After watching many V13.2 videos, it is very obviously not "solved", as is clear to see by all the watching international regulators Keep pumping. Next you will be claiming that Optimus will be available to the retail trade for $20K next year and able to successfully change a daiper.
Great take on the compact car version using a controller before it morphs into a Cybercab once FSD is unsupervised! Maybe a phone app if not through the infotainment screen!
I watched one those tests. The intervention was the car stayed in the right lane going 40. So he intervened into the far left lane. And came to a complete stop. And the slow moving right lane rolled on.
@@wincoffin7985 Oh yes; I can not drive a wheelchair-van and it is difficult/often impossible to find a good driver when you need one. A FSD van would be fantastic! But what I was replying to was the conversation about if it is possible to drive an electric FSD vehicle with a joystick a bit, when FSD fails. But then.. it does take some practice, otherwise human input = error.
@@GiesbertNijhuis An EV (and by this I mean ELECTRONIC Vehicle, not Electric Vehicle) should be easier to convert to a joystick arrangement. Where the legacy auto industry has gone wrong is in assuming something like a Tesla is merely an electrically propelled version of a 100+ year old ICE technology. It is not, it carries two [for redundancy] 34 trillion operations per second super computers. If you want a graphic difference between a Tesla and any other car on the road - look at one of the light shows..! Good Luck with the mobility and hoping you will be able to get something suitable real soon.
I can think of a few things that could be tested. Obeying signs such as “do not enter” and “no turn on red”, school zone speed limits, signs that are dependent on specific times such as lanes that change direction/traffic pattern, hand signals from road workers and police officers, etiquette at dead traffic lights, those weird pedestrian lights that flash red when you’re supposed to stop and look before proceeding. Those are the kinds of situations that I would love to see tested.
I also want to see testing for cases of objects on the road including a simulated prone human, dodging holes in the roadway, handling a missing bridge span, handling a wrong way driver on a multilane freeway. We may have to trust Tesla statements they tested some of those cases in simulation.
Ten years seems a safe enough bet, but it's also plenty of time for FSD competitors to improve as well. I expect fully autonomous robotaxis (or close enough) in my lifetime, and they'll be transformative, but Tesla will have competition.
@@restonthewindi disagree. There is no one else doing what Tesla is doing. They’re all on sensor and lidar and retrofitting cars. Zoos Waymo isn’t going to scale fast enough to be able to compete with Tesla which makes their own cars built in with their own unique system that doesn’t require the time and investment that the others have
@@restonthewind If Tesla stopped innovating, there will be competition. In the meantime, everybody will be falling light years behind as Tesla moves ahead. :)
@@Raylovepalomar I don't say that Waymo is doing what Tesla is doing, so you don't disagree with me. Again, if FSD robotaxis don't exist for another decade, Waymo also has a decade to improve its hardware and software. Again, Tesla robotaxis with remote operators may cost less than Waymo. Tesla is hiring remote operators.
you will soon have 3 destination options 1: park in a space 2: drop you off at the door and than park in a space 3: drop off and continue as a robotaxi
I'm skeptical still. There needs to be more data. It's getting REALLLLLLLLLLY good however. The merge of Highway and City Streets to full on neural net appears to be the move to get it there, along with all of the additional compute they are throwing at the problem.
@@tlow5766Don’t hold your breath on 2025. Unsupervised FSD has been promised for almost a decade now. It’s definitely almost there, but it’s incredibly difficult to know how long that last .01% is going to take to solve.
@ A few weeks ago, I would have agreed. But they have started implementing new features at an insane speed (reversing, turn in a terminal road, parking etc.). They didn’t do any of that in the last 5 years. And they already started fooling on the cybercab line. They are confident, it will work.
Farzad, one thing to remember about the current release of V13, this was trained on the small portion of Cortex that is currently on line! As more of Cortex is brought on line V13 will improve exponentially... The available training compute restricts the volume of training data that can be consumed by training in a given time unit, as Cortex expands more training data will be included in the trained models refining the results...
@@MrQuay03 Actually YES exponentially! With neural networks as more training data is added you start to see emergent behavious that fill the gaps... The march of 9999s is more a heuristic code thing.
Really appreciate AIDriver's(JD) perspective on V13.2 experience. I am super excited to get OTA on my MS hopefully soon. JD is totally relevant and needs to stay around with his excellent utube content.
I disagree with the idea of a joystick in the car. I think the joystick may be something that a teleoperator uses to intervene if the car gets stuck and the rider asks for help. Having said that, it's possible that you could implement controls on either the Tesla app and/or the robo-taxi app, for the user to intervene if needed. Although there's a danger that there could be a legal implications if a user miss-controls the vehicle using those types of devices without some training and especially when the user may not even have a license to drive.
A better analogy than fighter jets and such is construction equipment. Many modern machines transitioned from a steering wheel to a side stick and steer by wire. These are machines that can drive on the road today at full speed and do so with a stick. And they've been doing it for many years so it's not at all out there.
Given the logic of fully digital imput, I think that Farzad has a great point. In fact, to extend the joystick analogy it would make much more sense for Tesla to simply make wireless controller that not only allows you to play all the games in the infotainment system but that same controller should be made to drive/take over steering for the car.
