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

FSD rewrite will go out on Oct 20 to limited beta

This site may earn commission on affiliate links.
I'm sure they are under an NDA. it might show on firmware tracker, but only if someone that uses one gets the update. Something will leak sooner than later

I suspect that they've already contacted the individuals who are receiving this update (employees, high profile Tesla owners, etc.). I can't imagine some Early Access owners just discover a new update today that includes significant FSD changes. I think the initial testers will be briefed on the update - how to use it, limitations, etc. You can't capture all that detail in release notes.

For one thing, if this update does what Elon is hinting it will (turns at intersections, navigating unmarked streets, etc.), the AP interface and controls will need to change completely. Will you even still double tap the AP stalk to engage? Will there be a concept of engaging TACC but not AP? Do your hands need to remain on the wheel (makes turns difficult) or will they use the cabin camera to ensure driver attentiveness? What actions will require driver approval (similar to 'go on green').

I think they train the early testers, let them leak some videos, and then release a LOT more specific info on capabilities and how to use before they roll it to the general population.
 
I think the rewrite will bring enhanced L2. Elon has suggested the Dojo will be needed to tune it, and that is a year away. This will happen about the time HW4 arrives. I believe at this time they will also enhance the radar. The number of points an automotive radar can see, has been dramatically enhanced in the last year. From coarse cloud, to vision radar. Tesla has already upgraded once, from Bosch to Continental. TI is one of the companies with vision radar development kits out to automotive customers. It would be foolish for Tesla not to take advantage of this. More detail at the same cost.
So if you have HW4 and new radar, it might also make sense to look at the cameras.

FMCW radar can apparently sense stationary objects. So a change in radar would be one way to eliminate phatom braking. I don't know how Tesla releases HW4 without most HW3 owners joining a class action lawsuit.

But I doubt Mobileye has a phantom braking problem, so perhaps there is a camera solution to that problem.
 
I don't know how Tesla releases HW4 without most HW3 owners joining a class action lawsuit.

Easy:

Tesla said:
The currently enabled features require active driver supervision and do not make the vehicle autonomous. The activation and use of these features are dependent on achieving reliability far in excess of human drivers as demonstrated by billions of miles of experience, as well as regulatory approval, which may take longer in some jurisdictions. As these self-driving features evolve, your car will be continuously upgraded through over-the-air software updates.
 
Based on the incredibly, incredibly slow speed of actual, usable Autopilot enhancements and how functionally AP at the end of 2020 is 99% the same as the AP I had in the beginning of 2017, I will go out on a limb and say that Tesla fans are overvaluing the whole "We got a million cars on the road collecting data!! Nobody can even reach us!!!" argument; especially considering how other brands have pretty similar and well performing driving assistance systems that don't utilize any such data at all.
 
Based on the incredibly, incredibly slow speed of actual, usable Autopilot enhancements and how functionally AP at the end of 2020 is 99% the same as the AP I had in the beginning of 2017, I will go out on a limb and say that Tesla fans are overvaluing the whole "We got a million cars on the road collecting data!! Nobody can even reach us!!!" argument; especially considering how other brands have pretty similar and well performing driving assistance systems that don't utilize any such data at all.

If you haven't yet, I would highly recommend watching Karpathy's presentation from February. Much of it is outdated, and judging by the recent exchange posted above from Green all of the networks discussed were pre-rewrite, but it really highlights how Tesla is able to leverage their data collection in ways other automakers aren't.


The example he gives is stop signs, and from the fleet he's able to access thousands of sample images; of occluded stop signs, stop signs in varying different contexts, etc. Tesla is only getting started on the work that will be able to leverage the cache of data, but it will allow Tesla to progress very quickly with the areas of autonomy that even Waymo struggles with now.

From an interview with The Verge in December describing the role of Waymo remote operators:

These folks don't joy-stick the car or anything like that. But they can help answer specific questions that a car may have about an ambiguous situation.

