Do you mean it's not working with AP1 or that it is not officially a AP1 feature?Blind spot detection is NOT an AP1 feature so I am puzzled by your puzzlement
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
This video makes it pretty clear why autopilot is limited to 35mph. There is something unstable with that lane tracking. I haven't seen that behavior in AP1.
it will be interesting to track speed of learning. NVIDIA claimed that their system "performance continually improved" ( How Our Deep Learning Tech Taught a Car to Drive | NVIDIA Blog ), given that Tesla took similar approach though with their own code and with already setup 'fleet learning', we should observe improvements, but how fast - is really a question. I look forward for more videos. One thing is to know that there will be improvements, another to view it by own eyes
joebiker25
Member
Exactly! Hence my assumption that AP 2 blind spot detection will utilize the 2 side cameras as some point!
ZeroDarkSilver
HEPA cleaning your atmosphere, you're welcome
Just speculating here, but I wonder if the 4 vs 8 camera distinction between EAP and FSD will go away at some point. Meaning they will announce at some point that EAP is using 8 cameras too.
The idea here is, why not use all 8 cameras (even when EAP is not engaged) for "safety" functionality like avoiding collisions? More data from more cameras should yield a safer result regardless of which mode you're in. And the real difference between EAP and FSD would end up being not how many cameras it is using, but software lockouts/differences like where EAP is allowed to operate, and how integrated it is with the Navigation system to initiate lane changes, exits, turns, etc.
Maybe the 4 vs 8 camera thing is just a marketing trick to help justify the price difference in EAP vs FSD in people's minds.
Last edited:
but for different purposes. You have to understand that autopilot is not a one complete unified system. It consists of two completely seperate systems. Adaptive cruise control and lane keeping assist. Then you also have the automatic lane change system that currently uses sonar.
These are three completely different system in action.
The two rearward cameras will only be used for changing lanes. aka for the automatic lane change system.
similar to AP1 the main forward camera will be the only camera used for driving and the narrow forward camera will somehow assist it.
Again these are separate systems. You also have to take into account how their Neural nets are trained.
In theory you could cover with paper both the rearview camera today and AP2 will still function because it doesn't use it for its driving policy.
This won't happen partly because of the reasons i listed above.
Autopilot is not a unified system. Its multiple systems developed independently per-say that makes up autopilot.
Tesla won't use the 8 cameras because EAP is simply AP1 with more fancy features.
Alot of people refuse to get this and Tesla make it confusing aswell. But Its still level 2 autonomy.
Autosteer is seperate from TACC and Auto-Lane Change is seperate from Autosteer and TACC.
ZeroDarkSilver
HEPA cleaning your atmosphere, you're welcome
Notice how in my post I say "just speculating here" and "I wonder if...". And in your post you say "you have to understand...".
Unless you are Sterling Anderson, work on his team, or have stolen Tesla's IP, you don't know how all this actually works. Not sure why you are so insistent that you're 100% on all these details that we can't really know.
I think the mistake you may be making here is that you think the new cars still have the AP1 MobilEye-based hardware in there, running the AP1 software, and that AP2 is just added in as a separate system. The public evidence makes it pretty clear that the AP1 hardware is gone, and AP2 is new software running on new hardware. So, they are having to rebuild the basic AP functionality back up on a new platform. This is why it's taking so long and AP2/EAP doesn't already just work as well as AP1.
Most likely, although the drive to keep Model 3 costs down may mean it does not have the same equipment as the S/X but some lesser level. Really tough to know until Tesla puts more out with respect to what equipment the 3 will actually have. Think of it like car radios back in the 1990s. Most new cars came with an AM/FM/CD player car radio, but the premium cars from a manufacturer tended to have a better radio while the low end cars/trim packages got a cheaper radio. That's why so many audiophiles would put in after market radios...
It is entirely possible that Tesla may take a similar approach to keep 3 costs down near the price point Elon has said he wants to meet. On the other hand, the argument could be made that the parts will all be the same to the maximum extent possible to avoid the costs of different parts sets and enable even further economy of scale in parts purchases.
