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

Investor Engineering Discussions

This site may earn commission on affiliate links.
Moving to the photon thing is huge. Hasn't been as much discussion as there should have been on this.
  1. Seeing in the dark etc. has been mentioned but all sorts of great things will come from this - some that Tesla may not expect. Bad weather, seeing through trees etc.
  2. V11 - single stack has been delayed - presumably because effort has been diverted to "project photon"
  3. Project photon is in alpha but to get beta to 10.8 level could take a long time. Minimum March imo.
  4. Will 10.9 happen or will there be a wait to March? If 10.9 etc. continues it is presumably because they won't be throwing away the dev. However, I would think they will be throwing a lot of it away. Entire NNs that are 1) processing the individual cameras, 2) stitching together the surround vision, 3) creating the Birdseye view, 4) interpreting the Birdseye view will all be gone
  5. If they continue with 10.9 etc. and they get to 10.13 by March then project photon has to be even better to replace - ie. might not come until May
  6. My preference is to stop all non project photon work
 
Moving to the photon thing is huge. Hasn't been as much discussion as there should have been on this.
  1. Seeing in the dark etc. has been mentioned but all sorts of great things will come from this - some that Tesla may not expect. Bad weather, seeing through trees etc.
  2. V11 - single stack has been delayed - presumably because effort has been diverted to "project photon"
  3. Project photon is in alpha but to get beta to 10.8 level could take a long time. Minimum March imo.
  4. Will 10.9 happen or will there be a wait to March? If 10.9 etc. continues it is presumably because they won't be throwing away the dev. However, I would think they will be throwing a lot of it away. Entire NNs that are 1) processing the individual cameras, 2) stitching together the surround vision, 3) creating the Birdseye view, 4) interpreting the Birdseye view will all be gone
  5. If they continue with 10.9 etc. and they get to 10.13 by March then project photon has to be even better to replace - ie. might not come until May
  6. My preference is to stop all non project photon work
One can make a split between two parts of FSD:
Detection, identification, and kinematics of objects in the scene. (Translation to vector space)
Driving behavior based on those objects. (Reaction to vector space)

Raw photon processing mainly impacts the first part, so improving driving behavior with the current image type is not wasted effort.
Front end/ step 1 will require a bunch of work to recreate the data sets with raw data. Fortunately, Tesla has been improving their labeling infrastructure (both auto and manual) so that will go faster this time around.
 
  • Like
Reactions: Buckminster
How can photons travel through trees?
Way off the point, but X-rays are photons...
In the visual space: Processing data from multiple displaced images can penetrate foliage to some extent.

Other fun techniques use scattered/ bounced light to see around corners. Like headlights illuminating pole strung phone lines. Though this works better with a scanned source like a CRT.
 
And, just saying this in general: People are pretty blame good at spotting tigers through foliage; some are better at it than others. The reasons are obvious: Those who don't spot tend more to get eaten, leaving the uneaten with better photon processing to reproduce.
That which the human eyeball (not, I should mention, a very good camera) and built-in neural networking can do, so can a bunch of better cameras and some hot-dog neural networking also do.
Note that I'm not saying that any of this is easy, mind you: but we've got nature's proof of operation in front of us, meaning that it's possible for machinery to do the same.
 
Heat Pump

Tesla called out this potential failure mode in the patent in the context of a high waste heat mode which may nit have been implemented. While this extreme mode may not be implemented, other sections do also call out preheating with the compressor drive electronics and allowing metered amounts of liquid into the compressor .
Screenshot_20220116-075127_Adobe Acrobat.jpg
 
  • Informative
Reactions: CorneliusXX
Per HP patent, system is called out as having modes where the compressor is intentionally flooded (partially) to increase performance. However, if too much liquid enters, it may result in the damage seen in non-Tesla teardowns.
I think there could be something to this concern (damage to the compressor due to too much liquid), but Green is jumping to an extreme conclusion that this is happening "en-masse", due to a single post he came across. Also, from what I've read, most modern scroll compressor designs are capable of handling flooded conditions, so I'm not convinced that that's the cause.
 
  • Like
Reactions: MC3OZ and MP3Mike
I think there could be something to this concern (damage to the compressor due to too much liquid), but Green is jumping to an extreme conclusion that this is happening "en-masse", due to a single post he came across. Also, from what I've read, most modern scroll compressor designs are capable of handling flooded conditions, so I'm not convinced that that's the cause.
Yeah, that term is a bit much, if meant literally.
My limited understanding is that the compressor can handle some flooding; however, this condition can also flush out the lubricant/ oil resulting in metal on metal contact and accelerated wear. Depending on design (this may be more relevent to home HVAC units), the compressor bearing lubrication can also be impacted.
 
  • Like
Reactions: petit_bateau
I think there could be something to this concern (damage to the compressor due to too much liquid), but Green is jumping to an extreme conclusion that this is happening "en-masse", due to a single post he came across. Also, from what I've read, most modern scroll compressor designs are capable of handling flooded conditions, so I'm not convinced that that's the cause.
He often does. Like with Radar.
 
I would take those Russian sources with a grain of salt.
Tesla has no presence in Russia, they can only import privately.. so very few Teslas. Also model Y has only been in europe for 5 months, there is no used market.. where would Russians buy these "en masse"?
Agree, literal definition of en masse seems hyperbolic at this point. He does clarify later than % failure is not known.
As to Russia, it's status as an unsupported region is the reason why third party tear downs exist.
 
How many of these extreme cold heat pump 'failures' do we know have been an actual hardware failure of any sort? From what I've seen the biggest issue seems to be sensors thinking something has gone wrong due to the vent being held open by snow build up and the software shutting the pump off.
Apart from that one broken pump in Russia, what other evidence is there that shows hardware is failing?