I don’t have any more insight than anyone else, but my guess is that this has to do with the shuffle of the leadership on AP. The last major update was first released shortly before Karpathy took over and was pretty disastrous in just how long it took to get it working solidly. I’m assuming it’s taking so long because Tesla is getting their house in order, potentially even rethinking the software architecture.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
The next big thing in AP 2.0
- Thread starter Keith909man
- Start date
-
- Tags
- Model S
I don’t have any more insight than anyone else, but my guess is that this has to do with the shuffle of the leadership on AP. The last major update was first released shortly before Karpathy took over and was pretty disastrous in just how long it took to get it working solidly. I’m assuming it’s taking so long because Tesla is getting their house in order, potentially even rethinking the software architecture.
I also think the differing views on the forum are healthy.
I'd like to believe you.... but I've been diagnosed with TDHD, (Tesla Delusional Hyperactivity Disorder). Sadly, the doctors think I'm 3 to 4 months away from even entertaining the thought of Volvo or Mercedes having something close to autopilot... and let's just say that GM super-cruise has to be someone pulling ol'buttershrimp's legs.... I mean, "GM" and "super-cruise" go together like "Skoda" and "ultimate-handling"
timvracer
Member
By the way, just so that this does not get too off topic, there are nice threads and ponderings about these alternative driver's aids and self-steering systems:
Audi's L3 and now L4 being discussed: First L3 Self Driving Car - Audi A8 world premieres in Barcelona
GM Supercuirse: GM and Cruise announce first mass-production self-driving car
Audi's L3 and now L4 being discussed: First L3 Self Driving Car - Audi A8 world premieres in Barcelona
GM Supercuirse: GM and Cruise announce first mass-production self-driving car
Ulmo
Active Member
I think Elon's lack of growth of his management skills and his inability to handle politics and human resources optimally is not equal to his companies' lack of vision of helping the future, and I believe that the Tesla Semi is actually going to be a great thing. Everything Tesla is actually releasing as a product is part of that path. Let's see:
- Model 3: a few (hundred) have delivered. They have millions to go. Bad ratio. "S curve looks bad in the beginning" is the excuse. Seems believable (similar happened on prior models from same company). This is a case where they're currently actually delivering.
- PowerWall: Apparently, utilities and their butt-kissers (government, investors, etc.) just hate competition. Who knew! Everyone knew. The paperwork of introducing PowerWall to the California political landscape, much less the whole world, is daunting, and the ramp up is taking a huge amount of time. Once again, Elon's management and political skills lacking is an issue, but I think they are headed in the right direction, and they will figure it out. I can see first hand that this process is proceeding, since I actually have one set and it works great in everything that it does and its software potential is quite good. PowerWalls make Solar actually 100% clean day and night.
- Solar Roof: still in development. First alpha products on roofs now.
- Solar Panels: I haven't heard anything. Someone let me know.
- Tesla Semi: will reduce pollution for a lot of stuff.
- Boring Co: will create better driving and traveling opportunities. Still in infancy.
- SpaceX: will allow pressure relief for our tiny planet. Made great progress! No doubt about this.
- Hyperloop: better than the bogus Bullet Train (30 minutes from San Francisco to Los Angeles --- oops, that was a lie, Governor).
Has Tesla slowed in its rate of progress? Yes, from an external view. Simple obvious things to customers simply are not done, broken, done poorly. But, from the internal deep-in-corporate view of Tesla, things are going pretty well: Model 3 is actually delivering. Soon, it may even start delivering to real buyers that aren't themselves employees. This is going to be real.
Your position is that it is not real, and it's a ponzi scheme. I fear that too, but by every sign, that's not what I see; what I see, is something rising slowly up, but it doesn't go real fast.
Oh, hey, they built a battery factory. So what? It's not actually a product in itself. Elon said it's actually a product in itself. We don't see the product.
Um, they're already shipping. I have one (set of PowerWalls). Soon, others will have one (Model 3 car). They see Model 3's driving around all the time and post pics. Many of us are experiencing the first PowerWalls from the new battery factory. It feels minimal. It is. But it's not nothing. It's a monster being born and waking up. We get to hear the Goo Goo Ga Gas.
I owned a Model S (refresh 60D). It was real. Not fake at all. Did I want it to be better for that price? Yes. Was it a good electric car? Yes. They're coming out with Model 3 to fix the price issue. Model 3 is using the battery factory.
I own PowerWalls (v2). They are real. They are not fake at all. They work great. Do I want to the software to be improved a huge amount and integrated better with other companies' products? Absolutely. Does it still work OK? Yes, very much so. They're still working on the software end, and this is early in their ramp up. PowerWalls are using the battery factory.
