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]

Eh? How do you think the Space Shuttle worked?
The entire concept of the Space Shuttle is based on reuse.
Even Russia tried to copy it.

Just like the "Russian Concorde" ...
The Shuttle was also a Cold War tool, just like the Concorde blue print kept on being circulated and the Russians built their own very flawed "ruskie concorde", the Shuttle was a way to drain the resources of the soviet union, with a deteriorating economic outlook.

Locally, Space Shuttle was already a jobs program - yes at that time.
The gov't contractors knew making it really reusable would kill their chance of making more money.

 

[diplomat33]

I think we will know in about 24 to 36 months whether or not Lidar is helpful or not. (could be less than that, but given "Elon time" I will stick with this estimate)

We already know lidar is helpful since it has helped companies achieve SAE L4 before anybody else. And companies are also using lidar for driver assist features on consumer cars. So we know that lidar is helpful in different ways. The real question is whether lidar is necessary for FSD (SAE L4 or L5), which is different than just being helpful. We would need someone to achieve reliable L4 or L5 without lidar to prove that lidar is not necessary for FSD. So far, that has not happened.

 

[BlindPass]

Very early in the driverless game. No winner yet. Some of us predict the no lidar camp will win. To give an example of the folly of the masses: No one was doing reusable rockets, until SpaceX came around. Now everyone is talking about reusable rockets. It took 15 years for SpaceX to make its point. So perhaps by 2030 we will see that Lidar is helpful, but not absolutely necessary.

One can’t yet win this early in the game, but it’s possible to lose.

Lidar could be better now, even if not the end solution. There can be value in having better intermediary steps.

What happens if it takes Tesla too much longer to prove they not only have the right end game, but are far enough along for that to benefit end-users? I guess with the social capital Tesla has built, they’ll be able to indefinitely sell a dream for $8k

 

[mikes_fsd]

We already know lidar is helpful since it has helped companies achieve SAE L4 before anybody else. And companies are also using lidar for driver assist features on consumer cars. So we know that lidar is helpful in different ways. The real question is whether lidar is necessary for FSD (SAE L4 or L5), which is different than just being helpful. We would need someone to achieve reliable L4 or L5 without lidar to prove that lidar is not necessary for FSD. So far, that has not happened.

I updated it to "needed" in my post.
I think Tesla already knows from internal alpha build testing that they can get the necessary accuracy to meet the current FSD suite they are selling.
I think it will be clearly visible to us - the actual end used (what a concept) - that Tesla Vision is sufficient (i.e. Lidar is not needed) in ~24 months time.

 

[DanCar]

we have companies that are already doing FSD.

Joe Schmo from cocomo, doesn't car about companies losing billions and saying they have FSD when Joe can't use FSD.

And with companies like Waymo and Cruise, we already know that FSD is possible with cameras, radar, lidar and HD maps.

Again, Joe doesn't care what is possible. Joe cares about what is real and he can use.

In other words, FSD isn't real to most of us until it is usable. So your arguments about FSD existing are hollow.

 

[ItsNotAboutTheMoney]

Joe Schmo from cocomo, doesn't car about companies losing billions and saying they have FSD when Joe can't use the FSD.
Again, Joe doesn't care what is possible. Joe cares about what is real and he can use.

In other words, FSD isn't real to most of us until it is usable. So your arguments about FSD existing are hollow.

Aside: Kokomo (with Ks), IN has had a Supercharger open since 2/25/2020.
Passed through once and had some Mexican at a restaurant in a strip mall.

/Aside

 

[diplomat33]

I think it will be clearly visible to us - the actual end used (what a concept) - that Tesla Vision is sufficient (i.e. Lidar is not needed) in ~24 months time.

Yes, we will see one way or the other when we get the software updates. I am more pessimistic than you. I don't think we will know in 24 months. Tesla would need to make massive progress. So far, AP is still very far from reliable enough for FSD in my experience. AP can be fine and then suddenly phantom brake seemingly out of the blue. Just today, my car phantom braked going through a green light for apparently no reason. And there is still so much that Tesla vision does not do yet.

 

[diplomat33]

Joe Schmo from cocomo, doesn't car about companies losing billions and saying they have FSD when Joe can't use FSD.
Again, Joe doesn't care what is possible. Joe cares about what is real and he can use.

