AlanSubie4Life
Efficiency Obsessed Member
The plot thickens. I’d like to restrict the bet to just this left turn (not the other UPL which it has no problem with typically anyway). Haha.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD Beta 10.13
- Thread starter Todd Burch
- Start date
Daniel in SD
(supervised)
I guess I’ll see you in the Delaware Court of Chancery with the way you’re trying to alter the deal.
AlanSubie4Life
Efficiency Obsessed Member
I think I am reasonably safe on this one as he seems to limit the videos to one turn recently. But ok, I see you are not confident. Can keep that part of the deal as is.
But what about the gap timing? I feel like 5 seconds is possibly too generous to you! It certainly is on the far side of the road (as a human I’d probably do 3-4 seconds, though in your ponderous AWD you might have to allow 5-6)! The way it is written now, this timing is arbitrary (and determined by me, when I say the car should have gone).
tmoz
S85D, Y LR, ex Prius PiP
This thread is 2 months and one week old, with 34 pages.
That has to be a record of some kind, especially seeing that there is a chance that there will never be a 10.13 release.
That has to be a record of some kind, especially seeing that there is a chance that there will never be a 10.13 release.
If you subtract 200 posts about a .75 cent beer bet with 50 rules of compliance analytics It’s not that bad.
Daniel in SD
(supervised)
This is San Diego. $8 at the craft breweries. Now you understand how serious this is.
Yeah, with all the specific rule categories for compliance you should have negotiated that detail. My bet is that you are playing for a PBR or Old Style can. Lol
tmoz
S85D, Y LR, ex Prius PiP
We could let the general audience here vote who wins the bet, that would be a more balanced subjective decision.
AlanSubie4Life
Efficiency Obsessed Member
In addition to the exorbitant cost of the beer, you have to remember this will literally be the first time on TMC when someone admits they were wrong about their FSD prediction.
I’d also like to think many posts were dedicated to determining definitively and proving beyond doubt, that actually this is a very easy left hand turn. (Additional new evidence: you can sometimes see Chuck do the turns in his video…I know he’s a naval aviator, but he cruises through them without any difficulty, while talking, without hesitation!)
P.S. There will be a picture of the beer.
Last edited:
Iain
Member
AlanSubie4Life
Efficiency Obsessed Member
Looking good for me: it appears to be requiring a gap in excess of 6-7 seconds at the moment. This shows the importance of that metric! On the upside, it also implies that the car can see about 150m or more (assuming 50mph).
tmoz
S85D, Y LR, ex Prius PiP
The word "see" is not specific enough. Being able to detect that something is present isn't enough to say it "sees" something. It has to be able to calculate its velocity as well. If it can detect the object's velocity within reasonable error bars, as well as detect it, then I would say it could "see" the object. We don't have data on the velocity information as far as I can tell.
Also, sometimes human drivers will determine that they have enough time for a maneuver and that the other car can slow down if the other car gets uncomfortable with the position of the drivers car. Not saying FSD can do this, but that might be a subconscious calculation humans use to determine if they can go or not.
Daniel in SD
(supervised)
It was actually a ten second gap it didn't take (32s-42s) which is actually larger than the gap it did take (1:02-1:09).
The advantage that FSD has on this turn is that people behind you can't see if the path is clear and honk at you.
AlanSubie4Life
Efficiency Obsessed Member
Necessary but not sufficient is how I would describe it I guess. I was only addressing what appeared to be its ability to detect that a vehicle was present. It looks like it waits for there to be no vehicles to be detected within >150m, regardless of velocity. This range may depend on assumptions about the speeds on the road.
With a single test, you cannot conclude how far the car can see by the time gap when it entered the intersection. It's possible that the car cannot see that far and assumed there was nothing coming.
Had there been no cars for 60 seconds, you could hardly conclude that the car can see nearly a mile away.
Although you have to be cautious about forming any conclusions on a single test, it does look promising that the car creeped well, moved across the first set of lanes very confidently and positioned itself perfectly in the median. I certainly hope that this is a typical result!
Tesla needs to get this right. Every active beta tester (as opposed to users) should have a favorite occluded ULT scoped out for the next version. I have several tough cases that have been thoroughly tested with 10.12.2 and am looking forward to evaluating 10.13/69 against them.
AlanSubie4Life
Efficiency Obsessed Member
Yes, I was just looking for theminimum value that I noticed was missed. And that larger gap was right after getting to appropriate pose, so gave it a pass.
In any case, a clear failure per the terms, though of course big improvements in pose and strategy. And making the turn is very good too! It’s ok for it to be conservative, as long as isn’t due to major limitations in perception range. But not ok per the terms (which directly impact usability of course - I would just take over and go).
AlanSubie4Life
Efficiency Obsessed Member
Of course not, that would imply a different problem.
Yes, it was my assumption that the car would not sit at the turn for no reason. That is possible of course. The distance estimates were made with that underlying assumption, anyway.
As far as going when it can’t see, obviously we hope that is not the case as it has taken a long time for it to cross the road in the past (another way to guess the minimum distance it might be able to see). Though it looks like it was pretty quick (4-5 seconds) on that front in this clip - it kind of did not look like it was under FSD Beta control for the actual crossing, or at least got kicked a second or two in, in fact.
Last edited:
Similar threads