With ML and AI, people forget - "It is as bad as it will ever be right now." That's what's amazing - the change is accelerating, and improving dramatically with each release.
It’s cool that you think it’s solved. I’m skeptical bc there’s a lot of scenarios here in central MO that need a LOT of work. Ignoring speed limits signs, not staying centered on winding country highways, going too wide on a curvy road and crossing the double yellow line with oncoming traffic. I hope 13.x solves these, but I don’t think it’s OVER or SOLVED or DONE as YTers keep saying. But good job Tesla for continuous progress.
You have V13 ? If not your information is out of date LoL I agree that doesn't mean those perceived issues are absent in the current iteration of V13 however they can be tested when available... It is also true that these perceived issues may or not be crucial to unsupervised autonomy!
Just tried V13, it is solved ira just a matter in the future if it Will 2x times better than me or 10x better now. Its alresdy a better driver than me and I have been driving for 10+ years
When he said it is SOLVED, he doesn't mean that they are not going to improve it further.....They are going to improve it but the basic stuff are basically done. There are still edge cases so they still need further training.
No more FSD videos required. Onto Optimus robots 👉 I had heard that around the times of sunset, the change of lighting conditions seemed to cause some issues like phantom braking. Was that tested again? And the only other thing I have not heard anyone talk about is if there is any consideration towards updating the highway code so that signs, lines and tricky intersections, etc are improved or altered to assist FSD software. Imagine QR codes next to or a part of all signs. Perhaps for temporary ones as well to indicate construction ahead or road closures, etc.
Mine runs STOP signs. It also does not always pick up speed limit signs. Also flashing red lights still screw it up. I need to push the accelerator to get it to go.
The next challenge is auto callibration. FSD needs to be able to be dropped into any platform, with adequate compute, find the cameras, the controlls, calibrate the size, and performance of the platform and then drive it.
The car drives really good now but when it drives me home (the house is on a corner), it parks on the wrong side of house. Is there a way to put the final destination by or even in the driveway?
Welcome to Jonny Cab! All this talk of a joystick in a self driving car reminds me of the original Total Recall movie where Arnold jumps in an automated cab and the robot driver doesn’t understand when Arnold tells it to just drive. So he just rips the driver out of its mount up front and uses the joystick to drive it manually.
when I drive cross country, the most problems I have are bright moonlight causing problems and direct sunlight causing problems, which just turns off FSD altogether.
One of my questions is, the Tesla likes to drive straight down the middle of the road. Even when it’s in corners. With version 13, does it do like most normal people do and drive to the inside of the corners or does it still wanna place itself right in the middle of the road even when it’s on a corner like going around a loop to get on and on ramp for a highway
I wish you would have brought up the joystick concept with Chuck C. As a pilot, he commands a very expensive aircraft with a joystick. The lockout is a great idea when transporting kids or others that shouldn't be interferring.
The joystick makes so much sense…cheap, efficient, front load the ramp…. My FSD on V12 is phenomenal and I rarely have interventions. If Tesla isn’t headed in this direction, they should redirect because it definitely makes sense
Yeah, just use a (secure connection) Xbox controller. The military uses that type of system for so many things. Also most people (male?) under 40 have used and can use very well these types of controllers.
Farzad might be right-Tesla could include a fallback mechanism, like a joystick, for controlling the vehicle. However, from what I see, Tesla is aiming to get the RoboTaxi and Compact approved for self-driving with minimal human intervention and interaction, potentially eliminating the need for a joystick altogether. If a joystick is included, it will likely be an optional accessory you can purchase, rather than a standard feature. I believe Tesla’s ultimate goal is to achieve national approval for self-driving vehicles that function seamlessly without requiring a joystick. Given Tesla’s track record and design philosophy, a touchscreen interface may be all that’s needed. For instance, the touchscreen could allow you to interact with the vehicle to issue specific commands, like navigating around a car or parking in a certain spot. Tesla might already have metrics to determine when critical or convenience interventions are likely. This means the vehicle could run autonomously most of the time and interrupt only when necessary, prompting you for adjustments. For rare manual operations, like reversing the car to load groceries, a mobile app could provide remote control functionality. Instead of a joystick, you’d simply use your phone to direct the car while standing outside. This aligns with Tesla’s push for automation and minimalism, making a dedicated control interface unnecessary for most users.
Hey, if the edge cases are going to be so rare, why couldn’t an app on your phone get it out of being stuck, just like when using smart summon to get the requested info in?
There is a large amount of manufacturing equipment building up very quickly in a lot down in Texas. Tegmyer isn't saying it . But it obviously the next vehicle.
There will be fix functions for robo taxi to be command to move to a safe spot. Example move to the road side and pick a spot, park in a spot, you will giving an option to press on the screen and you press the button, no joystick needed.
The more data the better it gets. The odd windshield wiper thing makes sense to me as I am positive they are taking all their normal video footage and augmenting it with both rain and fog so they have 3x the training data. Their synthetic data capabilities are growing off the chart using unreal.
