As I noted above there are different issues with measuring the abilities of car. Clearly accident rate is the most important. Then there is legality .. does the car obey traffic laws (and know when they occasionally must be skipped). Then there is comfort .. does the car drive pleasantly for the occupants. Then there is accuracy .. does the car get you where you wanted to be?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
As I noted above there are different issues with measuring the abilities of car. Clearly accident rate is the most important. Then there is legality .. does the car obey traffic laws (and know when they occasionally must be skipped). Then there is comfort .. does the car drive pleasantly for the occupants. Then there is accuracy .. does the car get you where you wanted to be?
diplomat33
Average guy who loves autonomous vehicles
I think there was a miscommunication. If you look at my first post, I say that safety related interventions need to be 100k-200k miles. So I was comparing safety interventions to accident rate. I was not saying that all interventions should be compared to accident rates.
See my quote below:
But you disagreed with my first response quoted above, so I thought you meant we should use non-accident interventions also. The statistic you used was accident rates. People make all kinds of unsafe maneuvers that don't lead to accidents. I still say that if you want to use accident rates as a benchmark, you can only count interventions that would have caused an accident otherwise.
diplomat33
Average guy who loves autonomous vehicles
The reason I disagreed is because there are non-safety interventions that still matter. You still need to count them. You can't just ignore them completely. But you don't include them in your comparison with accidents. When measuring safety, you only count safety interventions. But you still need to look at other non-safety interventions to help improve your FSD.
Basically there are two different interventions:
1) safety interventions. You count those and compare them to accidents rate to measure safety of the AV.
1) non-safety interventions. You don't compare them to accidents but you use them to improve your FSD.
I think we are probably saying the same thing. We just need to be clear what intervention we are talking about.
In a Tesla it's just logical to be paranoid about curbs.
mikes_fsd
Banned
Just stating the obvious but unfortunately we won't know what percentage of "safety interventions" not taken by the driver would result in an accident.
From the discussion in FSD rewrite will go out on Oct 20 to limited beta about how not all networks/predictions are using surround camera view… It seems quite possible that static objects like cones and road closure signs might rely on main camera predictions whereas fisheye and pillar cameras are also used for road edge predictions. That could be why FSD beta happily creates a path towards the detected road when the barrier is just out of view of the main camera.
Would it be possible somehow for you to drive directly towards a road closure sign to see if FSD beta behaves any differently? Although doing that might not change the behavior of the surround camera predictions but might cause some other part of Autopilot to decide to throw up a big red take-over-immediately warning as it does for production Autopilot.
Pushing Tesla's FSD BETA over the limits - 26:45 - AI DRIVR
off the top of my head, both Chuck and Mars have had their car hit curbs, so AIDrivr isn’t the first. Still not a fan of how long he lets FSDBeta hang at stop signs, confusing and annoying other drivers on the road with him. Some very impressive performance on narrow residential streets with cars parked on the street in the later parts of the video before the car curbs the wheel
When you leave the Chicago Fed, FSD beta goes from the outer turn lane to the left-most lane and you say it's fine. It looks like the map data might indicate there's 4 lanes on Jackson, so that might be wrong, but I might be missing some behavior of that intersection with the bus-only and parking lanes?
No you are right. It's not "fine". I guess it's fine that there is nobody behind me but you are right that if it's on the far right, it should turn on the far right lane on Jackson. My comments are not always accurate during the drive. I'm mainly focusing on my safety and not disrupting other cars. Thanks Mardak!
Yeah, that's how I interpreted your comment as well. If you're looking for more double left turn lanes:
Edit: Separately, I noticed there was a U-turn only lane here as well as a strange left/straight lane just south of Union Station at Harrison & Canal (bottom left of the top map image):
Last edited:
Very nice improvement for Brandon, it looks like. The unprotected left at 2:15 was particularly impressive. Pretty flawless through the first part of the video, and while there were quite a few disengagements in the second half of the vid in the downtown area, still looked like a good improvement overall.
diplomat33
Average guy who loves autonomous vehicles
Dan Detweiler
Active Member
Elon stated yesterday that the version numbering system is changing. FSD will be numbered separately from the car. He said 8.1 is the version he was driving on and is the most current.
Dan
Dan
Similar threads