Last edited:
All the hardware in the world is not going to fix the fact they clearly don't have the software working yet.
NerdUno
Member
I think what worries me is seeing the minimal progress that's been made with AP2 over the past six months and then trying to measure how long it will take to roll out a functioning FSD. EAP is like walking to the grocery store while FSD is akin to flying to the moon in the scheme of things. I simply don't think they have the resources or the talent pool to pull it off successfully. I also worry that Elon has become bored and frustrated with the AP2 progress and may just decide to get out of the weeds and move on from Tesla and work on other projects. When is the last time Elon even tweeted about AP2? Competition from other companies producing autonomous vehicles (that actually work) will only hasten his departure.
I think Kapanarthy is rearchitecting the system / learning algorithm's. If you pay attention inside the choas there's evidence that something was clearly wrong with the neural network setup. Chris Latner had to do a complete rewrite of the architecture for better precision here from his resume Chris Lattner's Resume: "
Now, if you watch Kapanarthy on twitter he's a "HUGE" proponent of Pytorch since version .2 shipped a few weeks ago, I strongly believe he's redoing the image recognition system in pytorch and he advocates for a different approaching learning for the network, seen
here:
Andrej Karpathy on Twitter
and here:
Andrej Karpathy on Twitter
So, all this together to me explains the "stall" in AP features. wih .36 we got back "two cars" in front. I imagine, in October / November we're going to get multiple lane recognition and vechiles in those lanes.
No chance off a FSD drive at the end of the year. In fact, if we can get car recognition by the end of the year, i'll be surprised. Aka on the IC we show a truck for a truck or a bike for a bike etc.
And no, don't even think about sign recognition this year.
It's already mid september, given Tesla's pace of progress I estimate we have maybe 2 - 3 more releases this year....
- I advocated for and drove a major rewrite of the deep net architecture in the vision stack, leading to significantly better precision, recall, and inference performance.
Now, if you watch Kapanarthy on twitter he's a "HUGE" proponent of Pytorch since version .2 shipped a few weeks ago, I strongly believe he's redoing the image recognition system in pytorch and he advocates for a different approaching learning for the network, seen
here:
Andrej Karpathy on Twitter
and here:
Andrej Karpathy on Twitter
So, all this together to me explains the "stall" in AP features. wih .36 we got back "two cars" in front. I imagine, in October / November we're going to get multiple lane recognition and vechiles in those lanes.
No chance off a FSD drive at the end of the year. In fact, if we can get car recognition by the end of the year, i'll be surprised. Aka on the IC we show a truck for a truck or a bike for a bike etc.
And no, don't even think about sign recognition this year.
It's already mid september, given Tesla's pace of progress I estimate we have maybe 2 - 3 more releases this year....
Yes, it seems like they've had 2 pretty hard resets on AP progress with the new directors. There was no theory of operation that transcended AP directors - the directors have all brought their own approaches to the problem. Luckily, maybe, I think Elon always leaned toward a more (not completely, but more) black box NN approach and that is what Karpathy is more likely to be able implement. Whether it's actually possible or not is a completely different question.
I've been trying to reply to this thread for a few days,
The AP thing is sort of ridiculous - it's not just being upset with the future not being delivered today, it is that they really did in November sell features as imminent that still aren't here now. In fact, it wasn't advanced payment for features, it was a come bet that still hasn't played.
Just by coincidence, I got the Tesla when friends of mine got a E350, A6, and SRX. I know they're not the same, but I did get the "cheap" MS. You can go through all of them and do the standard pros/cons thing until the cows come home. The construction of the Tesla is definitely funny. The leather feels suspiciously like vinyl, fit and finish is acceptable (nothing more), CarPlay would be very welcome, and so on. The E350 has a beautiful interior, but the interface is really awkward just to put the seat warmers on and the back seat is tiny. The A6's interior is every bit as cheap looking IMO as the MS and way more busy. And the engine start/stop at lights is maddening. The SRX isn't in the same class but people keep bringing up Cadillac b.c of the AP thing, but it is an abomination of styling in and out and rattles abound.
There is no perfect car, but the Tesla is pretty awesome. Never going to a gas station is awesome. It is a fantastic car to drive; it's not like driving any other car. I wish I had gotten the air suspension though.
To me the real test of whether I'll get another one is what happens in the next year or so with M3 rollout. It's a tremendous challenge they've undertaken and one that is basically make or break. And I think a lot of the resolution of gripes go hand in hand with that rollout. Service has to improve to handle M3 demands. AP has to work for the M3 to be what I gather a large chunk of reservations holders expect it to be. Quality has to increase to be able to maintain M3 throughput, and so on. When my lease is up, they're either going to be running lean and mean or it's going to be a cluster; I don't think there's much room for a middle road.
