As for your title question asking if this version is ready for wider release, the answer is ready or not, here they come…I just received this version as a software update today and it’s my initial FSD beta inclusion.
it does quite well with low speed and good visibility but still does some wild things at higher speed and tricky visibility. I imagine it will still struggle with Kim Paguette's neighborhood and other places.
The non-recognition of an unobstructed view of a standard octagonal red stop sign tilted at a 60°, at 8:10, is concerning. That's the most recognizable of traffic signs for humans, and recognizing it is critically important. The sign shouldn't be tilted, but a program that relies on sign recognition shouldn't be so easily defeated.
Following on from your point, while the stop sign is tilted, every letter is clearly visible. If the car cannot handle a stop sign at a tilt, then it doesn't have a chance of handling stop signs covered in snow and leaves.
@2:20 + I think you had it right the first time...not the safest. A car that does not "take its turn" at an intersection is going to provoque uncertainty, confusion, irritation and possibly rash action by other drivers. "An abundance of caution" is not an absolute virtue as is so often assumed.
definitely a system problem - depends on whether it is video processor issue or corrupt info being fed to video processor as to impact on it's driving capability....
Perfect? Not at all. But getting pretty close to good enough in a lot of situations. Looking forward to more drives, especially (a) city-canyon(s) run! Thanks Frenchie 🙏!
Big fan of FSD Beta but it’s a far cry from being ready for prime time. I commend you for using it in traffic since most RUclips videos are users in little to no traffic. I’ve been using it from week one, and have yet to go 3 miles without an intervention. I won’t even use it with a passenger in the car because it’s so unnerving to them. That said, it’s way better than the earlier versions, so progress is being made even when it’s hard to tell from one version to the next.
Issues with the speed bump are 1 - it doesn't see and 2 - if it did it would go over at 17mph minimum and you went 7mph... They need to do something about this
Also not sure it saw the yellow diamond-shaped "SPEED BUMP AHEAD" sign, which could have been a cue to drive slowly along that block even if the speed bump were invisible.
@@boba6002 ya... That's a major issue - Elon talks so much about how roads are meant for eyes and brains but so many signs are ignored/not seen/definitely not visualized
wow great vid! the whole drive gave me anxiety. the left turns are cutting corners sharp. i can not stop looking at surrounding areas of your drive sorry for it the same homeless tent at 2:24.
I feel that it would be fairly trivial to take the suspension g forces to pinpoint where speed bumps and potholes are and send that date back to mummy for inclusion in updates. Although it wouldn’t help the first encounters it would be a lot better than relying on visually locating the suspension-wrecking high speed hits.
True but wouldn't that requires Tesla to store and manage all of that? Basically creating a map of speed bumps everywhere a Tesla goes? I think they decided to consider each drive as a "new drive" without relying on previous data or map data (Elon said it would be coming soon).
@@FrenchieFSD Given that deep potholes and vicious speedbumps are proliferating all over the world they are rapidly becoming a non-trivial issue for vehicles’ suspensions. Drivers are now having to be very aware of them: hitting a pothole at speed carries a real risk of wrecking tyre, wheel and/or suspension and unless autonomous cars can somehow equally miss them it’s a big problem. Certainly mapping potholes isn’t possible but speedbumps could be. I’ve watched many beta videos where the cars have crossed speedbumps at speeds that would bend suspensions with the heights that they are in many parts of the world. I think in the States they seem much lower?
@10:22 FSD does a better job at this intersection. Performance seems to be highly variable. My conclusion: Too variable for "average" driver. Definitely not ready for wide release. Will it every be? Jury is still out.
Not sure if this is new or I just noticed it more on this journey, but when ego got it wrong, it's own correction seemed very close to the intervention you would have done if you'd have interviened (if that makes sense...?). It also is behaving more and more humanly. If feels like it's getting close to actual FSD for large chunks of the journey.
I reported it multiple times to Tesla via email. The other thing is also, if I take over, that gets reported to Tesla automatically. I'm hopeful it will be fixed in future versions.
And the answer about your "question title" is ?!?? I suppose "no" ? Always happy to discover FSD beta because for France, we have to be satisfied with such an ugly option that cost 3.700 € for ... nothing useful compared to the first level (3.800 €) with "highway FSD" which is very useful. I'm preparing a video about that because it's very strange to pay for something not possible/available.
Very useful testing, but I gotta disagree that its level of caution is "pretty good" in those busy intersections. When it's creeping farther and farther into traffic but not decisively moving along once it has the right of way, it's confusing all the other drivers and making a fender bender more likely as people start proceeding out of turn.
