Blog Musk Touts ‘Quantum Leap” in Full Self-Driving Performance

A “quantum leap” improvement is coming to Tesla’s Autopilot software in six to 10 weeks, Chief Executive Elon Musk said a tweet.

Musk called the new software a “fundamental architectural rewrite, not an incremental tweak.”

https://twitter.com/x/status/1294374864657162240

Musk said his personal car is running a “bleeding edge alpha build” of the software, which he also mentioned during Tesla’s Q2 earnings. 

“So it’s almost getting to the point where I can go from my house to work with no interventions, despite going through construction and widely varying situations,” Musk said on the earnings call. “So this is why I am very confident about full self-driving functionality being complete by the end of this year, is because I’m literally driving it.”

Tesla’s Full Self-Driving software has been slow to roll out against the company’s promises. Musk previously said a Tesla would drive from Los Angeles to New York using the Full Self Driving feature by the end of 2019. The company didn’t meet that goal. So, it will be interesting to see the state of Autopilot at the end of 2020.

 

[mikes_fsd]

But not FSD. AP is L2.

Here is that cognitive dissonance again...

And there is the rub....
Even with their L2, Tesla is ACTUALLY making a quantifiable difference in safety on our roads.

Now, read that sentence again, slowly.
Their stated goal of creating/adding Autopilot to the original Model S was to make the roads safer, because someone close was killed on the road (I think it was a cyclist).
They are clearly already doing that. Nowhere done with it.

 

[diplomat33]

And there is the rub....
Even with their L2, Tesla is ACTUALLY making a quantifiable difference in safety on our roads.

Now, read that sentence again, slowly.
Their stated goal of creating/adding Autopilot to the original Model S was to make the roads safer, because someone close was killed on the road (I think it was a cyclist).
They are clearly already doing that. Nowhere done with it.

You are moving the goal posts. We were talking about FSD, about a car being able to drive with no human in the driver seat. But now, you want to talk about AP safety and claim Tesla is achieving their goals even if they don't have FSD.

 

[mikes_fsd]

But ok, let's talk about the Autonomy Day demo. ..... That's good enough for you?

Demo's are of no use to me, that was for you.
I have no idea if you would have been satisfied with 1000 miles or 1 million miles.

What good is the Tesla demo?
What good is they Waymo demo?
Does it save lives because it is actually on the roads, can Waymo quantify their safety like Tesla did in their safety report?

 

[mikes_fsd]

You are moving the goal posts.

No, I am pointing out the cognitive dissonance that you seem to have.
Becuase you have a Tesla, you daily can compare FUTURE promises by Tesla to your CURRENT car and come away disappointed that the FUTURE is not here yet.
You CANNOT do that with a Waymo or Cruise (pick a flavor) car.
So, you take their videos and their speeches as gospel and since there is nothing you can compare it to and test like you do daily with your Tesla, you do not have insight into any inconsistencies between their demo videos and their actual performance.

 

[Bladerskb]

And Tesla has the best ML/AI available to them... each one can claim the "best", does not make it so.

Every AI/NN breakthrough and advancement this decade has come from Google (DeepMind). Every single one.
Zero, Nil, Nada has come from Tesla. It has nothing to do with claiming.

Results, results, results. That will show the best. Not demo's, not speeches.

There's only one company with cars driving around an area the size of SF without a human driver 24/7 and that's Waymo.
If its so easy, why ain't anyone else doing it? There are hundreds of SDC companies. There are total tens of thousands of SDC engineers.
Why is only Waymo the only one currently running driverless fleets? Its so easy right? Shouldn't other companies be doing the same?

Training supercomputer should not be compared to the AI chip in the Tesla car itself, it should be compared to whatever actually goes into the Dojo supercomputer.

Just ONE single Google TPU (4.0) pod has over 300 petaflops and they come out with a new and improved chip every year. Yet Dojo is still years away in development. Also Google's Edge TPU (4.0) is also more efficient than HW3. The same can be said of Mobileye's EyeQ5 which is actually safety critical.

 

[mikes_fsd]

It seems that some ppl might be following this just from a technology point of view.
To me - as I stated in many places, FSD is a safety issue, manual (human) driving should be banned and considered dangerous, but we need a replacement to the manual driving, hence FSD.

My selfish goal is that my kid who is 10 years old, can get into my hand-me-down Model 3 and not have to do the driving by the time she needs to drive.
I only view the FSD solution through the lens of road safety, for those in and outside the car.
Manual driving can only be banned when FSD is ubiquitous and it can only be ubiquitous if it is affordable to the most ppl.

 

[diplomat33]

Demo's are of no use to me, that was for you.
I have no idea if you would have been satisfied with 1000 miles or 1 million miles.

