FSD discussion

[bkp_duke]

So why not watch the (hundreds) of videos of Driverless Waymo with dozens of instance of it handling double parked cars, constructions, lane-shift and road closures?

Why stay misinformed when you can easily be informed?
Youtube is your friend.

I've done better than that. I've repeatedly ridden in Waymo's when I went on business trips.

My experience matches what I see . . . on YT! LoL. Where the system actively avoids traffic, highways, and as many left turns as it can (it will usually make 3 right turns to avoid a single left).

This is a CLASSIC example released after Waymo expanded their Phoenix coverage, and mirrors what I have experienced as well. Waymo may get me there, but it's painfully slow (and not cheap):

Unlike you, I'm speaking from a position of 1st hand knowledge.

 

[JRP3]

Ouch, third degree burn!

 

[EVNow]

So why not watch the (hundreds) of videos of Driverless Waymo with dozens of instance of it handling double parked cars, constructions, lane-shift and road closures?

If Waymo can handle all these - why do they need HD Map in the first place ? Why do they need to test for months / years before expanding ?

Google had already spent billion+ on AVs before Tesla even started their in-house development in 2016. They have been testing actual driverless on streets for 5 years, nearly. Yet, they don't even operate in my area.

ps :

In 2017, Waymo sued Uber for allegedly stealing trade secrets.[17] A court document revealed Google had spent $1.1 billion on the project between 2009 and 2015. For comparison, the acquisition of Cruise Automation by General Motors in March 2016 was reported at just over $500 million, and Uber's acquisition of Otto in August 2016 was for $680 million.[45] Waymo began testing autonomous minivans without a safety driver on public roads in Chandler, Arizona, in October 2017.[46]

 

[Buckminster]

https://twitter.com/x/status/1662128733850062848

 

[SPadival]

The software probably won't allow activating FSD Beta due to hardware and region checks, but because the neural networks are available, it does clear the path for making use of the neural networks for replacing basic Autopilot lane and distance keeping to push vehicle safety to the next level even to the existing fleet of millions.

FYI, nothing yet in this version 2023.12.10 - just an odd separation of buttons for TACC & Autosteer. I was hoping at least for FSD Beta visualisation with this version.

imgur.com

Discover the magic of the internet at Imgur, a community powered entertainment destination. Lift your spirits with funny jokes, trending memes, entertaining gifs, inspiring stories, viral videos, and so much more from users.

imgur.io

 

[Buckminster]

https://twitter.com/x/status/1662226586194804736

 

[Mardak]

FYI, nothing yet in this version 2023.12.10 - just an odd separation of buttons for TACC & Autosteer

Thanks for checking. The odd separation is probably because FSD Beta can't be activated there in 2023.12.10… yet? Potentially some software check will make the hidden 3rd slider position become available:

 

[Buckminster]

 

[FSDtester#1]

I just completed a 22 mile round trip residential/city loop on 11.4.2.
I was specifically checking 2 areas where fsd would slow from 55 to 35 mph every time I drove past those areas. @Ramphex has a similar area where his MS does the same thing and even slows to 0. To my surprise both areas are pretty much fixed.
1 was non existent, 2nd it went down 2 mph, but not noticeable enough to make me or drivers behind me uncomfortable. It also completed a U turn on a 4 lane divided highway, where in the past would force me to take over. I then tested residential streets with cul-de-sacs which also in the past would be a forced takeover because the car couldn't "complete the maneuver". It successfully completed that multiple times and kept the speeds down due to "residential area detected ", which overrides any speed offset you have programmed. All this testing took place after taking the recommendations of other members to clear out the Teslacam USB. Which others have stated that the full or corrupt files in the USB may interfere with FSDb and it's potential processing power.

Tldr: Bullish for fsdb and it's progress coming from a 2 year, 30K plus beta mile tester.

 

[y_naught]

I just completed a 22 mile round trip residential/city loop on 11.4.2.
I was specifically checking 2 areas where fsd would slow from 55 to 35 mph every time I drove past those areas. @Ramphex has a similar area where his MS does the same thing and even slows to 0. To my surprise both areas are pretty much fixed.
1 was non existent, 2nd it went down 2 mph, but not noticeable enough to make me or drivers behind me uncomfortable. It also completed a U turn on a 4 lane divided highway, where in the past would force me to take over. I then tested residential streets with cul-de-sacs which also in the past would be a forced takeover because the car couldn't "complete the maneuver". It successfully completed that multiple times and kept the speeds down due to "residential area detected ", which overrides any speed offset you have programmed. All this testing took place after taking the recommendations of other members to clear out the Teslacam USB. Which others have stated that the full or corrupt files in the USB may interfere with FSDb and it's potential processing power.

Tldr: Bullish for fsdb and it's progress coming from a 2 year, 30K plus beta mile tester.

I’ve only made one round trip drive for 20 miles, but finding 11.4.2 much better than 11.3.6, especially the city street lane finding and planning.

edit: also, no safety-related disengagements!

