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.
By definition, if a car knows it can't handle something, then it recognizes what it is. The problem with this is that what the car doesn't know is an edge case. It's not possible to know edge cases unless it's seen them before. That's why we've yet to see an actual L3 car. If you can solve for L3, you can solve for "L4" and "L5".
What does that even mean? L3 has nothing to do with "edge cases", but rather with risk/capability trade-offs. For example, the few L3 systems we have seen so far (like Audi's traffic jam assist) have a speed limitation, which is probably related to sensor scan rates and a risk assessment that Audi made.
 
My big question is still this. What kind of visualizations will those of us with MCU1 get to see? All this is likely dev views but if I’m limited from seeing cones how the heck will all this other data that helps me understand the car is perceiving adequately be shown? In past Green said it wasn’t hardware limitations. I hope they open it up. Otherwise time to shell out 2.5k.
 
@verygreen are you implying that the FSD visualizations that we see in the current release would "embarrass" Tesla because it somehow resembles the point clouds a radar manufacturer generates?
upload_2020-10-22_16-34-39.png


If not, please explain the "loss of face" that Tesla is supposedly trying to suppress data.


Just so there is no confusion, here is the FSD "Neural Net" output.
upload_2020-10-22_16-34-14.png
 
Its one better forward looking radar unit and has absolutely nothing to do with 360 degree LIDAR cars...
Its the most normal thing in the world that Tesla continually improves their current sensors like everything else they improve in their cars all the time. The cars with the new sensors will have even more reliability and safety, shocking I know.

The sensor discussions are so boring, the neural nets is where the magic will happen ;)
 
@verygreen are you implying that the FSD visualizations that we see in the current release would "embarrass" Tesla because it somehow resembles the point clouds a radar manufacturer generates?
View attachment 601404

Not what he is saying.

I think he's implying that Tesla understands they need better hardware, which is coincidentally extremely similar to lidar, and gives the same output as lidar would, for FSD but so they don't lose face insist it can easily be done with the existing sensor & camera suite.
 
I'm not Green, and I don't know if it's true or not, but I believe what he means is that Tesla secretly wants to use lidar but can't because Musk painted them in a corner with his anti-lidar crusade, so they are now looking at high-res radar as an alternative.
Here is an output from Lidar...
4BFB8851-4BE2-40AE-B77E-FDD1391D6884.jpeg

How does the point cloud from greens post "resemble" the output from lidar (as shown above)?

He is a sharp dude, I know he would not confuse output form Lidar with Radar point clouds... (unless he has an ulterior motive... )
 
Last edited:
Just for clarification, can Tesla achieve L3 with this set of features by having the system programmatically determine when it needs human intervention and alerting the driver in enough time for them to resume control?

If it's able to determine that far enough in advance to give the driver plenty of warning? Sure.

The point of L3 is the driver does not have to actively paying attention. Just ready to take over with some notice.

So you can be reading a book or watching a movie instead of watching the road- get sufficient notice the car will need you soon- and have enough time to refocus from whatever ELSE you are doing, back to what is going on on the road, and then take over from the computer.

If all the computer can do is go TAKE OVER RIGHT THE HECK NOW that's not L3.



And then can they achieve L4 by having the car doing the same task as above, but also being able to find a safe place to pull over and putting on the hazards (same fallback performance of dynamic driving task as Waymo), correct?

Yeah the main difference between L3 and L4 is L3 can not fail safely. If there's no human to take over after a short warning period, the car can not safely drive and bad things may happen.

L4 never requires a human to even be present, let alone ready to take over, in order to be safe- if it finds it's no longer able to continue driving it's able to, entirely on its own, safely park the vehicle somewhere.


If Green is correct, the radar that he mentions is actually a high-resolution radar, which is very different from what we currently have in the automotive sector, and has been discussed as a potential alternative to lidar:

Here Comes High-Res Car Radar

Question then becomes- do you NEED that to get to at least L4?

If so- do they offer free retrofits and/or refunds? For everyone? For just the pre-3/19 buyers? For nobody and they defend lawsuits?

Or do they say the old radar is good enough for L4, the new radar is just "better"?
 
  • Helpful
Reactions: willow_hiller
Just for clarification, can Tesla achieve L3 with this set of features by having the system programmatically determine when it needs human intervention and alerting the driver in enough time for them to resume control?
It can be a lot more casual then what you are suggesting. Tesla can decide to turn on L3 for below 15 mph in stop and go traffic. Once speed goes above 15 mph then Tesla starts nagging. Speed that Tesla is comfortable at for L3, can go up over time. Also L3 can be limited to starts that have favorable laws.
And then can they achieve L4 by having the car doing the same task as above, but also being able to find a safe place to pull over and putting on the hazards (same fallback performance of dynamic driving task as Waymo), correct?
Good point. For Tesla that can pull over, there isn't much of a difference between L3 and L4. I guess Tesla can call their traffic jam assist L4 since I suspect they will have the ability to pull over.
 
Hopefully reverse summon will come soon:

View attachment 601428

I just want the car to be able to park itself with me inside, for starters. Not much point in navigating the parking lot if you can’t park in the furthest spot from the store between two lines with no adjacent vehicles (everyone knows you never park adjacent to other vehicles :eek:).
 
Lets summarize...
In this tweet, @verygreen is saying/implying that Tesla would be embarrassed and thus has to "save face" by hiding info of how this particular radar sees the world... because it resembles how lidar sees the world..
That's not the way I interpret it. He's just saying that Tesla is working on high resolution RADAR. I guess he might be implying that they're choosing high resolution RADAR over LIDAR because of Musk's past statements. There could be many other reason too.
What is a little bit odd about Elon's anti-LIDAR position is that LIDAR is basically just better RADAR. I was assuming that meant that they would be getting rid of RADAR (like Mobileye has done for its ADAS). The low resolution RADAR Teslas use has probably become pretty worthless as the camera vision system has improved. I wonder if they're using it for FSD at all?