What good is the Tesla demo?
What good is they Waymo demo?

Waymo has more than just demos. That's the point. The fact is that Waymo has FSD, Tesla does not.

Does it save lives because it is actually on the roads, can Waymo quantify their safety like Tesla did in their safety report?

Waymo has done 20M FSD miles with ZERO at-fault accidents. So yes, they can quantify their safety.

No, I am pointing out the cognitive dissonance that you seem to have.
Becuase you have a Tesla, you daily can compare FUTURE promises by Tesla to your CURRENT car and come away disappointed that the FUTURE is not here yet.
You CANNOT do that with a Waymo or Cruise (pick a flavor) car.
So, you take their videos and their speeches as gospel and since there is nothing you can compare it to and test like you do daily with your Tesla, you do not have insight into any inconsistencies between their demo videos and their actual performance.

That's ridiculous. I can see with my own eyes that Waymo has cars that can drive around with no human in the driver sear, not just once or twice for a demo, but for millions of miles. That's FSD. I am not making up that Waymo has FSD.

My selfish goal is that my kid who is 10 years old, can get into my hand-me-down Model 3 and not have to do the driving by the time she needs to drive.
I only view the FSD solution through the lens of road safety, for those in and outside the car.
Manual driving can only be banned when FSD is ubiquitous and it can only be ubiquitous if it is affordable to the most ppl.

I totally get that. And if you want Tesla to achieve FSD because you believe it will allow your kid to have a safer car, that's great. I support you. All I ask is that you don't lie about Waymo just because you can't ride in one. The fact is that Waymo has FSD and they have FSD that is better than Tesla. It is also a fact that Tesla does not have FSD yet. Don't deny reality just because it is an inconvenient truth to you.

 

[mikes_fsd]

All I ask is that you don't lie about Waymo just because you can't ride in one. The fact is that Waymo has FSD and they have FSD that is better than Tesla.

I can agree on most what you've said.
But here, you lie, in the same paragraph that you accuse me of lying!

Waymo does not have FSD that is better than Tesla.
First, you do not know what Tesla has.
If you are comparing Waymo to current Autopilot then you're deluding yourself.

ANYONE can claim to have better FSD, and even show video's (lucid air -- don't compare us to Tesla, but let us always compare ourselves to Tesla)
You know why? Because there is no independent testing of the two systems pitting each system against same set of criteria.
So, we need something like NCAP or IIHS testing of the systems.

 

[diplomat33]

Waymo does not have FSD that is better than Tesla.
First, you do not know what Tesla has.
If you are comparing Waymo to current Autopilot then you're deluding yourself.

Yes, Waymo has better FSD than Tesla. Waymo has cars that can completely drive themselves with no human in the driver seat, not just for a demo, but routinely, consistently, everyday, for millions of miles. They can handle complicated city driving, road hazards, construction zones, read hand gestures from police, pull over for first responder vehicles, all with no human input. Now you say we cannot compare Waymo to current AP. Fair enough. Tesla has better software than what we have in our cars. But is there any evidence that Tesla has that capability? I have not seen any indication of that. And if Tesla has that capability, why don't they release it then? The fact that Waymo has released L4 FSD to their fleet and Tesla has not released similar FSD to their fleet is evidence that Waymo is ahead.

 

[diplomat33]

ANYONE can claim to have better FSD, and even show video's (lucid air -- don't compare us to Tesla, but let us always compare ourselves to Tesla)
You know why? Because there is no independent testing of the two systems pitting each system against same set of criteria.
So, we need something like NCAP or IIHS testing of the systems.

Waymo is not just claiming they have FSD. They have shown proof that they have FSD.

And Waymo cars can easily pass these pedestrian detection tests since they have cars that routinely stop for pedestrians probably hundreds of times a day.

 

[Stratman]

These videos are almost a year old, but green posted them... what Autopilot sees vs what human sees.
Also remember that the recorded videos are very compressed where as the Autopilot computer gets HDR raw video.
Without autopilot overlays (what do you see?)
https://twitter.com/greentheonly/status/1192287875263541248
With AP overlays.
https://twitter.com/greentheonly/status/1192287994327244800

In those sub quality videos not much. But put any of us in that car at the same time and we would see and perceive a whole lot more.
I'm not convinced about the so called HDR raw feed either. I have driven at day and at night with cyclists beside me. In the day the cyclist is identified (not always) but at night the cyclist is just an ultrasonic flash on the screen at best. The software had no idea what it was because it could 'see' it low light with sufficient quality. I could clearly see and perceive the cyclist who was wearing lycra, and was peddling with his head down (yes it was a male) and was on the edge of the cyclist lane so even closer to the car.
These problems make me think FSD is a long way off, if ever with the current hardware. I am generally impressed with the auto pilot performance on a clear highway.

