My wife also thinks I’m nuts to have it on our two cars. When I bought it during the “sale” in 2019 it seemed like a good move ($5k for AP/FSD and $7k for EAP/FSD) and I think time has proven me right, at least in terms of cost.
-
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 10.12
- Thread starter EVNow
- Start date
FSDtester#1
Craves Electrolytes
Well said also, and I have to mention that the "Shill" debate on these threads has been quite entertaining!
Even though like many things we say are off topic or OT. (I thought that meant over time until last week!) Staying OT, I had family in town last week. They were quite impressed with the "Tesla Cam" and all it's features.After their first drives in an EV, they have renamed it "Gap Cam" and asked why I don't record every time and ICE tries to have a go with a Tesla....
n.one.one
Info w/o Drama
Just posted but not much activity yet on TeslaFi or Teslascope.
You inspired me to post also
. I agree! This update has been a huge improvement for my area as well (IL/MO). It is handling back roads around our subdivision so much better. It used to be capped. around 28 mph, now it works as you would expect. Turns, our subdivision island, and proceeding through our stop signs is much better (still some room for improvement though). This morning I had to run to Menards that was 25 miles away with 50% interstate and 50% city. I only had 1 disengagement in the 50 mile trip. it just seems much more capable. I have been in beta since Oct 2021. Most of the updates have been hard to tell much difference. This one seems huge in my area.I'm hopeful it's new cars. It seems weird to tweet that it's going to existing cars given that the entire current FSD fleet is on 10.12.2 according to TeslaFi. Also, no one was asking about existing cars lol.
I think thats the plan. As usual Elon considers any imminent plan already executed
sleepydoc
Well-Known Member
and any future plan will be done within 1-2 weeks.I think thats the plan. As usual Elon considers any imminent plan already executed
Its quite likely the communication from the team to him was "We are going to send the release to everyone" a few days back. So, he assumes it is being done / already done.
sleepydoc
Well-Known Member
People’s natural tendency is to focus on what‘s wrong more than what’s right. On a certain level that makes sense, since the problems are what you need to work on but it also tends to make things seem more negative.
The variability has always been present. I’m sure it frustrates the FSD programmers even more than it does people here. People have theories, but a lot of it is difficult to explain. My general take is that overall it’s a slight improvement with some things getting better and others a bit worse. The nuance to that is that the things that are worse should be easier to fix. Example - many people have posted that FSD is now more confident but actually borders on aggressive. Ok, so it’s not necessarily better, just different. Except the part that got better was the part that needed more work to achieve. Having the confidence to safely turn across 2 lanes of traffic requires significant judgement. Doing it at the right time but too aggressively not great but it’s easier to fix than figuring out the judgement component.
So even though I’m not blown away by this release I still see progression and improvement in the algorithms.
New car here! installing now to a 2016 Model X with camera retrofit done!
Teslafi is showing 390 cars now waiting to get 10.12.2. Many of them have to be new testers since only 90 are still on 10.11.2 !
100 score. Still shows score in app but car not done installing yet.
People are more motivated to complain than to praise. It's pretty universal. I expect it's because we feel that complaints may lead to improvement while praise does not.
I agree but I don't think that's enough. I think there is already some path continuity prediction of other moving users for intersections.
I think one input element into the neural network would still need to be "is a railroad crossing coming up on my route?" and at the moment that has to be provided by mapping. Humans use this as well---a person who knows the area and meets a stopped car will wait if there's a rail crossing ahead vs someone who knows nothing, and at the moment the nets are nearly 100% amnesiac.
"Once this occurs, they can leverage that to infer that the last car in a line at a railroad crossing is part of a line of cars, even if the other cars have become occluded."
Maybe but that's very hard to train, like in this current case the truck was blocking all view, there's no way to know if the cars ahead had already gone or not while occluded. In other cases, this will give many false positives and the computer will erroneously stop itself in traffic behind another truck and never move, causing a further backup.
BTW the radar bouncing under one car to the next one ahead could have helped here.
Planning will never be in NN for this reason. It needs to take inputs from NN and make decisions. One of those would be based on the map and navigation.
Similar threads
