Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
The next big milestone for FSD is 11. It is a significant upgrade and fundamental changes to several parts of the FSD stack including totally new way to train the perception NN.

From AI day and Lex Fridman interview we have a good sense of what might be included.

- Object permanence both temporal and spatial
- Moving from “bag of points” to objects in NN
- Creating a 3D vector representation of the environment all in NN
- Planner optimization using NN / Monte Carlo Tree Search (MCTS)
- Change from processed images to “photon count” / raw image
- Change from single image perception to surround video
- Merging of city, highway and parking lot stacks a.k.a. Single Stack

Lex Fridman Interview of Elon. Starting with FSD related topics.


Here is a detailed explanation of Beta 11 in "layman's language" by James Douma, interview done after Lex Podcast.


Here is the AI Day explanation by in 4 parts.


screenshot-teslamotorsclub.com-2022.01.26-21_30_17.png


Here is a useful blog post asking a few questions to Tesla about AI day. The useful part comes in comparison of Tesla's methods with Waymo and others (detailed papers linked).

 
Last edited:
Yes those are too. While I was replying to something that spoke of friction braking, in my post I was speaking to the dynamics of course, not the use of friction brakes (which I did not mention!), which are just a symptom of the problem. Using friction brakes is fine as long as it is planned.
You seemed to be specifically focusing on the braking, deceleration profile. I don’t dispute that it’s something that can/should be improved, but the ‘fundamental’ part is safely stopping, which it does. Managing a right turn at a yield sign or a left turn into traffic is much more fundamental than having a ‘perfect’ deceleration profile.
 
You seemed to be specifically focusing on the braking, deceleration profile. I don’t dispute that it’s something that can/should be improved, but the ‘fundamental’ part is safely stopping, which it does. Managing a right turn at a yield sign or a left turn into traffic is much more fundamental than having a ‘perfect’ deceleration profile.
I’d say everything, including deceleration, is important. Even though I’m half joking about having to shell out $$$ for brake replacement prematurely (my 2016 AP1 is 120k miles on first set of brakes) I’m not just worried about safely stopping, I’m worried about if the vehicle stops too abruptly, the vehicles behind me may be distracted and using my speed to determine the flow of traffic instead of looking ahead of me, then be unable to react fast enough to my abrupt slowing down and cause a rear end collision. From the very beginning Elon bragged about brakes lasting virtually forever, and I have to imagine many of us drive one pedal, so the data has to be there, but like you said, yielding and left turns into traffic are important to get right, I just think not only stopping to be safe for me, but other vehicles behind me should be in that list. I feel they may have underestimated how much they needed to get right for self driving to work flawlessly.
 
$15k FSD should come with first 2 brake and rotor changes included.
What's weird is that in my use today it seemed to hardly ever use the physical brakes. Riding the line between regen and physical brakes. Today was very warm, obviously in the winter with a cold battery the same deceleration profile would result in a lot more physical brake usage.

I'm not expecting Tesla to program FSD to act differently based on available regen.
 
I’d say everything, including deceleration, is important. Even though I’m half joking about having to shell out $$$ for brake replacement prematurely (my 2016 AP1 is 120k miles on first set of brakes) I’m not just worried about safely stopping, I’m worried about if the vehicle stops too abruptly, the vehicles behind me may be distracted and using my speed to determine the flow of traffic instead of looking ahead of me, then be unable to react fast enough to my abrupt slowing down and cause a rear end collision. From the very beginning Elon bragged about brakes lasting virtually forever, and I have to imagine many of us drive one pedal, so the data has to be there, but like you said, yielding and left turns into traffic are important to get right, I just think not only stopping to be safe for me, but other vehicles behind me should be in that list. I feel they may have underestimated how much they needed to get right for self driving to work flawlessly.
I have felt that they could engineer a combined braking action to start with, whenever using the brake pedal. Most of the work could be done by active regen / reverse-EMF on the motor, which can be very powerful. Only the extremes of emergency-level breaking really need to resort to the brake pads and rotors.

By considering it this way, that all braking activity starts with the motor reversal, it would have removed that whole engineering layer from the FSD programming tasks. They would simply have API routines to apply brakes at whatever strength, and the already-engineered brake pedal control system would take it from there. As a major side benefit, the brake wear would be minimized even for any manual driving from owners who never really get the hang of EV one-pedal driving, and/or for those who want to set Regen to low or zero.