I leased specifically because I didn't want it to be a "dream car" for me. It's transportation for 3 years, and with a lease I limited the down side. That being said, I share a lot of the complaints with posters here.So...what am I missing here? Why do (seemingly) most of you all have Teslas? Am I (and other potential buyers) making a mistake? Would you buy a Tesla again? Sorry...had to vent. Which apparently this forum is great for that
The AP thing is sort of ridiculous - it's not just being upset with the future not being delivered today, it is that they really did in November sell features as imminent that still aren't here now. In fact, it wasn't advanced payment for features, it was a come bet that still hasn't played.
Just by coincidence, I got the Tesla when friends of mine got a E350, A6, and SRX. I know they're not the same, but I did get the "cheap" MS. You can go through all of them and do the standard pros/cons thing until the cows come home. The construction of the Tesla is definitely funny. The leather feels suspiciously like vinyl, fit and finish is acceptable (nothing more), CarPlay would be very welcome, and so on. The E350 has a beautiful interior, but the interface is really awkward just to put the seat warmers on and the back seat is tiny. The A6's interior is every bit as cheap looking IMO as the MS and way more busy. And the engine start/stop at lights is maddening. The SRX isn't in the same class but people keep bringing up Cadillac b.c of the AP thing, but it is an abomination of styling in and out and rattles abound.
There is no perfect car, but the Tesla is pretty awesome. Never going to a gas station is awesome. It is a fantastic car to drive; it's not like driving any other car. I wish I had gotten the air suspension though.
To me the real test of whether I'll get another one is what happens in the next year or so with M3 rollout. It's a tremendous challenge they've undertaken and one that is basically make or break. And I think a lot of the resolution of gripes go hand in hand with that rollout. Service has to improve to handle M3 demands. AP has to work for the M3 to be what I gather a large chunk of reservations holders expect it to be. Quality has to increase to be able to maintain M3 throughput, and so on. When my lease is up, they're either going to be running lean and mean or it's going to be a cluster; I don't think there's much room for a middle road.
That's a really nice post.
Interesting points, thank you @BigD0g. Calling @verygreen @lunitiks for comments.
Thank you @_jal_ for a nice post.
Thank you @_jal_ for a nice post.
timvracer
Member
Being in the software industry, and while no expert on FSD, I can say that it is of worthy note that the hype around the various FSD prototypes (Google, Uber, etc.) have gone silent! Why do you think that is?
The number of variables involved making true FSD work is staggering, and far beyond what any today's ML/NN tech can solve. This isn't just about hardware, and processing power, but it is almost akin to sentience. We are nowhere near that. While it is incredibly impressive what the google car can do (with LOADS of hardware all over the car, spinning radars, etc.), getting it to 90% isn't enough (and I don't think it is there yet). The last 5-10% is the true challenge, and I venture to guess that last mile problem is staggering.
Our brains can rapidly process information and recognize/classify it extremely fast, like that ball rolling from the driveway, the motorcycle that is lane splitting, the patch of ice in the road ahead, the deep water under the overpass during a rainstorm requiring significant slowing to prevent hydroplaning, that 2X4 piece of lumber that fell off a truck, construction zones with altered lanes, and in many cases, no lane markers (not on GPS, not on any map, MUST be recognized by the car hardware and interpreted). The number of anomalies in trying to park and find a parking spot alone is staggering. The list goes on and on.
There is a reason the momentum for FSD stalled... we are talking 10 years at least before it is real in terms of true driverless solutions on a very limited stage. We can get better and better at AP solutions in the meantime, but folks, Elon is full of it when he promises FSD anytime soon.
And yes, he said it, it's in the marketing materials, it is BS. That said, anyone considering the car should just shelve the expectations and promises, and judge the car for what it CAN do, what it is today -- and with those expectations in place, you end up with what I consider one of the best cars on the road today -- at least for me!
The number of variables involved making true FSD work is staggering, and far beyond what any today's ML/NN tech can solve. This isn't just about hardware, and processing power, but it is almost akin to sentience. We are nowhere near that. While it is incredibly impressive what the google car can do (with LOADS of hardware all over the car, spinning radars, etc.), getting it to 90% isn't enough (and I don't think it is there yet). The last 5-10% is the true challenge, and I venture to guess that last mile problem is staggering.
