and I go to my in-laws about once a week so i've been able to try NoA over the same route multiple times and in two general conditions lite traffic and medium traffic situations.
FSD is like the perfect driver ed student. It knows and follows rules of the road pretty well and in these areas I actually trust it a great deal. If it's just a matter of keeping distance or following traffic in stop and go where there's zero decision making it's about as good as I can ask anyone to drive. Where FSD is really lacking are situations that involves human-factors or requires human experiences and the ability to predict potentially hazardous situations. FSD doesn't really do merging traffic well or at least I don't trust it enough to let it handle it.
What I mean by human factors are things like It doesn't watch cars coming onto the freeway on ramp, anticipating it and slow down/speed up so you'd end up parallel with a car that is trying to get onto the freeway. Speaking of, FSD have no issues with staying parallel with another car for super long stretches. It's not against the rules of the road but it's generally a bad idea as people in generally have pretty poor situational awareness of what's to the sides and rear of them while driving. I've found that when I do "take over" I generally have it change lanes to avoid a potentially problematic merge, or if I can't avoid being in the merging lane then I take over. In the parallel situation, I'd hit the accelerator until I'm ahead of the other driver. When I see a situation where I'm not sure how well AP will handing, I generally hover my foot over accelerator or brake peddle and so far I've generally found that I have not had to intervene in these situations.
From all the times I've used FSD on the hwy, my general conclusion is that under a lite traffic condition, you can pretty much trust that it will take you from on-ramp to off-ramp, especially if you tell AP to use the middle or left lanes on the freeway. Under medium traffic conditions you might have to direct it a bit more in terms of telling it to preemptively switch lanes or speed up based on traffic pattern. Under heavy congested traffic, FSD/AP works as long as you are not in the merging lane and if you have to be on the merging lane, just take over and drive yourself.
I've played with FSD on streets some but it is a far cry from what it can do on the hwy and in general this is because you are basically constantly in a merging lane in medium-heavy traffic condition on surface streets. One unexpected problem that I've found is that one part of my route to work have a section where the posted speed limit is 40mph, but because it is a huge stretch of downhill that the average traffic speed is 60+. You'd probably get run over if you are going 40 and AP wouldn't let you go over +5mph off highways which makes FSD unusable on my route to work. If my work route has almost no traffic and doesn't need to make turns at intersections it can probably get me to work and back w/out interventions. It will be interesting to see when we do get the Beta button how much of these human factor related problems Neural Net FSD have solved.
FSD is like the perfect driver ed student. It knows and follows rules of the road pretty well and in these areas I actually trust it a great deal. If it's just a matter of keeping distance or following traffic in stop and go where there's zero decision making it's about as good as I can ask anyone to drive. Where FSD is really lacking are situations that involves human-factors or requires human experiences and the ability to predict potentially hazardous situations. FSD doesn't really do merging traffic well or at least I don't trust it enough to let it handle it.
What I mean by human factors are things like It doesn't watch cars coming onto the freeway on ramp, anticipating it and slow down/speed up so you'd end up parallel with a car that is trying to get onto the freeway. Speaking of, FSD have no issues with staying parallel with another car for super long stretches. It's not against the rules of the road but it's generally a bad idea as people in generally have pretty poor situational awareness of what's to the sides and rear of them while driving. I've found that when I do "take over" I generally have it change lanes to avoid a potentially problematic merge, or if I can't avoid being in the merging lane then I take over. In the parallel situation, I'd hit the accelerator until I'm ahead of the other driver. When I see a situation where I'm not sure how well AP will handing, I generally hover my foot over accelerator or brake peddle and so far I've generally found that I have not had to intervene in these situations.
From all the times I've used FSD on the hwy, my general conclusion is that under a lite traffic condition, you can pretty much trust that it will take you from on-ramp to off-ramp, especially if you tell AP to use the middle or left lanes on the freeway. Under medium traffic conditions you might have to direct it a bit more in terms of telling it to preemptively switch lanes or speed up based on traffic pattern. Under heavy congested traffic, FSD/AP works as long as you are not in the merging lane and if you have to be on the merging lane, just take over and drive yourself.
I've played with FSD on streets some but it is a far cry from what it can do on the hwy and in general this is because you are basically constantly in a merging lane in medium-heavy traffic condition on surface streets. One unexpected problem that I've found is that one part of my route to work have a section where the posted speed limit is 40mph, but because it is a huge stretch of downhill that the average traffic speed is 60+. You'd probably get run over if you are going 40 and AP wouldn't let you go over +5mph off highways which makes FSD unusable on my route to work. If my work route has almost no traffic and doesn't need to make turns at intersections it can probably get me to work and back w/out interventions. It will be interesting to see when we do get the Beta button how much of these human factor related problems Neural Net FSD have solved.