I would love FSD to be possible with the cars we have now and admire the chiefs optimism, but at best I think we we will get a kind of enhanced autopilot that can be used in a geo fenced environment.

 

[mikes_fsd]

Waymo has cars that can completely drive themselves with no human in the driver seat

How do you know Tesla doesn't have "cars that can completely drive themselves with no human in the driver seat"...
we saw 2 videos of cars "completely driving themselves" with drivers in the seat, would it change your mind if the same video from Tesla was published but without the driver?
Really?

 

[mikes_fsd]

But put any of us in that car at the same time and we would see and perceive a whole lot more.

Green, in the same tweet thread, says that there are many times the AP recognized something before he did.
Also, this was a demo on a Tesla equipped with 2.5 hardware.

Is it that difficult to believe that computers dedicated to always be watching for objects in ALL directions are going to be better than humans in almost (for now) all situations... I believe that is the case already and will only get better with time.

 

[Redbrick]

No I am not saying that. It would be absurd to say that something will never ever be possible. Given infinite time, I am sure Tesla would eventually achieve FSD on the current sensor suite. Although, I do think the Tesla sensor suite has some weaknesses. I was merely responding to whether Tesla can "solve FSD" before lidar becomes ubiquitous. I think FSD with camera+radar+lidar+HD maps is a much quicker and more reliable path to get to FSD. In fact, we can say that camera+radar+lidar+HD maps will be the first to achieve FSD because it already has.

fair enough, and you're right....call it for what it is today. Since I'm new to tesla, I hope this year brings us much closer to fsd. (...cause if it weren't for my wife I would've gone with anything but tesla..)

 

[diplomat33]

How do you know Tesla doesn't have "cars that can completely drive themselves with no human in the driver seat"...
we saw 2 videos of cars "completely driving themselves" with drivers in the seat, would it change your mind if the same video from Tesla was published but without the driver?
Really?

Because all we've we've gotten from Tesla are 2 demos that are not as good as what Waymo can do. I've yet to see Tesla demonstrate the kind of city driving that Waymo can do. Heck, I'd love to see a video of Elon's commute that he says is with the latest alpha build of the new rewrite. Then, we could see what Tesla's latest software is capable of. Heck, if Tesla has good FSD, why doesn't Tesla report autonomous miles to the CA DMV? That would be another source of data that would tell us that Tesla as FSD. Dozens of companies that have FSD, report it to the CA DMV every year.

But more importantly, it is not really about demos. Demos are just nice to showcase capabilities. It's about what is released for public roads. Tesla has still not released FSD to the fleet. Waymo has actually released L4 FSD to their fleet. Waymo has a fleet of robotaxis self-driving on public roads every day. That's the difference. I know Waymo has FSD because they have released FSD to their fleet of cars that are self-driving on public roads every day.

 

[mikes_fsd]

...cause if it weren't for my wife I would've gone with anything but tesla.

Since this is a topic on FSD, what exactly were you going to go with, in place of Tesla?

 

[Uncle Paul]

Dip:
What Waymo vehicle can you buy? WHats the point?

Think they only can drive inside a geo-mapped bubble. Outside they are non functional.

 

[diplomat33]

Dip:
What Waymo vehicle can you buy? WHats the point?

This is such a cliche. Waymo has chosen to focus on ride-hailing first, instead of consumer cars first. That does not make their FSD less real. A ride-hailing fleet is a great platform for testing and validating FSD. Tesla has a large fleet of great EVs but they are not FSD yet.

What's the point? The point is to actually get FSD. Waymo is working to finish FSD.

Think they only can drive inside a geo-mapped bubble. Outside they are non functional.

Not true. Waymo cars can self-drive outside the geofenced bubble.

 

[diplomat33]

Is it that difficult to believe that computers dedicated to always be watching for objects in ALL directions are going to be better than humans in almost (for now) all situations... I believe that is the case already and will only get better with time.

This is only true if the software is good enough and the sensors are good enough. For example, if you just have cameras and your camera vision is not very good, then no, it will not be better than humans. Conversely, if you have 360 degree advanced camera vision and radar and lidar, yes, it absolutely will be better than humans since you have 3 different sensor modes scanning in every direction.

 

[Uncle Paul]

Guess my point is that your debate is purely academic. Nobody wanting to purchase Waymo FSD can do so.

They might be able to work within tight parameters, but cannot function out on the open highway.

We saw Tesla produce a slick video, automatously driving down a carefully scripted route, but still not free range.

Have not kept up with Waymo recently. Are their cars still requiring human minders, or are they good enough to be autonomous.