I think they have to do a wide release, from a business perspective. TSLA isn't looking so great at the moment and the whole TSLA/Tesla boom was mainly based on selfdriving cars within a couple years. This wide release might give retail the impression they "got somewhere" and pump that share price (as very much needed), including the FSD take rate, which dropped from 80-50% to 15-10%. Its a risk that outways the issues Tesla might run into, if they would have decided to just struggle along like they have been doing for the past years (in regards to FSD). Eventually people want to see actions / results, rather than promises. A wide release is a good way to collect more data faster (which they likely will use as the argument for it), but it's more likely to be the above. Desperation and fingers crossed customers will all use FSD wisely. Note : "available for all customers who have purchased before the end of 2022." Tesla is looking for way more not-yet-functioning-FSD sales in the upcoming months, due to the massive decline in the FSD take-rate. You can't make such sales when only a limited amount of beta-testers get access. If I'm correct with my statement, this either gives a needed boost of revenues for further development of FSD, or (when crashes start to accure due to misuse) it will bring back FSD to the slowest develoment speed we have ever seen. Worst case, it's over. So... fingers crossed.
I like this theory. I would also think they would bump the price of FSD to $20k. From a psychological point of view, it seems to me to be the highest someone would pay for any "add-on" in a car. I can be wrong though. Maybe people will see the benefit of L3/4/5 and pay a lot more for it but I doubt it.
@@FrenchieFSD Thank you, Frenchie. When/if FSD gets to L4-5, I also think they will bumb the price, but perhaps even way more than we expect. It will be a no-brainer for businesses that rely on people to drive cars, even at $100K. If FSD would cost sub $30K, everyone with half a brain would buy a Tesla to let it run as a taxi service, the second they see how lucrative this businessmodel is. When the prices for these taxi services are sharp for consumers, one might not even want his/her own car anymore. It should make all of us wonder, why Tesla would even sell this to people for sub $100K and not just run their own robot taxi service. Therefor I somewhat have a hard time believing the average person can eventually buy a L4-5 selfdriving option/car, just for personal use. Since the early days of FSD development, I constandly have the feeling we might just be following the plan below : Let retail fund the development, then raise the prices (including subscription models) to the actual value of a selfdriving car, the closer it gets to a working product / till it works. FSD is non-transferable as far as I know, so by the time the average person (who purchased FSD at these lower prices) can use FSD to its full potential, he/she is ready to buy a new car and with that, will likely no longer be able to afford/justify FSD for personal use. I seriously hope Tesla will announce a *consumer* and *business* purchase/subscription model. Perhaps then FSD could be around $20K for the consumer.
@@bigdougscommentary5719 TSLA is highly overvalued and FOMO surely gained TSLA the most, so one could make the argument that FOMO (based on other promises made during FSD development) is a massive driving factor for the last couple years. Toggling the chart to log-scale and mapping announcements that Musk made (and after personally analizing years of market sentiment), does reflect it was mainly based on selfdriving cars. But I would love to hear your argument, though. Hard to work with a "you are wrong", without being provided with a decent argument, ofcourse.
It’s a lot better than older versions where I live, but it’s still not that great. Completely mishandles lane picking and badly misses some turns on my daily drives.
if it cannot find unmarked speed bumps then it is hopeless on Indian roads as of now😂 in india there are more unmarked roads then marked one.......but i guess indian roads are gonna help a lot in the development of the self driving ai because if it comes to india then it's gonna get a lot of data for unmarked roads....forget about unmarked roads noone follow the traffic rules and even on marked roads people don't care about the markings
Apparently the answer was yes as it goes to new people lmao
Excited to have joined! With two 0 intervention drives to start!
As for your title question asking if this version is ready for wider release, the answer is ready or not, here they come…I just received this version as a software update today and it’s my initial FSD beta inclusion.
Its a learning algorothm, the more testers the faster it improves
it does quite well with low speed and good visibility but still does some wild things at higher speed and tricky visibility. I imagine it will still struggle with Kim Paguette's neighborhood and other places.
Incredible valuable to see comparisons of such good edge cases as closed roads and the edge edge crooked stop sign case :)
If you try the crooked stop sign at different time of day with new sun angle, it may work fine
The non-recognition of an unobstructed view of a standard octagonal red stop sign tilted at a 60°, at 8:10, is concerning. That's the most recognizable of traffic signs for humans, and recognizing it is critically important. The sign shouldn't be tilted, but a program that relies on sign recognition shouldn't be so easily defeated.
True, anything resembling a stop sign in an intersection should be treated like a stop. Situational awareness
Following on from your point, while the stop sign is tilted, every letter is clearly visible. If the car cannot handle a stop sign at a tilt, then it doesn't have a chance of handling stop signs covered in snow and leaves.
The training set would need to be seeded with damaged signs, because of how rarely they’re encountered.
謝謝！
Thank you!!!
Yes; I just got the software update with a 97 safety score.
@2:20 + I think you had it right the first time...not the safest. A car that does not "take its turn" at an intersection is going to provoque uncertainty, confusion, irritation and possibly rash action by other drivers. "An abundance of caution" is not an absolute virtue as is so often assumed.