My gut feel is that Tesla is going to try to keep the parts the same to the greatest extent they can but may get forced to "go cheaper" with some things as a cost control measure.
One example where we know of where they are NOT going with the same parts is battery packs where the 3's will get the new 2170 batteries while the S and X have 18650s and are not expected to change for at least a couple of years. In this case, the argument could be made that the 2170s are actually BETTER than the 18650s, but the driving reason Tesla is going to them is because of cell cost -- the 2170's are supposed to be MUCH cheaper coming from the Gigafactory than the 18650s are today. The expected higher energy density is simply a benefit of the newer design/advances in battery technology, but it is the cheaper battery cell cost that Tesla is counting on to help them hit the Model 3 cost targets.
It is entirely possible that Tesla may take a similar approach to keep 3 costs down near the price point Elon has said he wants to meet. On the other hand, the argument could be made that the parts will all be the same to the maximum extent possible to avoid the costs of different parts sets and enable even further economy of scale in parts purchases.
My gut feel is that Tesla is going to try to keep the parts the same to the greatest extent they can but may get forced to "go cheaper" with some things as a cost control measure.
One example where we know of where they are NOT going with the same parts is battery packs where the 3's will get the new 2170 batteries while the S and X have 18650s and are not expected to change for at least a couple of years. In this case, the argument could be made that the 2170s are actually BETTER than the 18650s, but the driving reason Tesla is going to them is because of cell cost -- the 2170's are supposed to be MUCH cheaper coming from the Gigafactory than the 18650s are today. The expected higher energy density is simply a benefit of the newer design/advances in battery technology, but it is the cheaper battery cell cost that Tesla is counting on to help them hit the Model 3 cost targets.
Precisely ! Thanks for confirming that AP1 doesn't do this.
Here is a video of that odd subtle lane-dance at a complete stop. I don't have the equipment to record while on the move ( yet)
I wouldn't jump to that conclusion. You are comparing the visual representation of two different systems. The jitter here comes from Tesla Vision and may be no worse or even better than uncertainty in AP1. I would bet any small uncertainty is just filtered out in AP1 until it becomes so great that the line is lost. I'd personally prefer to see indication of uncertainty and how much.
The jitter is also very similar in to the line jitter of the raw Tesla Vision line estimations shown in Tesla's self driving demo videos!
Update on TACC.
It's pretty poor with lanes mergers .Had TACC engaged merging from one major freeway to another. Since it was following cars in traffic, it seemed ok to engage it. However on a merge lane it seems to complete disregard the car in front. Perhaps because it was too far wide to the left but nonetheless in the same merge lane. Kept accelerating and I had to apply the brakes.
It's pretty poor with lanes mergers .Had TACC engaged merging from one major freeway to another. Since it was following cars in traffic, it seemed ok to engage it. However on a merge lane it seems to complete disregard the car in front. Perhaps because it was too far wide to the left but nonetheless in the same merge lane. Kept accelerating and I had to apply the brakes.
If I were designing this, I would filter out the jitter. Personally, I would prefer to be able to see it, but I understand why it's alarming to some. I think preferably I'd like to see indication of uncertainty in the form of wider lane lines (stabilized over a short duration).
Started pretty patient, but am getting annoyed at the lack of information and communication.
Now starting to feel like a lot of others with HW2, that it's simply unbelievable that Tesla didn't have the AP1 functionality even close to parity on the new hardware before release on NYE. I think we were all lead to believe the software was done but going through validation. It seems it's not even close to being done, and there's a long road ahead to even full parity let alone EAP.
Even some small updates would go a huge way for the new owners waiting for the software update.
Now starting to feel like a lot of others with HW2, that it's simply unbelievable that Tesla didn't have the AP1 functionality even close to parity on the new hardware before release on NYE. I think we were all lead to believe the software was done but going through validation. It seems it's not even close to being done, and there's a long road ahead to even full parity let alone EAP.
Even some small updates would go a huge way for the new owners waiting for the software update.
Last edited:
How do you explain the jitter at a complete stop ? My guess is something to do with the 3 cameras and trying to make sense of the depth.
It's continuously redetermining where the line is and coming up with different estimations. It's expected as long as it always quickly averages out close enough.
