Parking Lot Summon - Our Vehicles compared to Ants in search of Food

[SOULPEDL]

Anyone else notice these patterns in public parking lots? (I'm on AP 2.5, V42.4, Model 3 in Chandler).

Now compare same location on Google maps...

Notice upper right showing construction. Got me thinking, how will the car respond? I assume it will prioritize paths where it won't need to back up (doing serpentine or loops), but this construction area is a clear dead end and the car won't know it.

Now consider that the side cameras alone have a dead spot directly behind (is the rear camera even used for Navigation as it doesn't show up in the videos of "what the car sees"). So can they self drive in reverse? My prediction is no for now, Summon will fail, and you'll have to go rescue your car. I don't think this will be frequent, but this map data is also about 1 yr old now. So there would likely be a cleanup period while we collectively fix the maps through these failures. In other words, the cars will learn like ants, finding new paths around the failures... and that's really cool! Just think about the power of the collective!

This also begs the question: How are parking lots mapped? I assume it's tracking common paths (like TeslaFi maps of routes taken, getting darker with more travel), but how are some paths going to be deleted? I bet that's where our rescue operations come in so be prepared - at first

I simply love watching this unfold and was extremely impressed by Pilot on Navigation. Elon, if you thought PoN was "Difficult", Summon may be your most difficult challenge to date when you consider pedestrians, other cars, and no "center-lane markings." If you figure this one out, I'm convinced you're on your well on your way to full FSD (paid $3K for it up front and never looked back). And to think this is still just an EAP feature...

Realistically confident, cautious, and patient. A shame so many members don't see AI this way and just complain at the first hiccup. Oh well, I'm having a blast in the front seat!

 

[1375mlm]

I'm convinced you're on your well on your way to full FSD (paid $3K for it up front and never looked back)

You seriously deserve an award

 

[zhu-]

I didn't get FSD as even if it does become available in my state in the next year or so, I'd probably trust a 11 year old to drive my car more than an AI... for now. Ironic since I work in tech with machine learning, but our code isn't autonomously moving around 4000 pound hunks of metal that cost $60k and liable to cause additional damage.

 

[SOULPEDL]

You seriously deserve an award

Guessing you're a non-believer who'd rather pay $5,000 or don't think FSD is possible?
Sorry to hear that.

 

[SOULPEDL]

I didn't get FSD as even if it does become available in my state in the next year or so, I'd probably trust a 11 year old to drive my car more than an AI... for now. Ironic since I work in tech with machine learning, but our code isn't autonomously moving around 4000 pound hunks of metal that cost $60k and liable to cause additional damage.

Loving all the confidence in the room. Better get a Cadillac instead, especially coming from an AI expert on the matter.

 

[SSonnentag]

I paid for FSD, but have serious doubts about ever being able to trust it too. I'll be happy if I can just lose the nags while driving on highways and interstates.

I've not seen or noticed the parking lot "roads" before. Interesting.

 

[1375mlm]

Guessing you're a non-believer who'd rather pay $5,000 or don't think FSD is possible?
Sorry to hear that.

More so the opposite, just with all this "FSD isn't possible" talk, I think its nice to hear someone who does believe it's true.

I remember seeing a mobileye video with ONLY cameras, NO radar, NO ultrasonic sensors, and was said to be full FSD

It reminds me of the Skydio R1 Drone, which has an insane 12 cameras that use a similar approach to flying. They recognize objects and estimate distances, effectively making a 3D map of its surroundings.
It works exceptionally well, and even figures out alternate routes around things in its way.

Because of seeing the success with Mobileye and that drone, I actually do have faith for Tesla's approach, ESPECIALLY since they've added ultrasonic sensors and radar.
My only gripe with the approach, and the reason I left the comment, is because of the time it's taking.
The drone was developed for years, and it's still not 100% fool proof. And its just a drone. It can handle light impacts without an owner being frustrated..
For cars, not the other way around. When a Tesla bumps into something on AP, its a big deal. This means that before they release a feature (such as FSD features), they are making sure it's as foolproof as they can make it (while also emphasizing its beta form)
So while I do believe in FSD coming, I'm quite positive it's not coming anytime soon, which is my reason of commending your confidence in it.