There were plenty of intersections where the car was going to go, but a human driver didn’t wait their turn.
Merci pour ta vidéo. Version en net progrès mais il y a encore quelques bug. Les parcours en ville sont les plus complexes à réaliser
The graphics seems to be artifacting a lot here. Could be an issue.
definitely a system problem - depends on whether it is video processor issue or corrupt info being fed to video processor as to impact on it's driving capability....
@@rohansprenger6902 this is an issue with the HDMI recording straight from the screen. Has nothing to do with what's seen on the actual screen.
@@jakesta11 thank
Yeah it's not even the HDMI. It's the encoding from my laptop. Will fix next time.
the amount of people on the streets makes me nervous... I guess I need to try mine around pedestrians. thanks for the video.
I think obstacles that even humans fail at, should not be counted as disengagement/fails, such as the speed bump
Perfect? Not at all. But getting pretty close to good enough in a lot of situations. Looking forward to more drives, especially (a) city-canyon(s) run! Thanks Frenchie 🙏!
Big fan of FSD Beta but it’s a far cry from being ready for prime time. I commend you for using it in traffic since most RUclips videos are users in little to no traffic. I’ve been using it from week one, and have yet to go 3 miles without an intervention. I won’t even use it with a passenger in the car because it’s so unnerving to them. That said, it’s way better than the earlier versions, so progress is being made even when it’s hard to tell from one version to the next.
Issues with the speed bump are 1 - it doesn't see and 2 - if it did it would go over at 17mph minimum and you went 7mph... They need to do something about this
Also not sure it saw the yellow diamond-shaped "SPEED BUMP AHEAD" sign, which could have been a cue to drive slowly along that block even if the speed bump were invisible.
@@boba6002 ya... That's a major issue - Elon talks so much about how roads are meant for eyes and brains but so many signs are ignored/not seen/definitely not visualized
Did you already ask the city to mark that speed bump :-)? Tx for the video!
I did. And the defective stop sign too :(
wow great vid! the whole drive gave me anxiety. the left turns are cutting corners sharp. i can not stop looking at surrounding areas of your drive sorry for it the same homeless tent at 2:24.
I feel that it would be fairly trivial to take the suspension g forces to pinpoint where speed bumps and potholes are and send that date back to mummy for inclusion in updates. Although it wouldn’t help the first encounters it would be a lot better than relying on visually locating the suspension-wrecking high speed hits.
True but wouldn't that requires Tesla to store and manage all of that? Basically creating a map of speed bumps everywhere a Tesla goes? I think they decided to consider each drive as a "new drive" without relying on previous data or map data (Elon said it would be coming soon).
@@FrenchieFSD Given that deep potholes and vicious speedbumps are proliferating all over the world they are rapidly becoming a non-trivial issue for vehicles’ suspensions. Drivers are now having to be very aware of them: hitting a pothole at speed carries a real risk of wrecking tyre, wheel and/or suspension and unless autonomous cars can somehow equally miss them it’s a big problem. Certainly mapping potholes isn’t possible but speedbumps could be.
I’ve watched many beta videos where the cars have crossed speedbumps at speeds that would bend suspensions with the heights that they are in many parts of the world. I think in the States they seem much lower?
@@Wol747 depends which ones but you are right some are lower
@10:22 FSD does a better job at this intersection. Performance seems to be highly variable. My conclusion: Too variable for "average" driver. Definitely not ready for wide release. Will it every be? Jury is still out.
Not sure if this is new or I just noticed it more on this journey, but when ego got it wrong, it's own correction seemed very close to the intervention you would have done if you'd have interviened (if that makes sense...?). It also is behaving more and more humanly. If feels like it's getting close to actual FSD for large chunks of the journey.
That stop sign was bent over. You should send a snapshot of that in to Tesla.
I reported it multiple times to Tesla via email. The other thing is also, if I take over, that gets reported to Tesla automatically. I'm hopeful it will be fixed in future versions.
Thank you! :D
Nice
awesome vid! are the glitches on fsd screen tesla or your screen recorder..?
I think it's bad encryption on my side.I'll make sure it is fixed moving forward.
And the answer about your "question title" is ?!?? I suppose "no" ? Always happy to discover FSD beta because for France, we have to be satisfied with such an ugly option that cost 3.700 € for ... nothing useful compared to the first level (3.800 €) with "highway FSD" which is very useful. I'm preparing a video about that because it's very strange to pay for something not possible/available.
I agree with you. I would only buy Enhanced Autopilot in Europe right now. FSD package would be a waste of money.
great videos...
Very useful testing, but I gotta disagree that its level of caution is "pretty good" in those busy intersections. When it's creeping farther and farther into traffic but not decisively moving along once it has the right of way, it's confusing all the other drivers and making a fender bender more likely as people start proceeding out of turn.