Its clearly way better and smoother. But still not perfect or fully solved. Watch the Dirty Tesla video from this morning. He is also very impressed with how good V13 is but there are still issues. It tried to make a right on red in a "no turn on red" intersection. It also ran a red left turn traffic light.
there are several times i needed to step in, this was in busy-ish areas. it still gets confused with lanes that expand from one to two, or faded lines. parking wise, it will avoid striped areas when you can drive over them with no danger, i.e. next to my parking spot is a striped area that i drive over all the time to park. self-park avoids it as if it's a wall so it takes longer for it to position itself to reverse straight into the stall. doesn't need to do it so i hope they include something that addresses that in the future. that and wanting to park close to one side of a stall.
Can you guys go into more detail about the changes in training process / infrastructure / NN architecture that have made v13 ao great? And what is the outlook regarding that? Or does anyone have a good link / source?
6:40 If you watch the end of @DirtyTesla most recent video. At the very end of the video there is a person that the car is interacting with that is not shown on the screen.
I like Farzad's joystick idea, but let's go one step further. A gaming controller i.e. a joystick could multitask as the steering input too. Or what about an Atari style wireless controller, that might work even better. The younger crowd have great eye hand coordination & steering a car that way would be a no brainer.
halfway through, regarding the manual control, I'd guess grok integration for sticky situation, with voice and image recognition the passenger can tell it where to go
I like Farzad’s attitude. He’s like why l wouldn’t they on the joystick thang. It’s waymolicious. That’s what innovation is based off of: almost silly ez ideas that turn out to be genius. Copy n improve to win 🥇 Japan forgot that move 😂
Is there a point coming soon where you can tell GROK where you exactly want to go? Also there is the question if it can read road signs as well like parking between certain times of day or puling over for emergency vehicles or school bus stops, things like that are still an issue. What you say?
If there will be something like an Joystick configuration I assume it will be just some sort of Bluetooth Controller Option, e.g. bring your one Controller of choice be it an PS5 or xbox one. Most people know how to Drive with these ones
Its too late they already have nothing to stand on. Its hard for people to admit defeat and its honestly even more weird to bet against future tech in the first place.
that's good news but i rarely have to take control with my current version 12. the new Summon on my 2023 model Y is amazingly good. i think the biggest complaint i have is it waits until the last second to change lanes to exit the freeway or get into the correct lane to make a turn. worse case is that i have to take over or miss the exit and it makes a u-turn; no big deal and not dangerous.
This is an off beat example but All of the motorized wheel chairs are controlled by >> yes you guessed it >> a JOY STICK !!! It works exceptionally well !!!!!
Come to Park City Utah. Tight narrow streets and snow. I have model Y and Cybertruck. The hw3 model y does better than Cybertruck on my "fsd challenge" route saved to favorites.
This interview reminds me of the scene in Blade Runner where Deckard administers the Voight Kampff test to Rachel. One day we humans will be subject to the same rigor.
Does 13.2 handle hanging barriers at all? Railroad crossing gates, chains across driveways, etc. My experience with 12.5.x is that it will try to drive right into a crossing gate.
I would love to read a book that chronicles the past decade of Tesla’s software development. The UI/UX/QA/Firmware & ADAS systems development is incredibly interesting.
I live @ 200 miles north of the US border to Canada. It’s a rural town with @ 6 months of winter and the closest public charging is @ 50 miles away. Obviously I won’t be buying an electric vehicle until they can handle the weather and the infrastructure is better. Note: I occasionally have to deal with -40 degree days. If you were in my area would you buy an electric vehicle? lol
Solved or near as dammit. So exciting!!! For me the next question is, how much of an advantage is this FSD leap to a bot navigating the real world…are the algorithms transferable? What is the timeframe to them performing complex tasks and problem solving???
I think Farzad's joystick idea is brilliant. A cheap backup that is in no way acceptable for day-to-day driving, but will be sufficient to get you out of a situation where FSD fails for some reason or another in a non-critical way-- e.g., unable to turn around in some tight space. Kind of like that simple set of controls on the back of my smart-panel TV that will allow me to turn it off and on, adjust the volume, or move a cursor around the screen if I lose the remote. Something to use "in a pinch."
@@Manwith6secondmemoryI think most people don't even like using cruise control. I think even less people will want self driving. We shall see if it ever spreads beyond the fanboys.
What I think ppl are still missing is the fact that self driving is a safety feature. Sure it seems gimmicky now but as it improves it will be like putting on a seatbelt. Ofc you would like to drive without one but it's safer to have it on. Furthermore, I don't know anyone who enjoys driving in bumper to bumper traffic so you can always have it to turn on when needed and turn it off when you can speed or whatever else ppl like about driving.
Tons of people I know still drive without seatbelts because it’s “annoying” and they assume they won’t crash. I think many people do whatever they feel regardless of safety data. And while I agree convenience is a big motivator for consumers, ego is an even greater one. People really don’t like the IDEA of robots doing tasks they take pride in, like driving. This isn’t everyone of course, just what I’ve gleaned from conversations with non-tech friends
@@MrGoggug brother where do you live, most cars have beepers if you do not put on your seatbelt. People do not like automation? Look around you everything is/has been heading in that direction
Whenever it rains or gets dark, my tesla says vision is impaired or degraded of some sort. This new version fixes that? How does it handle the heavy rain?
