RobStark
Well-Known Member
Most of the 18650 lines were mothballed 4 years ago when Tesla came calling.
Most smart phone and tablet makers have moved on.
And Samsung/LG were growing their 18650 share down market.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
- Status
Most of the 18650 lines were mothballed 4 years ago when Tesla came calling.
Most smart phone and tablet makers have moved on.
And Samsung/LG were growing their 18650 share down market.
andrewt3000
Member
One possible explanation is they just want to test that the system works well without mobile eye processing in the mix.
But I agree. The delay of AP features till December is still strange to me... why can't it be done sooner? Especially if they have 200 million miles of training data and mobile eye only has 250k miles of data. Plus they have been collecting the data since September 2014 (that's 2 years). Elon said on the call they have been working on it for a year. What could they do between now and December... that they couldn't have done over the last year with millions of miles of data..... Did they just need 6 more weeks to write the software? If so... why not just wait 6 weeks and not have a step back in functionality. Is it because they are testing or do they need to collect more data? By my estimation they should be able to collect around 10 million miles of data by the end of the year on the new hardware.
At the least, the people on the conference call should have asked... "Why are we taking a step back on functionality? What is the impact of removing mobile eye from the mix? How applicable is the 200 million miles of data collected so far to the new platform?"
Regulations probably require proof of safety, given the entire sensor suite was switched out, new software logic also needed to control the new sensors. 2 months of shadow data to show it's absolutely safe, if AP was on it would take correct actions in any given circumstance. No such thing as "lost 200m miles of AP".
Disclosure: I wrote software for autonomous rover before in college sort of like AP but a billion times simpler (only 2 sonar sensors)
I think Panasonic has a number of lines completely dedicated to Tesla 18650 output.
From July gigafactory event:
- JB Straubel says Tesla developed this battery this size by starting without preconceived notions. They then optimized for efficiency, size and output. The 18650 standard was called an accident of history though it had served Tesla and others well. Tesla says it predicts that this new 21-70 battery size will become a new standard.
- The half centimeter height increase for the car packs would be offset with more efficient battery packaging which will make the packs actually the same thickness or less than current packs and obviously with a higher energy density.
- Musk noted that once the 21-70s were in mass production, they could find their way into existing Tesla car battery packs for the Model S and X as well as the Powerwall.
That seems like Tesla expects to phase out 18650s completely. Which implies that Panasonic will convert its 18650 lines to the new format. This seems pretty straight forward, the 2170s are only 3mm wider on the diameter and 5mm taller on the height.
Last edited:
This. I think given the mothballed status of the Panasonic plants that Tesla got a very good deal. This would have been probably part of GF deal #1. To me it seems very clear that Panasonic sees Tesla as a remarkable growth opportunity and is staying very close.
Ameliorate
Member
I know nothing of this organisation; however, it was nice to read a change in sentiment towards Tesla due to joining the utility service industry and offering multiple ways to generate income. Comparison to Amazon.
Why Tesla Motors Inc's Solar Roof Is A Game Changer For TSLA Stock - BNL Finance
I also watched and shared Before the Flood. Fantastic all around, definitely a nice bonus to see Elon and the factory getting some more visibility.
Why Tesla Motors Inc's Solar Roof Is A Game Changer For TSLA Stock - BNL Finance
I also watched and shared Before the Flood. Fantastic all around, definitely a nice bonus to see Elon and the factory getting some more visibility.
Here is some brief technical of how AP works. AP is a software that runs in a loop. Each sensor input ( or multiple) will interupt that loop to take appropriate actions (technically a function call) depending on what the sensors saw.
So AP 1 already created a great working framework which is the most difficult part. AP 2 will basically add some more functions based on additional sensors and how decisions are going to be made based on a combination of the sensors input. It is much easier to do, and would need a bit of time to test and refine the decision making functions.
Again, as Elon said, do not bet against AP 2. There is zero chance it would fail.
So in Q1 TE delivered 25 MWh total, in Q2 they didn't give a number and in Q3 they said TE delivered 300 MWh to date. Any estimates for Q4 totals?