 

[FSDtester#1]

I’ve only made one round trip drive for 20 miles, but finding 11.4.2 much better than 11.3.6, especially the city street lane finding and planning.

Also, it is now super confident in merging on 2 and 3 lane roads to find gaps to change lanes. A couple of them I was gripping the yoke and had feet ready to intervene.
They were all safe, but way more aggressive then prior versions.
Before the car or "ego" , would look for about 3 car lengths of room to change lanes.
Beta did 3 of them with maybe 1 to 1.5 car lengths of space.

 

[insaneoctane]

I will summarize the new 11.4.2 as yet another step clearly net forward. I still experience some non-perfect behavior, but it is smoother, in general, and I agree that it somehow appears to be looking further ahead somehow. Still not as far ahead as me though. This pace of improvement truly makes me optimistic.

 

[bkp_duke]

I just completed a 22 mile round trip residential/city loop on 11.4.2.
I was specifically checking 2 areas where fsd would slow from 55 to 35 mph every time I drove past those areas. @Ramphex has a similar area where his MS does the same thing and even slows to 0. To my surprise both areas are pretty much fixed.
1 was non existent, 2nd it went down 2 mph, but not noticeable enough to make me or drivers behind me uncomfortable. It also completed a U turn on a 4 lane divided highway, where in the past would force me to take over. I then tested residential streets with cul-de-sacs which also in the past would be a forced takeover because the car couldn't "complete the maneuver". It successfully completed that multiple times and kept the speeds down due to "residential area detected ", which overrides any speed offset you have programmed. All this testing took place after taking the recommendations of other members to clear out the Teslacam USB. Which others have stated that the full or corrupt files in the USB may interfere with FSDb and it's potential processing power.

Tldr: Bullish for fsdb and it's progress coming from a 2 year, 30K plus beta mile tester.

Thanks for the USB tip, will go clear out the footage today in both cars.

 

[FSDtester#1]

Thanks for the USB tip, will go clear out the footage today in both cars.

Before I did, I backed it up to my laptop fyi

 

[bkp_duke]

Before I did, I backed it up to my laptop fyi

I've got 100TB of SSD storage at home . . . I'll find a spot for it.

 

[Dave EV]

PSA: For those experiencing unusual amounts of hesitation, visualization glitches, or red-hands system errors: it could be your USB drive.

Two days ago, I had a system crash that threw up red-hands and disabled all TACC/FSDb for the remainder of my drive. Even a two-thumb reset didn't fix it.

When I got home, I pulled out my Sentry Mode SSD drive and ran a Check Disk utility on it: irreparable drive errors. I removed my music partition and formatted the drive, and let my Model 3 sleep over night.

The next day, not only was FSDb back, but it was driving smoother than ever. Much less hesitation and latency, much smoother driving, and several zero-disengagement drives.

I don't know if the FSD computer mounts the Sentry Mode drive directly (maybe it uses the space to cache clips), but it's pretty clear to me now that drive errors do seem to negatively impact FSDb performance.

I feel like USB has almost always been prone to flakiness to some degree in computers for as long as I can remember, so I'm not surprised that issues with USB storage could be causing issues with general hiccups, delays and latency in the operating system.

Combine this with trying to write data to a USB device at speeds nearing it's maximum bandwidth while maintaining low latency and it's easy to see how small changes in USB storage device performance can affect sluggishness.

An easy test might be just removing the USB drive. FWIW I had tried various Sandisk and Samsung (not the BAR that Tesla ships now) USB drives, but eventually those all started getting flaky. Most recent one is the 256GB Samsung BAR (same drive that Tesla ships, but bigger, Tesla ships the 128GB version) and so far this one has been good. IMO this application really needs a real SSD with a good size RAM/SLC cache and low latency, but I haven't found a small enclosure and drive that I like.

PS - Still waiting for the latest FSDb here...

 

[EVNow]

Fsd is getting closer to becoming a reality wall st will have to recognize...

https://twitter.com/x/status/1662202904151838720

After my first drive on 11.4.2 I can report that it is not that different from 11.3.x.

OTOH, Curt has nice video estimating the perceived range of side cameras to be 2x the rated 80 mtrs.

 

[bkp_duke]

After my first drive on 11.4.2 I can report that it is not that different from 11.3.x.

OTOH, Curt has nice video estimating the perceived range of side cameras to be 2x the rated 80 mtrs.

If you have a USB drive installed for sentry mode, clean it out. Per the FSD thread, this may actually be causing some problems with FSD computer response times. Not confirmed, but has been reported. I'm still waiting on 11.4.2, so can't comment on the drives yet, but 11.3 has been very good here in SoCal.

 

[heltok]

FSD Forward looking Side Camera can now see up to 190m sideways. A bit more than the stated range of 80m.

The way they accomplished this is mainly through having a better dataset thanks to autolabel, but also by having a bigger an d better neural network and using photon counts.

 

[SPadival]

What’s the reasoning behind sending two versions of FSD to the masses? Initially, I thought only the OG testers will get the latest beta while the masses get a slightly older version (for safety reasons while the latest version is trialed by experienced FSD testers)