It seems to me that to make FSD truly viable would need to acquire some local area knowledge. We humans drive in our local area knowing best routes and techniques to handle complicated parking lots. Custom routes, especially where roads and alleys and parking lots are not included in the navigation data. Finally, I have some bright tape I use to accurately back into my garage. Some type of standard markers ( tape) to signal routing in complex area Seoul be great.
There is one hardware aspect that is a blocker for full FSD at the moment imo. If the cameras are obscured due to heavy rain, mud, snow etc. the system wont be able to work. Tesla needs to work on a camera system that can ensure that the lens is ALWAYS able so clean itself. Something like a rotating and self-cleaning and heating lens shield.
Steer-by-wire is already in CyberCab, they have production steer-by-wire ready for Model Y refresh (same as used in CyberCab sharing parts with Model Y) no one will now have an issue with no stalks as the buttons will always be near to right way up! Model S3X are now legacy products and will not progress much further...
Sounds good, but here is a challenge I’ll bet 13 won’t come close to; take me into downtown Chicago in rush hour taking the Lower Wacker drive (famous in movies for chase scenes) while losing all GPS. Now take Lake Shore drive North and drop me off at Wrigley Field. This requires very assertive driving where you will need to “nose in” to move lanes. You can’t stop and hope someone will let you in, you must assertively make a move. I am a little tired of the California driving examples. I would do it myself but I am stuck on H3. I have tried many times in v12 and didn’t get very far…
We had over 100cm (>40 inches) of snow this past week. People still had to drive. I'm very skeptical that FSD can ever handle this. Maybe it's "finally solved" in fair weather conditions only.
AI DRIVR on YT: www.youtube.com/@AIDRIVR
AI DRIVR on X: x.com/AIDRIVR
After watching many V13.2 videos, it is very obviously not "solved", as is clear to see by all the watching international regulators
Keep pumping.
Next you will be claiming that Optimus will be available to the retail trade for $20K next year and able to successfully change a daiper.
Great take on the compact car version using a controller before it morphs into a Cybercab once FSD is unsupervised! Maybe a phone app if not through the infotainment screen!
So excited for a Tesla future
I watched one those tests. The intervention was the car stayed in the right lane going 40. So he intervened into the far left lane. And came to a complete stop. And the slow moving right lane rolled on.
It's difficult to realize maybe it's right. You just don't realize why yet.
I've been driving with a joystick for 28 years! Wheelchair. With my chin. Very precise. (Spinal Cord Injury C3/4, can only move my head).
So are you imagining an increase in independence with this new FSD technology? I'm thinking it will provide this for a lot of people & seniors too.
@@wincoffin7985 Oh yes; I can not drive a wheelchair-van and it is difficult/often impossible to find a good driver when you need one. A FSD van would be fantastic! But what I was replying to was the conversation about if it is possible to drive an electric FSD vehicle with a joystick a bit, when FSD fails. But then.. it does take some practice, otherwise human input = error.
@@GiesbertNijhuis An EV (and by this I mean ELECTRONIC Vehicle, not Electric Vehicle) should be easier to convert to a joystick arrangement. Where the legacy auto industry has gone wrong is in assuming something like a Tesla is merely an electrically propelled version of a 100+ year old ICE technology. It is not, it carries two [for redundancy] 34 trillion operations per second super computers. If you want a graphic difference between a Tesla and any other car on the road - look at one of the light shows..! Good Luck with the mobility and hoping you will be able to get something suitable real soon.
Gosh your certainly very restricted.. could applying to Neuralink be worth while..good luck...@GiesbertNijhuis
I’m getting a Neuralink implant as soon as practicable.
I can think of a few things that could be tested. Obeying signs such as “do not enter” and “no turn on red”, school zone speed limits, signs that are dependent on specific times such as lanes that change direction/traffic pattern, hand signals from road workers and police officers, etiquette at dead traffic lights, those weird pedestrian lights that flash red when you’re supposed to stop and look before proceeding. Those are the kinds of situations that I would love to see tested.
It will be interesting to see how it reads road instructions. It will need some sort of Grok AI brain to read text and understand the meaning.
I suspect that some of those cases have not been worked on at all and that is a big deal!
I also want to see testing for cases of objects on the road including a simulated prone human, dodging holes in the roadway, handling a missing bridge span, handling a wrong way driver on a multilane freeway. We may have to trust Tesla statements they tested some of those cases in simulation.
@@elviswsjr all necessary before vehicles can become autonomous
@@elviswsjr I think Speed limit signs are already being read.
In less than ten years, most people will regard autonomous vehicles the way they now regard Falcon-9 launches.
Less than 10 years is a huge range. What do your analytics point to more specifically? 6? 3? 9?😊
Ten years seems a safe enough bet, but it's also plenty of time for FSD competitors to improve as well. I expect fully autonomous robotaxis (or close enough) in my lifetime, and they'll be transformative, but Tesla will have competition.