Perhaps there's some kind of long-established NHTSA regulation, based on historical fail-safe considerations of mechanical backup for hydraulic and power brakes, that has to be considered. But I'm confident that any such issue could be overcome with a highly reliable redundancy element in the braking control, and the red-tape discussion could appeal to the obvious energy-efficiency benefits.
 
  • Informative
Reactions: pilotSteve
...obviously in the winter with a cold battery the same deceleration profile would result in a lot more physical brake usage.
One assumption behind this observation, is that motor braking must necessarily throw the regenerative energy back into the battery. That's obviously a major element of high efficiency EV design, but it's not really necessary from top-level considerations.

It would be a simple and relatively inexpensive matter to direct the generated current into a simple wire-wound load cell (or whatever). The resulting heat could be dissipated to the outside, or optionally used to contribute to heat pump efficiency in cold conditions. I've often wondered why Tesla chose simply to cancel Regen with a full battery and/or in cold conditions. It seems inadvisable to change the way the braking works just for temporary and unrelated reasons.
I'm not expecting Tesla to program FSD to act differently based on available regen.
As I mentioned in my prior post above, I think the entire conception of the braking system should integrate as a standalone motor first, pads and rotors second, unified system. That has benefits for any motive driving, and again it removes that layer of engineering from the FSD programming - those guys would simply interact with a unified brake software interface, and the details of Regen versus mechanical brakes would be taken care of within that module, separate from FSD/Autopilot.
 
Last edited:
layer of engineering from the FSD programming - those guys would simply interact with a unified brake software interface, and the details of Regen versus mechanical brakes would be taken care of within that module,
I would think it works this way already. I certainly hope so. Presumably FSD just calls for whatever stopping power.

Don’t think there is any evidence to the contrary. (And friction braking when regen limited implies this too.)
 
  • Like
Reactions: EVNow
It would be a simple and relatively inexpensive matter to direct the generated current into a simple wire-wound load cell (or whatever). The resulting heat could be dissipated to the outside, or optionally used to contribute to heat pump efficiency in cold conditions.

People keep saying that, but my guess is that they haven't actually looked at what it would take. Simple and inexpensive? Are you sure about that? How much does a 60kW heater cost? How much does it weigh? How big is it? A quick look at a standalone heater looks like a 60kW heater costs close to $5k, is very large, and weighs over 200 lbs. (Obviously Tesla could probably integrate something cheaper and smaller, but really it would likely add thousands of dollars to the cost of the vehicle, result in a reduction of range, and additional expensive parts to fail.)

And no, I don't think you are going to be feeding 60kW of heat into the heat pump system.

again it removes that layer of engineering from the FSD programming - those guys would simply interact with a unified brake software interface, and the details of Regen versus mechanical brakes would be taken care of within that module, separate from FSD/Autopilot.
I'm sure that is likely how it already works. But if people want FSD to drive like they do, which people have clearly stated is their desire, where they change their driving habits based on regen availability, FSD would have to take that into account and change its behavior as well. (And again I don't think that is ever going to happen.)
 
People keep saying that, but my guess is that they haven't actually looked at what it would take. Simple and inexpensive? Are you sure about that? How much does a 60kW heater cost? How much does it weigh? How big is it? A quick look at a standalone heater looks like a 60kW heater...
This is a good discussion. I certainly don't have a design blueprint for you to review , but one of the most simplifying thoughts towards answering your challenge (a good one) is to consider the heat-shedding issue of standard car and truck brakes. In those, essentially all the stopping power produces waste heat at the brake components (usually pads, discs and rotors these days)

Despite the very significant primary design constraints of creating reliable and long-lasting brlakes for very low cost, it turns out that the heat load can be dissipated without exotic materials or complex cooling techniques. Moderate performance vehicles may typically use ventilated rotors on at least the front axle; racing and other higher performance applications have upsized parts with fancy drilled rotors with enhanced alloys and specific pads. But the point is that there is nothing like thousands or even hundreds of dollars involved.

