Perhaps you would like to comment on the videos provided before attacking someone else’s analysis, or did you not watch them before spouting from the hip?
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
diplomat33
Average guy who loves autonomous vehicles
That is a bold prediction. What data do you base this on? What is FSD beta's accident rate? What is FSD beta's safety disengagement rate? What is FSD Beta's safety envelope violation rate? How many safety false positives or false negatives does FSD beta have per detection? Don't you think we need well, statistics, if we are going to make judgments about statistical safety?
It is fine to say that you think FSD Beta is improving in your opinion. But you cannot say that is "very clear" that Tesla will achieve statistical safety greater than humans just based on watching videos of easy routes with no traffic!
Chazman92
Member
FSDBeta 9.1 - 2021.4.18.13 - First Impressions Memorial Park Drive for Comparison to Previous v9
Check it out!
Check it out!
Terminator857
Active Member
Wild swerves at 21:30
Confirmation required for left turn at 22:30
Last edited:
Terminator857
Active Member
Instead of poo pooing others for being enthusiastic, you could give your assessment?
@Chazman92 is there a chance that at the start of end of your video (with a time stamp link) or a different vid, you give us a run some of the errors?
diplomat33
Average guy who loves autonomous vehicles
That's an average of 1 disengagement per 40 minutes of driving. Personally, I don't think that is a good disengagement rate for FSD. Although it is lacking a lot of info. We don't know the type of disengagement, type of driving, number of miles etc...
If only Tesla would release the disengagement data, then we could maybe quantify the progress better.
It ls not bad as a level 2 driving aid though. And what really matters how is how much it Improved vs it’s previous iterations
diplomat33
Average guy who loves autonomous vehicles
That is why I said "for FSD". Yes, for L2, it might be ok. I could see situations where for short commutes (say 10 minute drive), especially if traffic is not very bad, the car could probably do the commute with no interventions. So as long as the driver is paying attention, it could be a nice convenience feature for some owners. When Tesla improves the rate say 10x, it might good enough for L2 wide deployment. But it will be important to make sure the driver is paying attention especially as the rate gets better to avoid complacency. As the drivers do hours of driving with no intervention, they will start to get complacent.
scottf200
Well-Known Member
diplomat33
Average guy who loves autonomous vehicles
powertoold
Active Member
9.1 seems like 3 steps forward, no steps backward
Chazman92
Member
#FSDBeta 9.1 - 2021.4.18.13 - Unprotected Left Turns with Drone View - Looping behavior
My personal opinion so far is one step forward, no step backward.
powertoold
Active Member
I don’t base it on any data. It just ends up being my opinion, but it’s starting to seem like 9.1 will perform as well or better on Waymo-equivalent routes in Chandler, and we know that Waymo’s service in Chandler is human level+.
People talk about how easy some of the 9.1 routes are, but most of them are actually much more complex than the ones Waymo takes in chandler:
Last edited:
powertoold
Active Member
See my thing about “Waymo-equivalent routes in Chandler.” I’ve watched many Waymo videos. You should as well to understand my point about how easy they are. Also, Waymo’s performance on routes have gotten worse over the last 6 months. They’ve restricted routes that caused issues, so now, the Waymos take long detours to avoid more complex areas.
Whereas Tesla has completely revamped their approach to vision only AND dramatically improved the software within 3 weeks.
Last edited:
it’s too early to determine. Definitely not from a night video with literally no traffic or road user what so ever.
and “astonishing” shouldn’t be that it made this one turn better or it’s more stable in lanes. Astonishing is it went from 1-5 miles per safety disengagement to 500-1000 miles.
these are safety critical systems were talking about.
if alil comfort and stability is your barometer for “astonishment” then every version will be “astonishing”. Astonishing loses its meaning. Which is per for course when talking about Tesla. Everything is the best thing since sliced bread.
why can’t it be “the turns are better and it’s less shaky and more confident which is nice”
Not “OMG LOOK HOW IT MADE THAT TURN. That BLEW MY MIND OMG AMAZING ASTONISHING THE IMPROVEMENT IM BLOWN AWAY L5 IN TWO MONTHS OMG ELON YOU GOD”
Daniel in SD
(supervised)
Waymo can do a 10 mile loop 10,000+ times without a collision.
I'm also not completely convinced that they're greater than human level.
Similar threads
