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

Neural Networks

This site may earn commission on affiliate links.
Tried 10.4 on roundabouts last evening.

I'm afraid it's going to need 10x the processing power and better decision making than humans to successfully navigate one of those. :(

With that said, I was amazed at how well it was doing for the typical non-roundabout driving.

No, it’s programmed to limit the turn radious, it’s not a cpu problem, they haven’t enabled the logic yet for the extreme torque of the wheel needed to make a round about.
 
There are threads about HW 2 cars suddenly getting much better on EAP with the latest software down loads. Also, reports of large downloads of software and files related to self driving are happening. Side cameras are calibrating. All this points to something major coming soon. If this is foreshadowing FSD roll out, it will be interesting to see how Tesla handles rollout. They tend to trust their customers much more than most companies. AP was rolled out very beta and no one was killed. I’m betting they do the same thing in a short time with FSD. Even my HW1 car has gotten much smarter lately. I think someone at Tesla has solved a big problem.

Show me your dev screen with your side cameras calibrating, mine still says 0%, so I’m very eager to see yours please.
 
  • Like
Reactions: yak-55 and croman
Drove 200 miles today in North Georgia with 2018.10.4 on winding rural roads with plenty of crests, dips, poorly marked roads etc. It is a route I've regularly driven with AP1 and prior versions of AP2. Well, hats off to@Tesla for this version. It really is a significant step forward, and makes AP very useful on rural roads.

1) Tesla can consider the basic Autosteer on freeways and rural highways done. Works very well.
2) The problem of 'crest wandering' is fixed.
3) The detection of 'around the curve' moving in-path objects is much better.
4) Lane change very smooth

Next to do:
1) Visual recognition of speed signs - Come on guys, this should not be too hard with the NN...
2) Detection of stopped vehicles ahead - harder, but I guess possible with specific NN training...
3) Lane change on non-freeway roads - Also come on guys, just enable it like on AP1. Not hard...

After this, then all the edge cases of roundabouts, EAP lane changes, etc.

.... only then even think about FSD (sniff...)
 
You all convinced me to take a drive last night to see how improved things really were (they are; I won’t go into detail here since so many others have already commented and demonstrated and posted video...) but I did want to mention that I felt the auto high beams were working phenomenally last night.

Could totally be placebo, but it was something I noticed.

I also believe some adjustments had to be made to the car-control side of things. Previously, the car seemed aware it wasn’t in the lane - it KNEW it was over the lane lines, but last night, it actually steered correctly to stay IN the lanes that it knew were there. No doubt the NN updates probably play a very large role, but actual car control seems vastly improved too. (Maybe just a case of “less garbage in / less garbage out?”)

Agreed about the high beams. I used them last night too and they definitely worked better.
 
Can someone explain this? I tried the new 18.10.4 AP on local roads where you can go 50mph which at times drops to 30 mph. If have set my max speed to "relative +6mph" of the posted speed limit I would expect the speed to drop to 36mph when the posted speed drops to 30mph and there is a speed camera shortly after that. I noticed my car was still going at 56mph (50mph+ 6mph) even tho it noticed the drop in posted speed limit. i had to break quickly. This happened more after the new update. I noticed this on the hwy too where it still remembered my local speed limit and did not go to my set hwy speed limit 71 mph (65+6). I had to accelerate quickly as I was still traveling at 46mph still.
It is possible that I am noticing all this now as I am using AP most of the time both local and on hwy after the new update.
 
  • Helpful
Reactions: Snerruc
I thought shrimpy is signaling left? The Merc came in hot, but to see that you need a rear corner radar. Can't see in the IC what it warned for really.

To tell you the truth, I didn't realize this whole thing happened until watching the video... at the time I couldn't figure out why it beeped, it was different. It was only because of the motorcycle buzzing me from behind earlier in the day that I thought to look at this video again. My working theory is that these are the side rear facing cameras picking this up.

There seems to be a couple of new safety features I noticed that are very subtle. One other feature I noticed is that on particularly difficult roads where the turn may be tricky, the hold steering wheel alarm comes on without disengaging.... it's something different than normal. Or perhaps, the NN is so good now that the alarm is now noticeable because the car is staying on path.
 
There seems to be a couple of new safety features I noticed that are very subtle. One other feature I noticed is that on particularly difficult roads where the turn may be tricky, the hold steering wheel alarm comes on without disengaging.... it's something different than normal. Or perhaps, the NN is so good now that the alarm is now noticeable because the car is staying on path.