It has to be cautious because the other drivers didn’t adhere to driving rules.
I think they have to do a wide release, from a business perspective. TSLA isn't looking so great at the moment and the whole TSLA/Tesla boom was mainly based on selfdriving cars within a couple years. This wide release might give retail the impression they "got somewhere" and pump that share price (as very much needed), including the FSD take rate, which dropped from 80-50% to 15-10%. Its a risk that outways the issues Tesla might run into, if they would have decided to just struggle along like they have been doing for the past years (in regards to FSD). Eventually people want to see actions / results, rather than promises. A wide release is a good way to collect more data faster (which they likely will use as the argument for it), but it's more likely to be the above. Desperation and fingers crossed customers will all use FSD wisely.
Note : "available for all customers who have purchased before the end of 2022."
Tesla is looking for way more not-yet-functioning-FSD sales in the upcoming months, due to the massive decline in the FSD take-rate.
You can't make such sales when only a limited amount of beta-testers get access.
If I'm correct with my statement, this either gives a needed boost of revenues for further development of FSD, or (when crashes start to accure due to misuse) it will bring back FSD to the slowest develoment speed we have ever seen.
Worst case, it's over. So... fingers crossed.
I like this theory. I would also think they would bump the price of FSD to $20k. From a psychological point of view, it seems to me to be the highest someone would pay for any "add-on" in a car. I can be wrong though. Maybe people will see the benefit of L3/4/5 and pay a lot more for it but I doubt it.
@@FrenchieFSD Thank you, Frenchie. When/if FSD gets to L4-5, I also think they will bumb the price, but perhaps even way more than we expect.
It will be a no-brainer for businesses that rely on people to drive cars, even at $100K. If FSD would cost sub $30K, everyone with half a brain would buy a Tesla to let it run as a taxi service, the second they see how lucrative this businessmodel is. When the prices for these taxi services are sharp for consumers, one might not even want his/her own car anymore. It should make all of us wonder, why Tesla would even sell this to people for sub $100K and not just run their own robot taxi service.
Therefor I somewhat have a hard time believing the average person can eventually buy a L4-5 selfdriving option/car, just for personal use.
Since the early days of FSD development, I constandly have the feeling we might just be following the plan below :
Let retail fund the development, then raise the prices (including subscription models) to the actual value of a selfdriving car, the closer it gets to a working product / till it works.
FSD is non-transferable as far as I know, so by the time the average person (who purchased FSD at these lower prices) can use FSD to its full potential, he/she is ready to buy a new car and with that, will likely no longer be able to afford/justify FSD for personal use.
I seriously hope Tesla will announce a *consumer* and *business* purchase/subscription model. Perhaps then FSD could be around $20K for the consumer.
Anyone who knows anything about Tesla stock, knows FSD is NOT the driving factor in its valuation.
@@bigdougscommentary5719 TSLA is highly overvalued and FOMO surely gained TSLA the most, so one could make the argument that FOMO (based on other promises made during FSD development) is a massive driving factor for the last couple years. Toggling the chart to log-scale and mapping announcements that Musk made (and after personally analizing years of market sentiment), does reflect it was mainly based on selfdriving cars.
But I would love to hear your argument, though. Hard to work with a "you are wrong", without being provided with a decent argument, ofcourse.
At 4:00 the people are just acting like you aren't there waiting to go. I counted at least 3 cars that showed up after you and just proceeded.
But it needs to just go.... Needs to be more assertive
What city is this? I find the road lay out and marking is less defined than when I live.
Chicago
AWESOME
Wow, human drivers are going to have to get a whole lot better.
It would have been awesome if FSD could rely on Tesla Vision only but I feel it needs excellent map data to supplement real-time navigation.
But map data is not kept up to date on a consistent basis.
10.12.2 has tried to kill me on a daily basis. It’s more dangerous than any other recent versions.
hmm tesla visualization has 999 ping haha.
This might have to do with the plug as i didn't see anyone else having problems like this with the new update.
It’s a lot better than older versions where I live, but it’s still not that great. Completely mishandles lane picking and badly misses some turns on my daily drives.
At 3:42 you passed by a i8
The answer is NO. THere is a lot more word to be done here.
just watched a great vid on black tesla on utube posted 6-3 cool ride on a sharp turn road. check it!!
Humans were the main problems on this drive. Cyclists not following road rules, people jumping their turn at 4 way stops.
if it cannot find unmarked speed bumps then it is hopeless on Indian roads as of now😂 in india there are more unmarked roads then marked one.......but i guess indian roads are gonna help a lot in the development of the self driving ai because if it comes to india then it's gonna get a lot of data for unmarked roads....forget about unmarked roads noone follow the traffic rules and even on marked roads people don't care about the markings