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

FSD Beta 10.69

This site may earn commission on affiliate links.
Here we go! Hopefully they are applying my attention score metric for release priority! Seems like a great idea, if I do say so myself.:D

I would assume most of these are employees? No idea the numbers there (don’t keep track, I know some do) but that is what would make sense.

 
Last edited:
Here we go! Hopefully they are applying my attention score metric for release priority! Seems like a great idea, if I do say so myself.:D

I would assume most of these are employees? No idea the numbers there (don’t keep track, I know some do) but that is what would make sense.

Yeah, the thing about it being a "major release" was in fact mentioned before. The limited rollout over several weeks? That's brand new.
 
This is completely standard practice and what they do with most releases. It should surprise no one in the FSD Beta tester group.
That is correct. Although this is the first time that I have seen the number of cars specified for each release. It makes sense though with ~10x increase for each iteration.

I'm expecting a slower than average rollout as well given the supposed large number of changes. This should lead to larger numbers of serious bugs identified in the first two releases, which may take longer to fix.
 
  • Upgraded the occupancy network to detect dynamic objects and improved performance by adding a video module, tuning the loss function, and adding 37k new clips to the training set
Ashok Elluswamy's keynote at CVPR'22 WAD focused on the occupancy network improvements.

Not only is it doing static object predictions, it now handles dynamic objects to predict future occupancy based on how things might flow. Additionally, they're using 3D occupancy to determine if the view is occluded useful for unprotected turns they're focusing on with 10.69:
occupancy occlusion.jpg
 
Tesla has shown no interest in an L3 system. If that's your hope, and do not think that Tesla can achieve at least L4 with your car, then you will never be happy.
Elon may have never had interest in anything other than L5. That said, I see no evidence that Tesla is building anything other than, at best, an L3 system, and maybe L3 for highways only. Current Tesla platform will never be driverless, i.e. L4 or L5.

But perhaps your right, if they continue to bang their heads against Autosteer on City Streets and never improve Navigate on Autopilot to be L3, then I won’t be happy.

BTW, while not fully functional now, not Cadillac SuperCruise and the Mercedes system are designed to be L3 highway systems, and will both eventually achieve those goals. It’s looking like right now that will be before Tesla.
 
BTW, while not fully functional now, not Cadillac SuperCruise and the Mercedes system are designed to be L3 highway systems, and will both eventually achieve those goals. It’s looking like right now that will be before Tesla.
err no, that is not accurate at all
Supercruise is handsfree, not L3. You are still driving and paying attention, the car uses cameras to make sure you are looking. Just like Tesla uses steering wheel torque to check.
Mercedes does have a limited (<30mph) L3 system, but it really only applies to slow stop/go traffic. In any other situation its just like everyone else and not L3.
 
err no, that is not accurate at all
Supercruise is handsfree, not L3. You are still driving and paying attention, the car uses cameras to make sure you are looking. Just like Tesla uses steering wheel torque to check.
Mercedes does have a limited (<30mph) L3 system, but it really only applies to slow stop/go traffic. In any other situation its just like everyone else and not L3.
It is certainly possible that Mercedes will expand their L3 ODD to normal highway speeds. At least, if it is permitted in Europe. And, I would expect GM to have goals to ultimately have L3, or higher, systems. How long these may take is unknown, and neither company has announced any sort of projected time frame for this so far as I know.
 
Ashok Elluswamy's keynote at CVPR'22 WAD focused on the occupancy network improvements.

Not only is it doing static object predictions, it now handles dynamic objects to predict future occupancy based on how things might flow. Additionally, they're using 3D occupancy to determine if the view is occluded useful for unprotected turns they're focusing on with 10.69:
View attachment 843073
This was pretty interesting. It’s always unclear exactly what is running in the car. Is this running now or not (he may have said; if so, I missed it)?

The one thing that was interesting is it looked like the range of the occupancy network in the videos was something like 100m (just my estimate from what I saw; hard to tell exactly). That’s not enough for a lot of turns, so I wonder what sort of secret sauce they add to get the 200m+ detection they have.
 
Last edited:
Yes, but you’re hypothesizing about how GM’s, MB’s and Tesla’s systems are designed as well as what will happen in the future so people can say you’re wrong with just as much accuracy as you can say you’re right.
I think one can reasonably derive from statements from each company and knowledge of the hardware/autonomous systems being used by each manufacturer that GM SuperCruise and Mercedes DrivePilot are being conservatively developed to achieve L3 highway driving. Compare this with Tesla that seems to have abandoned progress on Navigate on AutoPilot to pursue Autosteer on City Streets, which is a long way from ever being L3. Accordingly, it also seems reasonable that GM and Mercedes will achieve their goals of L3 autonomous driving systems on highways before Tesla.
 
Yes, but you’re hypothesizing about how GM’s, MB’s and Tesla’s systems are designed as well as what will happen in the future so people can say you’re wrong with just as much accuracy as you can say you’re right.
Anybody but Tesla.

Coming soon! Per the supercruise web site, Cadillac will have an upgrade to automatically perform a lane change lane change. Groundbreaking technology - someday.

Also per the supercruise web site, the system sets its speed to the current speed when engaged. So no adapting to changes in speed limits. The system cannot be used where lane markings are not clear, cannot be used in tunnels, cannot be used on exit lanes, cannot be used on any road other than a limited set of divided highways and does not avoid any obstacles or safety hazards.

But, you can take your hands off the wheel.
 
I think one can reasonably derive from statements from each company and knowledge of the hardware/autonomous systems being used by each manufacturer that GM SuperCruise and Mercedes DrivePilot are being conservatively developed to achieve L3 highway driving. Compare this with Tesla that seems to have abandoned progress on Navigate on AutoPilot to pursue Autosteer on City Streets, which is a long way from ever being L3. Accordingly, it also seems reasonable that GM and Mercedes will achieve their goals of L3 autonomous driving systems on highways before Tesla.
yeah, like I said. You've got an awful lot of conjecture there.

I looked at MB's Level 3 system, it is incredibly limited and essentially does what AP has been capable of for quite some time. It's true that a key difference is MB applying for it to be true level 3 so I don't want to ignore that, but in terms of technological capabilities it's nothing huge.

I'd also point out that much of what Tesla has been working on with FSD for city streets can be expanded to highway driving which is considerably simpler. We'll see what the next release brings; at some point Tesla will be merging the two stacks so AP will realize some of the efforts put into FSD.