In other words, FSD isn't real to most of us until it is usable. So your arguments about FSD existing are hollow.

Except that is not what we are talking about. You are changing the question. The question was not when will people be able to use FSD. The question was whether FSD exists and whether lidar is necessary. That is a purely technical question that is completely different.

 

[DanCar]

... The question was whether FSD exists and whether lidar is necessary. That is a purely technical question that is completely different.

Yes, FSD doesn't exist for most people. Exist by my definition = available and usable. I know you definition doesn't match that. Lidar being necessary, is purely hypothetical at this point, since no one is winning. I know you think Waymo is winning. I don't because they have been floundering for past few years and losing billions and no timeline when widely available. People can raise arguments why they think one is better than the other, but they aren't going to prove anything. When there is widely deployed FSD, then we can say there is an early winner. When FSD is ubiquitous, then we can say there is a definite winner and the technology used by the winner will provide some level of proof of which approach is better.

We may find that what is best for taxi service is not best for personal car ownership.

 

[Redbrick]

We already know lidar is helpful since it has helped companies achieve SAE L4 before anybody else. And companies are also using lidar for driver assist features on consumer cars. So we know that lidar is helpful in different ways. The real question is whether lidar is necessary for FSD (SAE L4 or L5), which is different than just being helpful. We would need someone to achieve reliable L4 or L5 without lidar to prove that lidar is not necessary for FSD. So far, that has not happened.

I've read several of your posts....so you really don't think fsd can be resolved with just cameras before lidar becomes ubiquitous? Comma ai thinks fsd in the wild with just cameras will be possible and tesla will be the first to get there, but will need driver monitoring. Heck, even comma's system does not use lidar.

 

[diplomat33]

I've several of your posts....so you really don't think fsd can be resolved with just cameras before lidar becomes ubiquitous?

No, I don't.

Here's the problem. With only one sensor, you cannot afford a single critical mistake since there is no back up. Keep in mind the complexity of driving too. Driving a complex intersection in the city, there might be a lot of cars moving in different directions, pedestrians, cyclists, traffic lights, yielding to other vehicles etc and your camera only has to do it all and cannot make a single critical mistake. It will take a lot longer to achieve that level of reliability. In fact, solving all the problems in driving to that level of reliability with camera-only is a daunting challenge. But if you have multiple sensors, then they can help other out. It's easier to solve the problem. And we already see FSD with lidar that works. Several companies have already been able to demo FSD with lidar, with no driver in the car at all. We have not seen that with camera-only yet. Mobileye has a great demo with camera-only but they say that camera only is not reliable enough to fully remove driver supervision so they plan to include lidar. So I do think that we will see lidar FSD become more ubiquitous before camera-only FSD is solved.

Comma ai thinks fsd in the wild with just cameras will be possible and tesla will be the first to get there, but will need driver monitoring.

IMO, Hotz is a joke. He's a hacker who pulled off a cheap knock off of AP1. He has no FSD. I don't value his opinion at the same level as the many experts working on autonomous driving who actually have FSD. And they are including lidar in their sensor suite and they say it is needed.

And if it requires driver monitoring then it is not FSD, it is a driver assist. Hotz is talking about doing driver assist that drives around, not FSD. Yes, you can absolutely do driver assist with camera only which Tesla is doing now. And that driver assist may even be able to drive around a bit but that is not FSD. FSD is defined as L4 or L5 where the car can drive without any human intervention or monitoring.

 

[mikes_fsd]

Here's the problem. With only one sensor,

That is the problem with your assessment, it is not "one sensor"...
It is redundant cameras, plus radar and ultrasonics.

 

[mikes_fsd]

I don't value his opinion at the same level as the many experts working on autonomous driving who actually have FSD. And they are including lidar in their sensor suite and they say it is needed.

But to be fair, you do not value Karpathy's opinion either, so @Redbrick take this with a large brick of salt.

 

[diplomat33]

That is the problem, it is not "one sensor"...
It is redundant cameras, plus radar and ultrasonics.