@@restonthewindi disagree. There is no one else doing what Tesla is doing. They’re all on sensor and lidar and retrofitting cars. Zoos Waymo isn’t going to scale fast enough to be able to compete with Tesla which makes their own cars built in with their own unique system that doesn’t require the time and investment that the others have
@@restonthewind If Tesla stopped innovating, there will be competition. In the meantime, everybody will be falling light years behind as Tesla moves ahead. :)
@@Raylovepalomar I don't say that Waymo is doing what Tesla is doing, so you don't disagree with me. Again, if FSD robotaxis don't exist for another decade, Waymo also has a decade to improve its hardware and software. Again, Tesla robotaxis with remote operators may cost less than Waymo. Tesla is hiring remote operators.
you will soon have 3 destination options
1: park in a space
2: drop you off at the door and than park in a space
3: drop off and continue as a robotaxi
For real!
How delusional is that. From the guy who has been promising fsd for ten years, and was sued for it. But yeah keep believing that guy
4. *wish list item* Option to open drivers side door upon arrival of summon point
Bought my first Tesla Saturday night after hearing people rave about v13
I call FSD solved. It is just debugging at this point.
I'm skeptical still. There needs to be more data. It's getting REALLLLLLLLLLY good however. The merge of Highway and City Streets to full on neural net appears to be the move to get it there, along with all of the additional compute they are throwing at the problem.
@@tapasero we are close, but even Tesla does not regard it as feature complete. But this will happen in 2025.
have a similar feeling
@@tlow5766Don’t hold your breath on 2025. Unsupervised FSD has been promised for almost a decade now. It’s definitely almost there, but it’s incredibly difficult to know how long that last .01% is going to take to solve.
@ A few weeks ago, I would have agreed. But they have started implementing new features at an insane speed (reversing, turn in a terminal road, parking etc.). They didn’t do any of that in the last 5 years. And they already started fooling on the cybercab line. They are confident, it will work.
Farzad, one thing to remember about the current release of V13, this was trained on the small portion of Cortex that is currently on line! As more of Cortex is brought on line V13 will improve exponentially... The available training compute restricts the volume of training data that can be consumed by training in a given time unit, as Cortex expands more training data will be included in the trained models refining the results...
Not exponentially, more like the march of .9999% like in every neural network.
@@MrQuay03 Actually YES exponentially! With neural networks as more training data is added you start to see emergent behavious that fill the gaps... The march of 9999s is more a heuristic code thing.
It would be good if we could get James Douma’s impressions of V13.
+
Absolutely agree!
Where is he? It’s like he fell of the earth or something
I've seen some video's from people driving through Manhattan in pooring rain and it was really incredible.. INSANE how FSD was performing..
Thanks!
You guys killed it! Thank you for us waiting V13 😊
Cyber Cab with a joystick - thats a Johnny Cab in Total Recall
Really appreciate AIDriver's(JD) perspective on V13.2 experience. I am super excited to get OTA on my MS hopefully soon. JD is totally relevant and needs to stay around with his excellent utube content.
The joystick in the middle console would make it universal for all countries no more left hand or right hand driving modifications.
I disagree with the idea of a joystick in the car. I think the joystick may be something that a teleoperator uses to intervene if the car gets stuck and the rider asks for help. Having said that, it's possible that you could implement controls on either the Tesla app and/or the robo-taxi app, for the user to intervene if needed. Although there's a danger that there could be a legal implications if a user miss-controls the vehicle using those types of devices without some training and especially when the user may not even have a license to drive.
A better analogy than fighter jets and such is construction equipment. Many modern machines transitioned from a steering wheel to a side stick and steer by wire. These are machines that can drive on the road today at full speed and do so with a stick. And they've been doing it for many years so it's not at all out there.
Given the logic of fully digital imput, I think that Farzad has a great point. In fact, to extend the joystick analogy it would make much more sense for Tesla to simply make wireless controller that not only allows you to play all the games in the infotainment system but that same controller should be made to drive/take over steering for the car.
For emergency take over if nothing else.
Voice control makes more sense to me but I can see your point
With ML and AI, people forget - "It is as bad as it will ever be right now."
That's what's amazing - the change is accelerating, and improving dramatically with each release.
It’s cool that you think it’s solved. I’m skeptical bc there’s a lot of scenarios here in central MO that need a LOT of work. Ignoring speed limits signs, not staying centered on winding country highways, going too wide on a curvy road and crossing the double yellow line with oncoming traffic. I hope 13.x solves these, but I don’t think it’s OVER or SOLVED or DONE as YTers keep saying. But good job Tesla for continuous progress.
@rickhaller3328 Amen and amen..lots of road repair and construction projects in cenral Missouri makes for complicated situations.
You have V13 ? If not your information is out of date LoL I agree that doesn't mean those perceived issues are absent in the current iteration of V13 however they can be tested when available... It is also true that these perceived issues may or not be crucial to unsupervised autonomy!
Just tried V13, it is solved ira just a matter in the future if it Will 2x times better than me or 10x better now. Its alresdy a better driver than me and I have been driving for 10+ years
When he said it is SOLVED, he doesn't mean that they are not going to improve it further.....They are going to improve it but the basic stuff are basically done. There are still edge cases so they still need further training.