I think Tesla won't need human remote operators for ambiguous situations that can be answered with context clues. They'll be able to source enough examples containing context for the network to learn that behavior itself. And when Dojo is utilized and those fleet-sourced training images transform into fleet-sourced training videos, the contextual learning will be even faster.
 
Last edited:
  • Like
Reactions: mikes_fsd
I suspect that they've already contacted the individuals who are receiving this update (employees, high profile Tesla owners, etc.). I can't imagine some Early Access owners just discover a new update today that includes significant FSD changes. I think the initial testers will be briefed on the update - how to use it, limitations, etc. You can't capture all that detail in release notes.

I think they train the early testers, let them leak some videos, and then release a LOT more specific info on capabilities and how to use before they roll it to the general population.

Yes agreed on all of this..

For one thing, if this update does what Elon is hinting it will (turns at intersections, navigating unmarked streets, etc.), the AP interface and controls will need to change completely. Will you even still double tap the AP stalk to engage? Will there be a concept of engaging TACC but not AP? Do your hands need to remain on the wheel (makes turns difficult) or will they use the cabin camera to ensure driver attentiveness? What actions will require driver approval (similar to 'go on green').

uhh... no.... why would any of the things you mentioned here change?

"the AP interface and controls will need to change completely." --> disagree

"Will you even still double tap the AP stalk to engage? Will there be a concept of engaging TACC but not AP? Do your hands need to remain on the wheel (makes turns difficult) or will they use the cabin camera to ensure driver attentiveness?" --> don't think any of these things need to change


"What actions will require driver approval (similar to 'go on green')." Go on green may or may not change, auto lane change in city may or may not require confirmation, and confirmation before turn is likely..... but none of this requires a complete change in AP interface and controls.

"Do your hands need to remain on the wheel (makes turns difficult)". --> this is not true... and I don't know why people believe this and keep suggesting this.... First, of all, it is not vital that you keep your hands on the wheel the entire turn... its important you you give AP some indication that you are paying attention at the start of the turn (such as wheel torque or other confirmation).... 2nd of all... why do you think it is hard to keep your hands on wheel during an autonomous 90 degree turn??
 
Based on the incredibly, incredibly slow speed of actual, usable Autopilot enhancements and how functionally AP at the end of 2020 is 99% the same as the AP I had in the beginning of 2017, I will go out on a limb and say that Tesla fans are overvaluing the whole "We got a million cars on the road collecting data!! Nobody can even reach us!!!" argument; especially considering how other brands have pretty similar and well performing driving assistance systems that don't utilize any such data at all.


Curr: Red SP100D(L) - MCU2 - AP2.5 - FSD.
Prev.: Red SP90D(L) - AP1. Be careful when buying a SP90DL.
Prev.: Red S85D - AP1
It seems that you've forgotten what your 2016(?) Red SP90D(L) - AP1 was actually capable of.
As someone who still owns an AP1 vehicle, let me tell you there is night and day functionality difference that has been introduced on FSD cars vs AP1.
The AP1 car will plow right through traffic lights, stop signs.
The AP1 car has a hard time detecting pedestrians as well (vs the FSD one)
The AP1 car still works decently well for highway lane keeping. But will not pass others automatically will not take exits, etc.
 
  • Informative
Reactions: loquitur
Big news! FSD going out on Oct 20 to limited beta.

ddnU7xI.png

If we don't get in the limited beta, I am sure there will be lots of leaked videos of the FSD.

If we don't get it, we'll know that we are not expert and safe drivers which I've long suspected
 
Highway FSD is a trivial task nowadays. Google and other companies are doing great in limited set of cities, but Im curious of their roll out plan to some city in XYZ country and how much effort it is. I'm not even talking about different conditions (snow/fog), different road rules, signs all over the world... Tesla has the data but can they leverage it properly? Will see soon)
I think this upcoming rollout will tell me if Tesla is capable of delivering the feature (within a year) because today's AP is not usable in the city at all and has incredible amount of low level issues and limitations that are NOT minor bug fixes and more like flows in their pre FSD release approach / software.