Tesla only uses one front radar and ultrasonics are very short range. So yes, in the front, Tesla has more than one sensor, I agree. In the front, Tesla has multiple cameras and a radar. The front is fine. But for the other directions, Tesla is only "one sensor" since at some angles, Tesla only has the one camera. Again, the ultrasonics only work for short distances so they are only useful in parking scenarios. Ultrasonics won't help you see objects that are further away, like a pedestrian or car that is several yards away. And to do reliable FSD, you need multiple sensors for all 360 degrees. Having multiple sensors for one direction but not in other direction is not good enough. What happens if you need to respond to objects in one of the directions that is only "one sensor"?

 

[diplomat33]

But to be fair, you do not value Karpathy's opinion either, so @Redbrick take this with a large brick of salt.

@Redbrick That is not true. I value Karpathy's opinion. Karpathy has ML expertise and years working with Tesla. I view Karpathy as far more of an expert than Hotz. I value Karpathy's opinion much higher than Hotz's opinion. But I don't "only" value Karpathy's opinion and no one else's.

 

[whitex]

I understand some of the anger because FSD isn't here yet, but seriously? It seems like no one has ever written software or designed hardware, let alone both. I know many of us have written or built stuff, and you should know it doesn't always happen as quickly as you think. And do you remember the last car you owned before a Tesla? How is that car doing with OTA updates? Is there anyone who can honestly say they'd rather go back to ICE?

Why would anyone go to ICE over auto pilot? It's not as if there is ICE you can buy which has level 5 autonomy. The biggest reason which might drive customers to go to ICE might be Tesla quality, parts, and service and not having an adequate EV alternative to Tesla today.

As for your question about OTA for my last pre-Teasla car, well, it didn't need OTA as it had all the features that I paid for available for me to enjoy when I picked it up, vs. Tesla customers who years later are still waiting (e.g. anyone who bought FSD in 2016), some of whom have sold their cars before seeing a single benefit for their money. Oh, and one OTA update once almost killed me (from windshield button became a frost button in one of the updates which happened just before i left on a coast to coast trip), and another update costed me $600 in damage to the car, because Tesla decided to auto unfold mirrors as soon as the car starts moving even if the driver manually folds them, which resulted in a different behavior than the prior year of driving, so when the car unfolded its mirrors where it never did before, one of the mirrors crashed into a wall (Tesla fixed it almost immediately in the very next update, but didn't cover the $600 mirror repair). So, after 7 years of Tesla OTA, not so crazy about - it allows Tesla to sell unfinished product (sometimes complete vaporware which is never delivered because it's always "coming sooner than anyone expects") and allows Tesla to use their customers as QA. I do believe OTA can be a great benefit to automobiles, but not the way Tesla uses it. Security and bug fixes should be included and supported for the life of the car, while new features (including any new UI) should be optional and could even be paid, making money for the manufacturer, but sold only after they are ready to use.

 

[DanCar]

... Lidar could be better now, even if not the end solution. ...

Lidar could be worse now, adds complexity, power, compute cost.

What happens if it takes Tesla too much longer to prove they not only have the right end game, but are far enough along for that to benefit end-users?

What happens if it takes Waymo too much longer while losing billions?

I guess with the social capital Tesla has built, they’ll be able to indefinitely sell a dream for $8k

I suppose with the social capital Waymo has built, they'll be able to indefinitely fool the gullible while losing billions.

 

[Stratman]

That is the problem with your assessment, it is not "one sensor"...
It is redundant cameras, plus radar and ultrasonics.

I think another big issue for Tesla with the current suite of cameras is how poorly they perform at night. What was clearly seen in daytime is totally missed by the camera at night. Our eyes perform so much better in low luminance than the cameras.

 

[DanCar]

I think another big issue for Tesla with the current suite of cameras is how poorly they perform at night. What was clearly seen in daytime is totally missed by the camera at night. Our eyes perform so much better in low luminance than the cameras.

Can Tesla turn up the gain on current cameras in dark situations? The original cameras were described as optimized for low light, if I remember correctly.

 

[powertoold]

Achieving reliable fsd with just cameras is impossible to me, but Elon has achieved the impossible (for many) before, and he seems very confident at this time. I'd like to be surprised. Again, I think the rewrite is the "final straw." If the rewrite doesn't allow us to see the light, then it'll likely never happen.