Both you and Xpert know much more about this system and technology than I do so please correct or ignore as appropriate.
I think Tesla has to start over from scratch and the only really useful data from the previous AP archive is the mapping provided by gps. Based on my limited understanding of the Dave paper by Nvidia it seems the local computer within the car, the Drive PX2 has to learn clues from a much greater number of sensors--the cameras, the car's response by the driver to the images, the radar, and the ultrasound sensors. Much more data than the old system before the hardware upgrade. It's a new ball game.
The Drive PX2 can provide situational experience connected to the driver's responses. Over a very short period of time it can derive enough experience which Tesla can massage, just as did Nvidia before testing the resultant program, and let it drive a car. In the demo of the Dave paper about 100 hours of driving over 100 miles and varied terrain, including traffic, they had enough to demonstrate the concept and successfully try on new and challenging situations. Plus Tesla's new cars have much more sophisticated sensors. Nvidia had three cameras in one direction collecting information, not surround vision.
I haven't the foggiest what the new AP software by Tesla will have to do. Maybe the only thing needed is correlating the sensory learns from each car to the core of another computer and apply deep learning techniques to that data. Or, they need to review the data collected by the newer hardware vehicles in corner conditions where the AP screwed up but the driver did not. I think Tesla has confirmed that is what they are working on now--calibrating the system.
Certainly with two months of MS/MX cars producing data is much more than Nvidia needed for its demonstration so the lack of folding in data gathered before is not a huge problem imo.
There's a lot of speculation here but youse guys and perhaps others can set me straight.
1) Are they really gathering shadow data from all vehicles produced at this time forward? Tesla has said so.
2) Is it true the only useful data from previous iterations of AP is the gps input?
3) Can they use the new radar experience or do they even need to do so if the car is doing deep learning by itself as assumed above? Maybe all that is needed is catching the echoes of the radar as it does now. Processing their meaning may be learned by the Drive PX-2.
4) Do they have to control things afterward to deal with speed limits, etc., and right of way rules? Or can each car learn this too?
5) What about geographic differences? Is it necessary to intervene in the learning process or ensure that right hand drive countries follow different rules?
6) Of course there will also be limits placed by regulating bodies.
7) Et cetera, et cetera, et cetera, as a king of Siam was made to say.
Edit: Apologies to Xpert: I didn't see your most recent post. Also, I may be completely wrong about the capabilities of Drive PX-2 which you can probably correct. If so, then sorry for the noise here.
Last edited:
DragonWatch
Small FootPrint
B4 the Flood ~ well done. While I have been reducing my (our) foot print, still have a ways to go. Focus now is on Model X and to follow with a Solar Roof, this will be our second home with Solar cell capability. I put my life where my mouth is!
mmd
Active Member
You are asking way too many relevant questions. How about using the Occam's razor, that the sudden end of MobileEye-Tesla relationship has forced Tesla to redo AP w/o MobilEye in a hurry? To make it sound good, call the fresh start version 2.0 instead of version 0.5. We also don't know for sure, who has the right to the data collected using MobilEye systems.
IIRC, some questions were asked in the conference call. Elon said it is getting too technical.
racer26
Active Member
While you might be correct that the MBLY breakup accelerated the release of AP2 (to guard against being hamstrung and having the product held hostage by a vindictive MBLY), it had been under development for a while beforehand, and so it's hardly fair to characterize it as 0.5.
We do know for sure that it is Tesla, and only Tesla, that holds the rights to the data collected on AP1 equipped cars. That was publically articulated as one of the demands MBLY made that TSLA refused, was the sharing of said data. Please stop spreading misinformation.
The Occam's razor answer is simple. In order to train autopilot's neural network, and to prove it is safe before unleashing it on the driving public, you need thousands of miles of data from the hardware suite it's attached to. The earlier hardware's data is no good for this purpose. The fastest way to get that data is to get that hardware on the customer cars. The 6 to 10 weeks of customers being upset they have to wait for their autopilot to work is a small price to pay compared to being at the mercy of MBLY refusing to sell (or grossly overpricing) the hardware you need to make AP1 work for however long it takes to acquire that data on company owned test cars.
