Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

FSD Beta Videos (and questions for FSD Beta drivers)

This site may earn commission on affiliate links.
How fast can cloud network respond?

Not nearly fast enough.

The onboard NNs are fed frames of data every 27ms.

Average 4G LTE latency is roughly double that time budget- and that's just the delay of the roundtrip data ping to the cloud and back, even if the back end cloud stuff spent -0- time actually doing anything with the data... once they do you're obviously even more late.

And that's average pings, some will be much slower. Or non-existant without cell coverage.

Nobody, including Waymo, uses "cloud" for real-time driving.


5G might offer a usable connection for this stuff (latency in the 1-5ms range) but by the time it's in common enough use you could rely on having it available everywhere, on-car compute will already be far more powerful than it is today.



Cool. But that's AP not NOA. Hope it's both.

it's the same limit- you can't have NOA without AP being on.
 
Omar

Frenchie
02:00 No Turn On Red

Dirty Tesla

Rocco
1:32 Improvement from 10.11.2 - slowed for car pulling out in front better 3:06 A First - didn't stop for this turn! 4:28 Disengagement 1 - failed to turn 5:35 A First - made this turn! 6:52 Improvement from 10.11.2 - checks for traffic better 7:05 Disengagement 2 - didn't appear to stop for oncoming car 7:54 Improvement from 10.11.2 - didn't slow for blinking yellow
 
Last edited:
Rob Maurer from Tesla Daily, downtown Milwaukee. Overall impression: progress, but there were some issues.
3:28 Re-routing 4:37 Yielding 5:07 Bridge out 6:45 Crosswalk confusion 7:45 Cross-traffic 9:00 Bicyclists 10:40 Intersection issue 11:45 Hairpin 13:20 Forward collision warning 13:45 Routing issue 15:10 Short speed ramp 15:33 Turn lane issue 16:40 Crosswalk yielding 19:15 Turn lane confusion 19:56 Great left turn 20:35 Red light issue 22:35 Busy area / construction 25:40 Turn signal visualization 27:10 Speed ramp 28:10 Yellow light 28:25 Open door visualization 28:45 Roundabout 1 29:56 Car pulls out in front 31:24 Roundabout 2 32:15 Turn lane issue 34:00 Cross traffic issue 34:24 Roundabout 3 35:08 Closing thoughts
 
  • Like
Reactions: Phlier and JHCCAZ
Rob Maurer from Tesla Daily, downtown Milwaukee. Overall impression: progress, but there were some issues.
3:28 Re-routing 4:37 Yielding 5:07 Bridge out 6:45 Crosswalk confusion 7:45 Cross-traffic 9:00 Bicyclists 10:40 Intersection issue 11:45 Hairpin 13:20 Forward collision warning 13:45 Routing issue 15:10 Short speed ramp 15:33 Turn lane issue 16:40 Crosswalk yielding 19:15 Turn lane confusion 19:56 Great left turn 20:35 Red light issue 22:35 Busy area / construction 25:40 Turn signal visualization 27:10 Speed ramp 28:10 Yellow light 28:25 Open door visualization 28:45 Roundabout 1 29:56 Car pulls out in front 31:24 Roundabout 2 32:15 Turn lane issue 34:00 Cross traffic issue 34:24 Roundabout 3 35:08 Closing thoughts
Maurer's drive was an unmitigated disaster. Sadly, he kept blaming the issues on "Navigation". Passing twin Do Not Enter signs to go the wrong way on a one way road is a major fail. Those are standard signs the car should be able to read. The one-way road was a temporary construction change, but, if the car is going to rely on vision, it needs to not go around a barrier against Do Not Enter signs.

Twice, the car went down a dead end. The first time in response to a road closure. It turned down a street that wasn't even on the map. Second time it just didn't make an easy right turn. That was unexpected.

Went straight from a left turn only lane. That's 10.11.2 behavior that apparently was not fixed. I was really hoping that there were improvements in that area.

Rob's drive looked like one step forward, 10 steps back. I give him credit for posting it.
 
Maurer's drive was an unmitigated disaster. Sadly, he kept blaming the issues on "Navigation". Passing twin Do Not Enter signs to go the wrong way on a one way road is a major fail. Those are standard signs the car should be able to read. The one-way road was a temporary construction change, but, if the car is going to rely on vision, it needs to not go around a barrier against Do Not Enter signs.

Twice, the car went down a dead end. The first time in response to a road closure. It turned down a street that wasn't even on the map. Second time it just didn't make an easy right turn. That was unexpected.

Went straight from a left turn only lane. That's 10.11.2 behavior that apparently was not fixed. I was really hoping that there were improvements in that area.

Rob's drive looked like one step forward, 10 steps back. I give him credit for posting it.
Many comments in the video seem to be quite positive. As far as going down the wrong way, Rob didn’t notice it at first because that is normally not the case there. And he brought up a good point that simply relying on mapping data can have shortcomings. Even Google maps still has that street wrong. AI ability to read more signs is critical.
 