That's odd, I've been in cars that can keep centered
30:00 why even a joystick when you can also just have an app?
No more FSD videos required. Onto Optimus robots 👉
I had heard that around the times of sunset, the change of lighting conditions seemed to cause some issues like phantom braking. Was that tested again?
And the only other thing I have not heard anyone talk about is if there is any consideration towards updating the highway code so that signs, lines and tricky intersections, etc are improved or altered to assist FSD software. Imagine QR codes next to or a part of all signs. Perhaps for temporary ones as well to indicate construction ahead or road closures, etc.
Mine runs STOP signs. It also does not always pick up speed limit signs. Also flashing red lights still screw it up. I need to push the accelerator to get it to go.
The next challenge is auto callibration. FSD needs to be able to be dropped into any platform, with adequate compute, find the cameras, the controlls, calibrate the size, and performance of the platform and then drive it.
That is exactly why the cybertruck took a while calibrate before it was just released.
The car drives really good now but when it drives me home (the house is on a corner), it parks on the wrong side of house. Is there a way to put the final destination by or even in the driveway?
local steering might be able to control by screen, like its now with parking.
Welcome to Jonny Cab! All this talk of a joystick in a self driving car reminds me of the original Total Recall movie where Arnold jumps in an automated cab and the robot driver doesn’t understand when Arnold tells it to just drive. So he just rips the driver out of its mount up front and uses the joystick to drive it manually.
Excellent, did you drive in direct sunlight which blinds you and maybe blind the camera
when I drive cross country, the most problems I have are bright moonlight causing problems and direct sunlight causing problems, which just turns off FSD altogether.
@@deerock7
Hw 4?
control will be via touch screen
One of my questions is, the Tesla likes to drive straight down the middle of the road. Even when it’s in corners. With version 13, does it do like most normal people do and drive to the inside of the corners or does it still wanna place itself right in the middle of the road even when it’s on a corner like going around a loop to get on and on ramp for a highway
I wish you would have brought up the joystick concept with Chuck C. As a pilot, he commands a very expensive aircraft with a joystick. The lockout is a great idea when transporting kids or others that shouldn't be interferring.
The joystick makes so much sense…cheap, efficient, front load the ramp….
My FSD on V12 is phenomenal and I rarely have interventions.
If Tesla isn’t headed in this direction, they should redirect because it definitely makes sense
Yeah, just use a (secure connection) Xbox controller. The military uses that type of system for so many things. Also most people (male?) under 40 have used and can use very well these types of controllers.
Farzad, even Hyundai has mentioned having a joystick instead of a steering wheel a few years ago.
It would make sense to clarify in the beginning of navigation if you want drop off or park. ❤
Wow. Great job guys. I was withholding judgment til JD chimed in. Thank you.
Farzad might be right-Tesla could include a fallback mechanism, like a joystick, for controlling the vehicle. However, from what I see, Tesla is aiming to get the RoboTaxi and Compact approved for self-driving with minimal human intervention and interaction, potentially eliminating the need for a joystick altogether. If a joystick is included, it will likely be an optional accessory you can purchase, rather than a standard feature. I believe Tesla’s ultimate goal is to achieve national approval for self-driving vehicles that function seamlessly without requiring a joystick.
Given Tesla’s track record and design philosophy, a touchscreen interface may be all that’s needed. For instance, the touchscreen could allow you to interact with the vehicle to issue specific commands, like navigating around a car or parking in a certain spot. Tesla might already have metrics to determine when critical or convenience interventions are likely. This means the vehicle could run autonomously most of the time and interrupt only when necessary, prompting you for adjustments.
For rare manual operations, like reversing the car to load groceries, a mobile app could provide remote control functionality. Instead of a joystick, you’d simply use your phone to direct the car while standing outside. This aligns with Tesla’s push for automation and minimalism, making a dedicated control interface unnecessary for most users.
Hey, if the edge cases are going to be so rare, why couldn’t an app on your phone get it out of being stuck, just like when using smart summon to get the requested info in?
When the car gets to your house, does it have the ability to open the garage door drive in and close the garage door?
Why joystick when you would have that visualization control on the screen
There is a large amount of manufacturing equipment building up very quickly in a lot down in Texas. Tegmyer isn't saying it . But it obviously the next vehicle.
There will be fix functions for robo taxi to be command to move to a safe spot. Example move to the road side and pick a spot, park in a spot, you will giving an option to press on the screen and you press the button, no joystick needed.
i love hearing this. i so am thrilled what Elon and Tesla has done!!! Bless you all so much! Thank😅-you
I was screaming at my phone , then you at last said it😅why need a joystick if already have touch screen🤣
looking forward to the rollout on my cybertruck
The more data the better it gets. The odd windshield wiper thing makes sense to me as I am positive they are taking all their normal video footage and augmenting it with both rain and fog so they have 3x the training data. Their synthetic data capabilities are growing off the chart using unreal.
Its clearly way better and smoother. But still not perfect or fully solved. Watch the Dirty Tesla video from this morning. He is also very impressed with how good V13 is but there are still issues. It tried to make a right on red in a "no turn on red" intersection. It also ran a red left turn traffic light.