 

[SOULPEDL]

So while I do believe in FSD coming, I'm quite positive it's not coming anytime soon, which is my reason of commending your confidence in it.

Two thumbs up to that. Why I ended it with "patience" - this generation has none, along with general consumers where most technology today is so simple so they expect it all. Meanwhile, I bet it will be another 5 years before EAP is out of Beta, and I doubt you'd get it for under $10K - It could even go subscription based if you value upgrades with continued improvements (and Tesla insurance, literally).

I point to Moore's Law which is still playing out. AI will be the same and we're still just in the Atari ages by comparison. I think part of this perception problem is not just the FUD (who also convinced most that Tesla would fail), but that people don't understand statistics. And when this car goes FSD, when I'm not in it, great! When I am, I'll still probably take over on imminent threats.

I'm keeping it real in that it will likely not be a better driver (overall) than I for another 10-15 yrs (although it could be said that's some have already been surpassed). What I do see is that it compliments my driving today for a safer net result. For example, it's really good at keeping between the lines 99.9% of the time and warning/stopping when something is ahead... especially people! So we're a team, the car and I, and we're getting better everyday.

What I also expect is that it will become more and more an extension of my eyes and ears and our aggregate chance of survival out there will continue to increase over time. My favorite part is watching this play out everyday! And I'd bet my prediction in parking lots has already been addressed. How else could it work?

 

[zhu-]

FSD will absolutely become a reality, I have zero doubts about that. I don't think anybody here is really questioning the validity of the long term possibilities of FSD. Many of the arguments are in the short term value of the $3000 commitment. I am 100% sure in the future, majority of cars will be self driving. Question is, is that a future 5 years away? 10? 30? 50? If 10 years away, I'm personally not going to risk my time dealing with insurance in the next few years if/when my car hits something or someone. I'd rather just pay for a driver or Uber as the "insurance premium" of my car not hitting anything because it isn't self-driving and wait for the technology to mature. And with that said, in 10 years I can double my $3000 as an absolute minimum so I don't see much value in pre-ordering at the discount available today.

 

[croman]

Tesla just needs to release some features. FSD, like Autopilot, is a misnomer. FSD could merely be a collection of features like AP. Tesla can incrementally provide benefits as those features are developed and mature.

Tesla hints those features depend on deploying new hardware featuring a custom TPU (Jimmy, rightfully calls out a systolic array). Once that is in the wild I do expect basic FSD features and improved functionality of the capabilities of the NNs that can't even run on the current hw.

 

[1375mlm]

it's really good at keeping between the lines 99.9% of the time and warning/stopping when something is ahead...

FSD will absolutely become a reality, I have zero doubts about that. I don't think anybody here is really questioning the validity of the long term possibilities of FSD.

I saw an article (business insider?) that summed up every significant Tesla (negative) event into a list. And reading through it, I kept thinking “oh yeah, man that sucked!”
But now it’s all water under the bridge. Stuff like the funding secured fiasco, April fools joke, the 2nd death on AP, & even the whole Elon vs. media thing...
There has been such a negative focus on Tesla, that I feel like Tesla has no other choice except to play it safe.
Although I’m extremely sorry for the families of the two people that lost their lives on AP, it’s pretty clear that they weren’t using it correctly. And to be killed in the safest cars on the road speaks volumes about the severity of the crash.
I think if news outlets didn’t jump at the chance to make a clickbait title that bashed one of the most innovative companies existing right now, Tesla would be much more open to sharing AP features still in development.
Perhaps we could have ~80% accuracy red light recognition...
And by testing the dev features, we may be able to offer more valuable insight to the AP development.
My point is that the constant negative attention has only retarded Tesla’s growth, which is quite ridiculous since the mission of the company is for a clean earth... they’re not trying to fight anyone

 

[djgarrett21]

Guessing you're a non-believer who'd rather pay $5,000 or don't think FSD is possible?
Sorry to hear that.