In fairness, a lot of the pro-FSD argument is that it should be able to drive anywhere without needing to know anything about the roads and NOT be heavily map dependent like super geofenced stuff like Waymo.

But the beta has made it glaringly obvious Tesla relies on maps pretty heavily and isn't remotely close to a drive-anywhere-without-maps state.
 
  • Informative
Reactions: scottf200

huh, first vid I can remember from Mars where he had commentary. At 3:50 pretty impressive job moving around a double parked Amazon van with some traffic around. Probably a bit flukey, but it did handle it. The more correct thing to do would have probably been to stop further away from the van to maintain some sight of oncoming traffic, but otherwise good.

4:34 - kind of interesting to see a trjack passing on the cross street elongating on the visualization the car’s cameras see it go by and adjust the image to match the size of the semi.

7:30 - interesting to see the car not show the large container on the side of the road. I’m pretty sure my car has shown those for me as semi trucks before, but nothing on the screen here.
 
Last edited:
Rob Maurer from Tesla Daily, downtown Milwaukee. Overall impression: progress, but there were some issues.
3:28 Re-routing 4:37 Yielding 5:07 Bridge out 6:45 Crosswalk confusion 7:45 Cross-traffic 9:00 Bicyclists 10:40 Intersection issue 11:45 Hairpin 13:20 Forward collision warning 13:45 Routing issue 15:10 Short speed ramp 15:33 Turn lane issue 16:40 Crosswalk yielding 19:15 Turn lane confusion 19:56 Great left turn 20:35 Red light issue 22:35 Busy area / construction 25:40 Turn signal visualization 27:10 Speed ramp 28:10 Yellow light 28:25 Open door visualization 28:45 Roundabout 1 29:56 Car pulls out in front 31:24 Roundabout 2 32:15 Turn lane issue 34:00 Cross traffic issue 34:24 Roundabout 3 35:08 Closing thoughts

handled the roundabout pretty well. It yielded properly. The car pulling out in front was interesting. The car was displaying the car changing orientation and it looked like it was displaying it starting to encroach into the road, but didn’t have it highlighted in blue to indicate it was tracking it and didn’t seem to react.

Roundabout 2 it yielded well and then took way too long to commit after and impeded a car that it had plenty of time to get in front of smoothly.

roundabout 3 it had some issues with lane changes but I was impressed with how it didn’t just stop at the yield sign with no cars in the roundabout as I’ve seen in most cases but smoothly entered.

The FCW at 13:20 was interesting too. The delivery truck driver was kind of hopping into the truck and the car apparently thought he would come out on the other side into the road and displayed an open door for the truck on the visualization as it buzzed an alarm.
 
Last edited:
Rob Maurer from Tesla Daily, downtown Milwaukee. Overall impression: progress, but there were some issues.
3:28 Re-routing 4:37 Yielding 5:07 Bridge out 6:45 Crosswalk confusion 7:45 Cross-traffic 9:00 Bicyclists 10:40 Intersection issue 11:45 Hairpin 13:20 Forward collision warning 13:45 Routing issue 15:10 Short speed ramp 15:33 Turn lane issue 16:40 Crosswalk yielding 19:15 Turn lane confusion 19:56 Great left turn 20:35 Red light issue 22:35 Busy area / construction 25:40 Turn signal visualization 27:10 Speed ramp 28:10 Yellow light 28:25 Open door visualization 28:45 Roundabout 1 29:56 Car pulls out in front 31:24 Roundabout 2 32:15 Turn lane issue 34:00 Cross traffic issue 34:24 Roundabout 3 35:08 Closing thoughts
The navigation at 13:57 was definitely correct, so something else is going on with those last second reroutes. It seems much quicker to reroute when something problematic comes along than before, like with the road blocked by construction elsewhere in the video. So this seemed more like a false positive of a road blockage etc. Hard to tell since those don’t show up on the visualization.
 
Many comments in the video seem to be quite positive. As far as going down the wrong way, Rob didn’t notice it at first because that is normally not the case there. And he brought up a good point that simply relying on mapping data can have shortcomings. Even Google maps still has that street wrong. AI ability to read more signs is critical.
Imagine how hard it is for lidar to read signs ;).
 
Summary from Chris for city driving at 14:10 : Still needs more work, too timid, FSD needs to be aggressive for city driving. Disengagement stats versus other versions at 17:50 :
1653425938511.png
 
Last edited:
Summary from Chris for city driving at 14:10 : Still needs more work, too timid, FSD needs to be aggressive for city driving. Disengagement stats versus other versions at 17:50 :
View attachment 808332

Huh so he had lower intervention/disengagements on v8.2 and v9 than any recent 10.X version? Wow that’s disappointing progress.
 
  • Like
Reactions: Terminator857