What will be the effect on car insurance as accidents start going away?
16:18 great example. How does it react to motorcycles lane sharing in heavy traffic?
Joy Stick driving has been around for over 20 years, John Deer bulldozer. Works great for me.
if only Biden would pardon all our interventions.
there are several times i needed to step in, this was in busy-ish areas. it still gets confused with lanes that expand from one to two, or faded lines. parking wise, it will avoid striped areas when you can drive over them with no danger, i.e. next to my parking spot is a striped area that i drive over all the time to park. self-park avoids it as if it's a wall so it takes longer for it to position itself to reverse straight into the stall. doesn't need to do it so i hope they include something that addresses that in the future. that and wanting to park close to one side of a stall.
Can you guys go into more detail about the changes in training process / infrastructure / NN architecture that have made v13 ao great? And what is the outlook regarding that?
Or does anyone have a good link / source?
6:40
If you watch the end of @DirtyTesla most recent video. At the very end of the video there is a person that the car is interacting with that is not shown on the screen.
I like Farzad's joystick idea, but let's go one step further. A gaming controller i.e. a joystick could multitask as the steering input too. Or what about an Atari style wireless controller, that might work even better. The younger crowd have great eye hand coordination & steering a car that way would be a no brainer.
halfway through, regarding the manual control, I'd guess grok integration for sticky situation, with voice and image recognition the passenger can tell it where to go
A joystick would obvious be for limited operational movements, like non-road related activities, such as garaging, servicing etc
Thanks guys!
I like Farzad’s attitude. He’s like why l wouldn’t they on the joystick thang. It’s waymolicious. That’s what innovation is based off of: almost silly ez ideas that turn out to be genius. Copy n improve to win 🥇 Japan forgot that move 😂
why joy stick when you have big touch screen
I have FSD v12 on my Cybertruck, it went thru red light twice in a couple of hundred miles of driving
I’ve been waiting for Ai driver. He’s my fav
Is there a point coming soon where you can tell GROK where you exactly want to go? Also there is the question if it can read road signs as well like parking between certain times of day or puling over for emergency vehicles or school bus stops, things like that are still an issue. What you say?
This is all cause for great optimism but it won’t go wide as Ashok said unless they can roll it out to HW3 (the great majority of the fleet).
If there will be something like an Joystick configuration I assume it will be just some sort of Bluetooth Controller Option, e.g. bring your one Controller of choice be it an PS5 or xbox one. Most people know how to Drive with these ones
So nice to see how far it’s coming. Soon the naysayers will have no ground to stand on at all.
That won’t stop them 😂
@ very true 🤣
They’ll just go quiet.
We haven’t heard from Gordon recently 🤣
Its too late they already have nothing to stand on. Its hard for people to admit defeat and its honestly even more weird to bet against future tech in the first place.
that's good news but i rarely have to take control with my current version 12. the new Summon on my 2023 model Y is amazingly good. i think the biggest complaint i have is it waits until the last second to change lanes to exit the freeway or get into the correct lane to make a turn. worse case is that i have to take over or miss the exit and it makes a u-turn; no big deal and not dangerous.
Nice to see the face behind the Channel
This is an off beat example but All of the motorized wheel chairs are controlled by >> yes you guessed it >> a JOY STICK !!! It works exceptionally well !!!!!
Come to Park City Utah. Tight narrow streets and snow. I have model Y and Cybertruck.
The hw3 model y does better than Cybertruck on my "fsd challenge" route saved to favorites.
Have to see it happen on my car to believe it
Love you guys! Thanks for the awesome content, exciting times
The potential of XAI501x is unreal! Excited to see where this goes after watching your video!
Lol, took this serious for a second.
This interview reminds me of the scene in Blade Runner where Deckard administers the Voight Kampff test to Rachel.
One day we humans will be subject to the same rigor.
Does 13.2 handle hanging barriers at all? Railroad crossing gates, chains across driveways, etc. My experience with 12.5.x is that it will try to drive right into a crossing gate.
Biggest problem on Hardware three right now is excessive slowing down when approaching most crosswalks. I believe it thinks their speed bumps.
Curious how they will solve the issue with direct sunlight
I would love to read a book that chronicles the past decade of Tesla’s software development.
The UI/UX/QA/Firmware & ADAS systems development is incredibly interesting.
I live @ 200 miles north of the US border to Canada. It’s a rural town with @ 6 months of winter and the closest public charging is @ 50 miles away. Obviously I won’t be buying an electric vehicle until they can handle the weather and the infrastructure is better. Note: I occasionally have to deal with -40 degree days. If you were in my area would you buy an electric vehicle? lol
@@kipcoulter2641 if you have a charger inside a garage to store it, you'd be fine. It's the most popular vehicle in Norway 🤷
I would, yes. Our only vehicle in Minnesota is an EV. You only need public charging for road trips.
Solved or near as dammit. So exciting!!!
For me the next question is, how much of an advantage is this FSD leap to a bot navigating the real world…are the algorithms transferable? What is the timeframe to them performing complex tasks and problem solving???