I like how you phrased that, "non-believer". The level of faith some have in Elon boarders religious devotion, everyone else to them is a non-believer.

 

[SOULPEDL]

The level of faith some have in Elon borders religious devotion, everyone else to them is a non-believer.

It is a cult of sorts. Too bad people are so polarized, mostly fueled by the FUD as @1375mlm mentioned. It's a natural reaction to bears trying to destroy the company (this well documented as I mostly hang out in the market analysis forum). It will settle out once they exit their short positions in the next 2 quarters and Tesla gets on the S&P 500 perhaps. Powerful Chairwoman helps too.

So I like the cautious approach... people out there are naive and/or love the edge and then sue for any reason and the bears are all over it. Case in point, car fires... ridiculous history on that.

When I drive, I enjoy waiting until the car has a complete fail before I take over. I tell my wife (while she grabs the door), "I wonder what it's going to do now..." But I also grew up in Canada driving in snow so I'm used to last minute corrections. I even applied to be a "Test Operator", there's actually only one job posting in Phx - ground zero for FSD and a slight change in tactics for Tesla beyond just their simulations.

So back to the Summon. Remote control from your phone before X-mas will be a $tock(ing) stuffer for sure. And if that's possible, then the car must be self aware or it would be crazy to let it crash while playing it like a drone. Summon is imminent and it will fail before it doesn't. Nothing new, just a heads up.

 

[SOULPEDL]

Perhaps we could have ~80% accuracy red light recognition...

I test this occasionally, but I don't think they'd dare try it until AP3 hardware + time. People would ignore driving for sure. However... a red warning, slow down, then disable FSD for the drive might save lives. But if they stop the car on red, then we're back to ignoring it again. IMO we need 99.999 accuracy with the sun as your biggest enemy. It needs to be even more aware of the situation and traffic flow. Then what if there's a traffic cop instead?

I bet it's not about detecting lights, instead more about learning all the different light patterns and rules by city. There are some odd-ball ones out there like 6 way intersections, left turns after the green, not to mention international. This knowledge is in deep memory and we underestimate the complexity of just a red light.

 

[BOBTHEJOCKEY]

Anyone else notice these patterns in public parking lots? (I'm on AP 2.5, V42.4, Model 3 in Chandler).
What's EAP?
View attachment 350817

Now compare same location on Google maps...
View attachment 350818
Notice upper right showing construction. Got me thinking, how will the car respond? I assume it will prioritize paths where it won't need to back up (doing serpentine or loops), but this construction area is a clear dead end and the car won't know it.

Now consider that the side cameras alone have a dead spot directly behind (is the rear camera even used for Navigation as it doesn't show up in the videos of "what the car sees"). So can they self drive in reverse? My prediction is no for now, Summon will fail, and you'll have to go rescue your car. I don't think this will be frequent, but this map data is also about 1 yr old now. So there would likely be a cleanup period while we collectively fix the maps through these failures. In other words, the cars will learn like ants, finding new paths around the failures... and that's really cool! Just think about the power of the collective!

This also begs the question: How are parking lots mapped? I assume it's tracking common paths (like TeslaFi maps of routes taken, getting darker with more travel), but how are some paths going to be deleted? I bet that's where our rescue operations come in so be prepared - at first

I simply love watching this unfold and was extremely impressed by Pilot on Navigation. Elon, if you thought PoN was "Difficult", Summon may be your most difficult challenge to date when you consider pedestrians, other cars, and no "center-lane markings." If you figure this one out, I'm convinced you're on your well on your way to full FSD (paid $3K for it up front and never looked back). And to think this is still just an EAP feature...

Realistically confident, cautious, and patient. A shame so many members don't see AI this way and just complain at the first hiccup. Oh well, I'm having a blast in the front seat!