We're not talking about a continuous application of braking power, but a much lower duty cycle with high-dissipation transients separated by much longer cool-off intervals. And in the EV case, a significant mitigation is the very application of Regen that we're discussing, used whenever possible - i.e. unlike a conventional vehicle, we don't usually have to dispose of nearly as much heat energy because of the very fact that we put it back into the battery if we can.

So, I take your point about a 60 KW heater, but the application is extremely different because the task is much simpler and the duty cycle is far lower. And by the way, we're not talking about a controlled, ducted and distributed commercial building heating plant, we're talking about something ancillary to the kinds of heatsinking that already exists in the car. Acually not even that sophisticated, because a transient load cell resistor doesn't need to be kept as cool is the motors, battery sales and power electronics that already exist.
And no, I don't think you are going to be feeding 60kW of heat into the heat pump system.
You don't have to feed transient maximum stopping power heat into the heat pump. I was merely suggesting that the waste heat from this load cell could contribute to available heat through the kinds of heat exchanging and piping that already exists in the car. Recall for example that Tesla is already cooling and collecting heat from the motors like this, and they even wrote special control software to run the motors at very low efficiency when it's helpful to have them generate more heat. By the way, I'd assume that that technique would also be programmed in to generate deliberately inefficient reverse EMF for braking, when the batteries are too cold or too full to take full regen charge - and this already-engineered trick would further reduce the energy absorption / dissipation requirements on the proposed load cell.

Finally, one might persist and ask, well then what if we have an edge case that becomes too much for the cheapo energy dissipation element that I'm suggesting? I think the answer to that is pretty simple: we still have the friction brakes. The suggestion is certainly not to eliminate those*, but to have a low-cost feature in the car that minimizes the use of friction brakes in more circumstances, contributes to a more consistent user experience and (possibly) allows more otherwise wasted energy to help warm things up just when that's needed.

* I don't think they'll ever be eliminated, but I do think that ongoing refinements of the motor-braking system could push the conventional brakes further into a backup role. This may already be happening, I'm not sure - but there's been a minor scandal reported by Performance model owners, that some newer P cars are being shipped with red caliper covers to cosmetically dress up ordinary brake components, rather than the originally-specified real high-performance calipers. I'm not saying that I know this has anything at all to do with an increased role of motor braking to justify such a change, but it's an interesting thought.​
Just as we now realize that a deep exhaust note from a high-flow muffler is no longer the giveaway signature sound of a super-quick car, we might find that the conventional wisdom of high-performance braking will no longer come down to expensive lovely calipers and fancy cross-drilled rotors. it'll have more to do with highly engineered deceleration capabilities of the electric driveline.​
 
  • Helpful
  • Informative
Reactions: jebinc and navguy12
Ask 5 different people what an "HD map" is and you'll get 5 different answers.

We already know Tesla uses map data for driving, including things like speed limits, stop sign locations, number and type of lanes, etc.

We also know they update this data in real time each time you enter a destination into the nav.

But they're not the cm or mm accurate things many LIDAR fans insist are "HD maps"
this is not true. There are not 5 different ANSWERS.

Is 11.4.1 using some form of HD maps? You be the judge. The visualization shows the road blocked here before we can actually see it?

Yup and its even more evident in this video:

1:14 is good example (the curvy side road on the right). 3:33 as well, also on the right. At 4:46 it's showing the two side roads up ahead (past the planned left turn) and it matches the map data perfectly!


Extra metadata <> HD map.

HD map means higher resolution i.e. say if standard map is accurate to one foot, HD would be accurate to one inch or one mm … and they might also be 3D maps like the LiDAR ones.

Standard maps always had extra metadata that we can’t always see. You can see tons of extra fields in Open Source map for eg … most of them not populated. Even long time back GPS units would show more details (like lanes) if you bought a higher end version.
These are not just metadata. These are accurate lane-level map that is detailed above with 1:1 match.

They even now have full exact parking layout and slots in the map even when its impossible for car to see it.

Very detailed HD map is clearly being fed to the BEV/ Lane Network

Basically what you are arguing is that no matter how much detailed the maps are, its not an HD map because it would expose another tenet of the FSD narrative.
 