I’ve heard that sound (the same put-your-hands-on-the-wheel-or-else beeps) used in other situations than wheel torque timeouts for quite a few versions now.

I think it happens when the perception confidence drops below a certain threshold. You can try it by enabling Autosteer on well marked roads, and then moving to difficult (e.g. very curvy) roads with faded lines: the car will give you a seemingly infinite number of beeps asking for your hands, but those are not counted towards the three-strike quota.
 
  • Informative
Reactions: buttershrimp
Its not $3k (unless you mean AP1 which was $3k at the end). EAP is $5k ($6k after delivery). I'm not sure you need a change order to add software (I bought it all including FSD). FSD should be worthwhile as well because they finally have a NN that could enable those features.

AP = Auto-Pilot
This is not listed as an option or as a standard feature. Is this just the old version of EAP or is it part of the standard software?

EAP = Enhanced Auto-Pilot
This sells for $5000 extra and I bought it. It's $6k after delivery.

FSD = Full Self-Driving
This requires EAP and sells for $3000 extra and I did not buy it. It's $4k after delivery.

My original question was whether the improvements you all are seeing are part of EAP or FSD.
 
AP = Auto-Pilot
This is not listed as an option or as a standard feature. Is this just the old version of EAP or is it part of the standard software?

EAP = Enhanced Auto-Pilot
This sells for $5000 extra and I bought it. It's $6k after delivery.

FSD = Full Self-Driving
This requires EAP and sells for $3000 extra and I did not buy it. It's $4k after delivery.

My original question was whether the improvements you all are seeing are part of EAP or FSD.

There’s still no E so I’ll say it’s AP functionality
 
I just came across this thread on reddit and I'm curious about the OP's claims. He claims that for driving only the wide angle camera is used and there's no loss of functionality when the main and narrow camera are covered.

How does this fit with the things @jimmy_d and @verygreen have found out? I was under the impression that the wide angle camera was only used for rain detection via the fisheye_wiper net. Didn't AP2 originally use some sort of stereo approach with the main and narrow cam running the same NN for driving?
 
I just came across this thread on reddit and I'm curious about the OP's claims. He claims that for driving only the wide angle camera is used and there's no loss of functionality when the main and narrow camera are covered.

How does this fit with the things @jimmy_d and @verygreen have found out? I was under the impression that the wide angle camera was only used for rain detection via the fisheye_wiper net. Didn't AP2 originally use some sort of stereo approach with the main and narrow cam running the same NN for driving?

Things have moved on a long way since that post!
 
I just came across this thread on reddit and I'm curious about the OP's claims. He claims that for driving only the wide angle camera is used and there's no loss of functionality when the main and narrow camera are covered.

How does this fit with the things @jimmy_d and @verygreen have found out? I was under the impression that the wide angle camera was only used for rain detection via the fisheye_wiper net. Didn't AP2 originally use some sort of stereo approach with the main and narrow cam running the same NN for driving?

That post is wrong it’s main and narrow being used for driving wide is just rain sensing currently.
 
  • Informative
Reactions: zmarty
Things have moved on a long way since that post!

Which post? I'm talking about @jimmy_d 's post from last thursday where he analyzed the 2018.10.4 NNs. He's still talking about one NN for the main and narrow cam with no mention of it being potentially also used on the fisheye cam.

That post is wrong it’s main and narrow being used for driving wide is just rain sensing currently.

So how does the reddit OP's Autopilot still operate when everything but the fisheye camera is blocked off?
 
Fantastic thread.... Probably the best way to look at this is to consider the new NN's have given AP2 a "new pair of glasses". It can see better, but is not making any new decisions nor is it recognising anything new. It's much much better at seeing what we thought it saw before, road markings, rain, headlights.

Also I implied is that the NN are computationally more efficient which may support Elon's tweets about AP2 vs AP2.5 hardware.

Roll on updates to decision making.

-malc-
 
Which post? I'm talking about @jimmy_d 's post from last thursday where he analyzed the 2018.10.4 NNs. He's still talking about one NN for the main and narrow cam with no mention of it being potentially also used on the fisheye cam.



So how does the reddit OP's Autopilot still operate when everything but the fisheye camera is blocked off?

I don't recall the Reddit OP providing a picture of how he did this test. I would suspect that the cameras weren't truly blocked off to the point that it impacted AP performance. The car actually doesn't need much of a field of view, just like how you can still sorta drive when your vision is partially obstructed.

Taping the mirror housing that's several inches away is a terrible way of testing which cameras are active. All the cameras would have a somewhat shared field of view at windshield distance.
 
  • Like
Reactions: lunitiks