this is 100% my experience driving FSD Beta for over 6 months ... many situations where I take control are as follows:
This attitude will eventually lead to major problems. If you trusted my car on FSD, you would be injured or dead right now.
I honestly don't understand his rant. He is acting like he doesn't understand FSDb. People who have used it awhile understands where it would fail and where it would succeed. Fail area are usually places incompatible with the map data or the car chooses the wrong lane where it sacks its probability of success. This is rather consistent and with every update we usually test the same area to see if there are any improvements.
He has a long history of both having engineering rigor and having very strong opinions.
I'm embarrassed to say that I said exactly that to the poor tech guy at Tesla about 4 years ago, after AutoPilot made an unexpected turn on a 2-lane road, because I got excited in a not-good way, but wanted to make sure they saw exactly what happened and took appropriate driver assistance programing action.
He has a long history of both having engineering rigor and having very strong opinions.
With that being said, this video is highly inconclusive and I hope he posts more videos. Looks like someone posted a comment that breaks down his issues
Would be nice to see the build #, drive at the speed limit, capture the video directly from the MCU and have a better view of the road. And in general, FSD is overly cautious currently, the beta has been extended to >1/4 million cars so safety is paramount. Yes, some turns are going to seem to take forever and there will be issues. I get frustrated as well from time to time so I get it. 0:40 NP (Neural Planner) plans a path close to the stop sign. Passes the poles fine however. Due to takeover, not sure if this would have resulted in a good or bad turn. To the left is occluded by a chainlink fence: Google Maps 1:11 Lanes net issue, detecting a turn only lane should be fed into the inference stack through sparse map data as a hint to lanes net. 3:27 NP incorrectly plans a path to the left of a waiting car. DE (Data Engine) incorrectly classifies the waiting car as parked and LN (Lanes Net) incorrectly. 3:28 The visualization can't be seen at all due to an edit so I can't tell what was happening afterwards. 3:40 After the edit, it appears that after the takeover, FSD switches to AP. You can see the lane lines that AP sees vs the FSD visualizations. At 3:51 it switches back. I've not seen this occur without a UI warning popping up so I wonder what build is being used in this video and if it has somehow been modified. 4:10 FSD seems to do the right thing as the NP decides to not go (line turns grey) when the takeover occurs. Notice that it again switches over to AP visualizations without a warning.
If it cant work in a tunnel, its pretty terrible. There are lots of tunnels in the real world, FSD should find the boring company tunnels extremely easy. No pedestrians and all vehicles are the same size/type.
A tunnel is possibly the easiest place to run FSD. OccNet is on vacation, just need lanes net.
Precisely my point and calling him a whiner. When you provide a report, you need to provide a full picture of where it works find and where it fails. Looking at umpteen videos, it appears all the vanilla drives - simple UPLs, Right turns seems to be largely intervention free. Just can't believe when he says it fails 75% of Right turns
Here's one of those strong opinions inbound. Your "diagnosis" is mostly nonsense.
It's been in the description since I posted the video.
lol. That would probably be the least safe thing someone could do around here, or in most places.
Feel free to shoot me something I can use to do this. Until then a GoPro aimed at the screen will have to suffice.
I can see the road fine...
The issue is that it was driving off the road into the adjacent parking lot. I've no idea if it would have hit the stop sign or whatever because I wanted to stay on the road.
Chainlink fences are not opaque last I checked.
The visualization is front and center throughout the entire video. The only edits made are to speed up the video, rewind and replay sections, etc. Saying I edited and hid the visualization somehow is ludicrous. It's right there the entire time. Where I rewind at 3:28, it just replays that entire section again slowly, again uncut. So don't lie.
Again, there was no edit.
Mine fails 100% of right turns.
Feel free to post another video and we'll see what changes you makeHere's one of those strong opinions inbound. Your "diagnosis" is mostly nonsense.
It's been in the description since I posted the video.
lol. That would probably be the least safe thing someone could do around here, or in most places.
Feel free to shoot me something I can use to do this. Until then a GoPro aimed at the screen will have to suffice.
I can see the road fine...
The issue is that it was driving off the road into the adjacent parking lot. I've no idea if it would have hit the stop sign or whatever because I wanted to stay on the road.
Chainlink fences are not opaque last I checked.
The visualization is front and center throughout the entire video. The only edits made are to speed up the video, rewind and replay sections, etc. Saying I edited and hid the visualization somehow is ludicrous. It's right there the entire time. Where I rewind at 3:28, it just replays that entire section again slowly, again uncut. So don't lie.
Again, there was no edit.
Any time I disengage by turning the wheel it switches back to the AP visualization for a moment. I've no idea why, but it does this on every vehicle I've used the beta in. Reengaging gets it back, or just drive a moment and it'll come back.
Nothing's been modified. In fact, this is my wife's 3 and is completely stock.
Nope!
The little blue creep wall in the visualization moved backwards, so the car decided it was no longer bound by this and fully let off the brake and started accelerating. Me clearly seeing the car to the left, my only real option was to nail it. In hindsight, should have jumped to the other lane entirely.
Again, the AP visualization thing happens on every car I've used this on when disengaging.
---
Any chance you could find someone who says FSD works great for them and have them run the same routes in their vehicle?
. 
Terrible misleading article.
Mercedes-Benz Gets Approval to Deploy Level 3 Driving Tech in Nevada
Mercedes-Benz is the first automaker to receive regulatory approval to launch a Level 3 driver assistance system in Nevada.www.thedrive.com
Mercedes is the first automaker to receive such approval in the U.S., beating out names like Tesla, General Motors, Ford, and even Honda. The auto giant holds the title as the first automaker to put out a production vehicle with Level 3 autonomy in its Japan-only Honda Legend sedan.
I find this to be 100% suspect... as in you do not have FSD B enabled... in my vehicle FSD beta handles majority of RH turns even on one way streets with cars parked to the corners of the intersection...
Yes we know that your non-Tesla car fails 100% of right turns. Also fails to lane keep 50% of the time, couldn't engage most of the time, and doesn't stop 20% of the time for the car in front. This is with over 50k miles of trying to use these other "motion drive pilot assist" or whatever the F they try to brand them.