Our brains can rapidly process information and recognize/classify it extremely fast, like that ball rolling from the driveway, the motorcycle that is lane splitting, the patch of ice in the road ahead, the deep water under the overpass during a rainstorm requiring significant slowing to prevent hydroplaning, that 2X4 piece of lumber that fell off a truck, construction zones with altered lanes, and in many cases, no lane markers (not on GPS, not on any map, MUST be recognized by the car hardware and interpreted). The number of anomalies in trying to park and find a parking spot alone is staggering. The list goes on and on.
There is a reason the momentum for FSD stalled... we are talking 10 years at least before it is real in terms of true driverless solutions on a very limited stage. We can get better and better at AP solutions in the meantime, but folks, Elon is full of it when he promises FSD anytime soon.
And yes, he said it, it's in the marketing materials, it is BS. That said, anyone considering the car should just shelve the expectations and promises, and judge the car for what it CAN do, what it is today -- and with those expectations in place, you end up with what I consider one of the best cars on the road today -- at least for me!
kvandivo
Member
This.
I think a mind like Elon's has to be doing crazy new and exciting things, and AP no longer qualifies as being crazy new to him. For where your tweets are, there your heart will be also. I'm seeing SpaceX, Boring, Tesla Semis.. Nothing related to "yesterday's news".
It seems that the way it was introduced to the AP2 team is that Elon got this crazy idea, and introduced it for the team as a goal with a very steep progress curve and a great height of failure. Result is that even though the work might have been doable, several key engineers quit due to pressure.
Of course if the AP team was involved in the decisionmaking and with setting the timeline, they would feel that they have been a part of the decision and be motivated to deliver it on time. They would view the Autopilot as their baby, and not a threat from outside.
That is if the public statements made by the engineers that quit is true. If it's true, it's clearly a basic management mistake #1: Top-down management, and not involving the correct people in the decision-making.
Of course if the AP team was involved in the decisionmaking and with setting the timeline, they would feel that they have been a part of the decision and be motivated to deliver it on time. They would view the Autopilot as their baby, and not a threat from outside.
That is if the public statements made by the engineers that quit is true. If it's true, it's clearly a basic management mistake #1: Top-down management, and not involving the correct people in the decision-making.
OilSucks
Member
Wow! I really like Tesla and am very serious / very close to buying a model S...I was thinking some time in Q4. But the more I read the posts in this forum the more I second-guess my thought process. Is it really that bad out there? I hear about how terrible navigation is. I hear about how there's really truly been no meaningful updates in the last few years (aside from facelift). I hear about the lack of compatibility with our mobile phones. I hear about the poor quality of materials used in the cars. I hear about how terrible service is. I hear about empty promises. I hear about cars that cost tens of thousands of dollars less yet are even better equipped and more advanced in some ways. I hear about the severe lack of creature comforts...The list kind of goes on and on. To add salt to my wounds - one of my friends works for a Tesla supplier and says the same thing AmpedRealtor said in his post (hence the reason I copied his post; I however am not trying to single out this particular individual). So...what am I missing here? Why do (seemingly) most of you all have Teslas? Am I (and other potential buyers) making a mistake? Would you buy a Tesla again? Sorry...had to vent. Which apparently this forum is great for that
Don't take those criticisms to heart too much. Most people complain louder than those who praise Tesla. I have my gripes with Tesla but by and large, I wouldn't drive any other car even if it was free. I love my car and just wish there was a little more honestly behind the sale and owner process but again, there is no other car I want or would drive. Other than a Model X....
Having been involved in several software intensive startups, there's a really nasty trap companies often fall into. It consists of management first promising stretch goals to customers and/or VC lenders based on nothing but aspiration. Then the engineers, generally being good soldiers, actually try to do it without really having any idea or plan of how it will work. Everything is focused on the schedule (and maybe budget).
Time marches on.
Sometime between 6 and 18 months in, it becomes apparent that the now partially implemented system is never going to work and a different approach is necessary. If you're lucky, and smart, you learned a lot the first time, and could actually do it the second. If not, it's just a stinking pile.
However, in either case, time and money are gone.
Indeed. But that is also caused by a really basic management issue: Not involving the people essential for doing the work enough in the decision-making process.
By involving the developers first hand in decision-making you:
- Make sure the developers are motivated to get it working by taking their input seriously from the beginning.
- Make sure development starts in the right end by getting essential input.
- You get realistic schedule and cost expectations from the beginning.
- Prevent demotivating the developers / make their guard raise when put to solve someone else's problem.
It's really basic modern leadership. To get the company going you have to work with the entire group and treat them like humen whose input is important. I'd be surprised if the truth turns out the management forces their vision down on sceptical engineers. That's not a way you'd want to go if you want to reach your goal.
Similar threads
- Locked
