But then - when that happens - far in the future - may be the driving will be better when the map & visuals don't agree.
Very true
Yeah, that second attempt was incredibly dumb. I stopped watching there since it was clear that there was either zero improvement or huge regressions in UL logic. Absolutely terrible.
Had my first curb strike recently on a left turn from a light, into a divided Ave. where the front of the vehicle cleared the median but the rear wheel caught the unforgiving concrete curb. I was paying close attention; but, I‘ll admit I had begun to get used the the vehicle correcting itself and being consistently successfully through these sorts of things—even if at the last min. Not this time!
Wait - is the car turning right after waiting for a while or is he disengaging ?
I stopped watching after the first 2 times this happened but those were manual disengagements.
This is a terrible regression, perhaps one of the worst regressions since the beginning of FSD beta videos which started in late October 2020. The only one that might have been worse was the 10.3>10.3.1 debacle which didn't have much video because it was pulled and replaced within one day.
You’d think they would would have a very good model of Chuck’s intersection by now and they could just run millions of simulations. I guess you have to assume that all this works in simulation… these don’t seem like edge cases.
Personally I don't wait for a long time to make such left turns. I turn right and take alternate route. I think thats what Waymo does too ... High speed unprotected left turns aren't the most important cases to solve first.
If the vision system can't see, no amount of training will improve matters. Chuck picked this intersection originally because it was such an extreme case. The visibility to the left is obscured by vegetation. There are no nearby
either left or right on the highway to mediate cross traffic. The speed limit on the highway is 55 mph. That intersection is very difficult.If the vision system can't see, no amount of training will improve matters. Chuck picked this intersection originally because it was such an extreme case. The visibility to the left is obscured by vegetation. There are no nearby
Taking the turn by first stopping in the median might improve the success rate but it might be difficult for the car to exactly pull into the median without blocking traffic. Besides, this is a tee intersection. If the neighborhood road continued straight, pulling into the median might be even more difficult to design.
The question as always is - what is the intent - entertainment or information. Picking a typical ULT is more useful, IMO.