Last edited:
  • Disagree
Reactions: EVNow
Incorrect.
Please provide proof of all the accidents caused by FSD’s ‘unsafe’ braking profile. Otherwise you’re stating an opinion that is contradicted by facts.
I have felt that they could engineer a combined braking action to start with, whenever using the brake pedal. Most of the work could be done by active regen / reverse-EMF on the motor, which can be very powerful. Only the extremes of emergency-level breaking really need to resort to the brake pads and rotors.
For the last couple of versions Tesla has already done this by adding the option to use the mechanical brakes automatically when regenerative braking is unavailable. FSD essentially does the same thing.
It would be a simple and relatively inexpensive matter to direct the generated current into a simple wire-wound load cell (or whatever). The resulting heat could be dissipated to the outside, or optionally used to contribute to heat pump efficiency in cold conditions. I've often wondered why Tesla chose simply to cancel Regen with a full battery and/or in cold conditions. It seems inadvisable to change the way the braking works just for temporary and unrelated reasons.
I’ve read about some other hybrid vehicles that essentially use large capacitors. As someone who lives in a climate where regenerative braking is all but unavailable for 3-4 months this would increase efficiency significantly.
 
  • Disagree
Reactions: AlanSubie4Life
Please provide proof of all the accidents caused by FSD’s ‘unsafe’ braking profile. Otherwise you’re stating an opinion that is contradicted by facts.

For the last couple of versions Tesla has already done this by adding the option to use the mechanical brakes automatically when regenerative braking is unavailable. FSD essentially does the same thing.

I’ve read about some other hybrid vehicles that essentially use large capacitors. As someone who lives in a climate where regenerative braking is all but unavailable for 3-4 months this would increase efficiency significantly.
The capacitors would be prohibitively large.
 
I’d say everything, including deceleration, is important. Even though I’m half joking about having to shell out $$$ for brake replacement prematurely (my 2016 AP1 is 120k miles on first set of brakes) I’m not just worried about safely stopping, I’m worried about if the vehicle stops too abruptly, the vehicles behind me may be distracted and using my speed to determine the flow of traffic instead of looking ahead of me, then be unable to react fast enough to my abrupt slowing down and cause a rear end collision. From the very beginning Elon bragged about brakes lasting virtually forever, and I have to imagine many of us drive one pedal, so the data has to be there, but like you said, yielding and left turns into traffic are important to get right, I just think not only stopping to be safe for me, but other vehicles behind me should be in that list. I feel they may have underestimated how much they needed to get right for self driving to work flawlessly.
Technically someone rear ending you at an intersection is 100% their fault but I get your point - suddenly slamming on the brakes unexpectedly is not safe behavior. It’s the same point I made in other threads regarding phantom braking. The current deceleration profile (11.3.6) is significantly improved from earlier versions, though, and you’d be hard pressed to call it unsafe.

I suppose one could say that the way in which the car comes to a complete stop at a stop sign is unsafe since it’s actually unexpected behavior compared to how virtually everyone drives, even if it’s following the law. We’d have to bring that up with the NHTSA, though.
 
this is not true. There are not 5 different ANSWERS.

There are though.

Hell you can find more than 5 different answers to what an HD map "really" is right here in the FSD threads just in the last 6 months-- this gets debated often.

A quick search in that period found mm-accurate, cm-accurate, accurate lane-level, lidar-mapped, larger than 10kb/km, and any map with road views rather than just the roads themselves given as >5 different definitions of what an HD map "really" is.

Generally the definition seems to just be used to try and make a point for whichever side of the debate you're on... if you want whatever you're discussing to be an HD map, you find some aspect of it you think can go into a nebulous "HD maps" term and call it an HD map... if you want whatever you're discussing to NOT be an HD map you just no-true-scotsman whatever the specific thing is as not making it an HD map.
 
  • Informative
  • Like
Reactions: nvx1977 and aronth5
This is a lie.

If you had any proof - bring it on
E2165E4XMAQSIWl



We pioneered commercial high-definition (HD) mapmaking, and we have been leading the way ever since. Today, we are a market leader for HD and automated driving (AD) map tech. TomTom’s HD Map has been successfully integrated into R&D projects at 9 of the top 10 global carmakers and over 10 million vehicles on the road use our maps for AD. We were the first company to offer broad highway coverage across the U.S., Europe and East Asia – and continue to expand our AD coverage and features.

@verygreen could you provide outputs of the latest hd maps kind of how you did with the initial AP maps?
 
  • Like
Reactions: jebinc