I don't think this is how it works. If there's enough new code in the system they have to be extremely careful in how and when they enable it. So they need to see how it works out in the wild first (without it activated, in shadow mode), iterate with fixes until it's working well and then enable it. I suspect the fallout with MBLY put some extra pressure but some amount of feature loss would be inevitable regardless of MBLY situation. This is a major overhaul and they simply need time to gain confidence in it.
MitchJi
Trying to learn kindness, patience & forgiveness
Guys, please start using the ignore function, instead of responding to this nonsense!
Using 220170's will alow increased capacity and lower costs. Tesla said that they'll do that eventually. Didn't sound eminent.
It's not that simple. The length used for the steel cylinders would be 65 mm vs 56, and the separator inside the cells would need a be increased proportionally as well.
One of the principal advantages of the scale of the GF is that made it feasible to develop custom large scale cell manufacturing equipment. Now that they've finished the first phase, succeeding phases will go faster and cost less. That also means that Panasonic and Tesla can build smaller factories at other locations using the new equipment that will produce cells at a similar cost to the GF. I'm not sure if it's cost effective to retrofit the old equipment at a large enough scale to replace the MS-MX cells.
He said don't bet against it being twice as safe as manual drivers using the AP 2.0 sensors.
Last edited:
mmd
Active Member
Please point to the source of the highlighted claim. Typically, these data sharing agreements are made before partnering, not after 2 years of working together. See below how Google and FCA are discussing data sharing even before their pilot project begins.
Fiat Chrysler CEO - Unclear who owns self-driving cars' data
So, I'm quite unclear why MobileEye will demand data sharing now if it wasn't in original agreement. It's also possible, there are multiple sets of data for the same miles driven.
About Tesla developing their own system for a year: May be it was only a background pilot project. If even the active safety features are off till December, what state of development was it in? These don't need millions of miles of driving data. And if they already have 200 millions miles of data, even easier to activate right away. And that's why I called it version 0.5. Hardware may be version 2.0, but whole system is at 0.5 right now. Eventually it will get to 2.0, but it is not there now.
Last edited:
Not wanting to defend the German car industry (there certainly has been some arrogance), however, part of it is that the public opinion has strongly turned in favor of EVs with media and now regularly saying EVs are the future. Before that was so, the whole topic was not as interesting as it is now. Also, naturally a game becomes more interesting when your home team is competing, right?!
I type on iphone so its kinda hard. I did not read Drive PX2 paper, but from quick read on Nvidia website, heres what i am guessing tesla AP2 might be.
They are using PX2 platform which provides all basic functions for mapping, positioning, AP, then modify them to fit their own sensors, radar and logic. Most higher level functions should be customized, PX2 probably was created so car companies can easily modify them. PX2 is sorta like android or IOS, you need to build your apps and data.
At first, I would run PX2 simulation on 200m miles of data in tesla lab, with adding fake/test map/sensor/radar/etc data to account for new sensors to work with the new software so test cars can start out with some basic brain.
Then, the shadow sensors start to collect real data and refine the AP brain and HD map, some automatic such as mapping and some might need engineer interventions such as decision making or random cases. Finally, in a couple months, it will be safe enough for basic AP to be turned on.
PX2 is great as huge amount of AP data can be processed in each of the thousand/million cars and learned results sent to Tesla over the air to update HD map and AP brain then update all cars. The new powerful chip allows this to be done within the car computer, which was not possible previously. Speed limit/right hand are the least of the problems. Decision tree and dealing with sensor limitations and random cases are likely the most difficult.
Tesla is way ahead of everyone as they have figured it all out. The demo video you saw was proof of AP 2 being in place and working quite well, just need more refinement. Level 5 will be achieved with enough data and refinements, only a matter of time. 2 years time frame is very doable in my opinion.
- Status
Similar threads
- Locked