We can already move the car in the app, send the car places and summon it. The best part is no part…
Joystick? No. RC Car controller for the cybercab? I can totally see that.
Great video JD, I can't wait for the AI DRIVR video!
For park-to-park, does it actually park on a house driveway? With v12 it didn't even recognize it as a parking spot.
I think Farzad's joystick idea is brilliant. A cheap backup that is in no way acceptable for day-to-day driving, but will be sufficient to get you out of a situation where FSD fails for some reason or another in a non-critical way-- e.g., unable to turn around in some tight space.
Kind of like that simple set of controls on the back of my smart-panel TV that will allow me to turn it off and on, adjust the volume, or move a cursor around the screen if I lose the remote.
Something to use "in a pinch."
You've neglected to realize that many drivers enjoy driving, at least some of the time. It will take a generation till that becomes truly rare.
People will choose convenience over effort. Always, this is always the case. We will pick the easiest option
@@Manwith6secondmemoryI think most people don't even like using cruise control. I think even less people will want self driving. We shall see if it ever spreads beyond the fanboys.
What I think ppl are still missing is the fact that self driving is a safety feature. Sure it seems gimmicky now but as it improves it will be like putting on a seatbelt. Ofc you would like to drive without one but it's safer to have it on. Furthermore, I don't know anyone who enjoys driving in bumper to bumper traffic so you can always have it to turn on when needed and turn it off when you can speed or whatever else ppl like about driving.
Tons of people I know still drive without seatbelts because it’s “annoying” and they assume they won’t crash. I think many people do whatever they feel regardless of safety data. And while I agree convenience is a big motivator for consumers, ego is an even greater one. People really don’t like the IDEA of robots doing tasks they take pride in, like driving. This isn’t everyone of course, just what I’ve gleaned from conversations with non-tech friends
@@MrGoggug brother where do you live, most cars have beepers if you do not put on your seatbelt. People do not like automation? Look around you everything is/has been heading in that direction
Berkeley hills are hard. What about Boston?
If there’s a joystick that would mean every compact car will come with FSD.
Screen based steering and acceleration control.
🤣🤣🤣🤣 the joystick rant was sooo funny. You guys definitely investigated that 😂
I encourage you to take a road trip to Manhattan New York for the ultimate FSD test.
Whenever it rains or gets dark, my tesla says vision is impaired or degraded of some sort. This new version fixes that? How does it handle the heavy rain?
Just when it’s dark? That’s abnormal
I sometimes get that warning but I ignore it and FSD just continues to operate fine.
It seems to me that to make FSD truly viable would need to acquire some local area knowledge. We humans drive in our local area knowing best routes and techniques to handle complicated parking lots. Custom routes, especially where roads and alleys and parking lots are not included in the navigation data. Finally, I have some bright tape I use to accurately back into my garage. Some type of standard markers ( tape) to signal routing in complex area Seoul be great.
15:38 How do you know these things? You just make it up. Incredible. Btw I'm here only for AI Driver if you are wondering.
Try driving from San Diego to LA and find out where you nearly have an accident because it gets confused. 3 hot spots.
And most people don't realize it's coming.... Until they see an optimus walking down the street.
@@mattmcfly2165 We won’t see Optimus walking down the street. The lizard people will put skin suits on Optimus.
Or, to add a joystick seamlessly, just update the scroll wheel slightly.
There is one hardware aspect that is a blocker for full FSD at the moment imo. If the cameras are obscured due to heavy rain, mud, snow etc. the system wont be able to work. Tesla needs to work on a camera system that can ensure that the lens is ALWAYS able so clean itself. Something like a rotating and self-cleaning and heating lens shield.
Also sun blindness. FSD will disengage with an alarm when hit with the low winter sun on an east coast afternoon.
So like eyes
@@nyanbrox5418 yes but when there is no steering wheel the car cannot move until the cameras vision is restored.
@@timelschner8451 Your idea about rotating and self-heating lensshields is excellent! Have you already sent it to Tesla????
@@andre495 I dont think Tesla accepts any external ideas since they do not want any questions about possible ownership later on. ;)
Steer-by-wire is already in CyberCab, they have production steer-by-wire ready for Model Y refresh (same as used in CyberCab sharing parts with Model Y) no one will now have an issue with no stalks as the buttons will always be near to right way up! Model S3X are now legacy products and will not progress much further...
Sounds good, but here is a challenge I’ll bet 13 won’t come close to; take me into downtown Chicago in rush hour taking the Lower Wacker drive (famous in movies for chase scenes) while losing all GPS. Now take Lake Shore drive North and drop me off at Wrigley Field. This requires very assertive driving where you will need to “nose in” to move lanes. You can’t stop and hope someone will let you in, you must assertively make a move. I am a little tired of the California driving examples. I would do it myself but I am stuck on H3. I have tried many times in v12 and didn’t get very far…
We had over 100cm (>40 inches) of snow this past week. People still had to drive. I'm very skeptical that FSD can ever handle this. Maybe it's "finally solved" in fair weather conditions only.
All AirBus airliners have been controlled with joysticks alone for decades. It's completely not ridiculous.