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

Autonomous Car Progress

This site may earn commission on affiliate links.
George Hotz answers the question: How close are we to self-driving? Seven minute video with Lex Fridman.


TLDW: George’s thinking is similar to Elon’s. They both have figured out how to do it, and just need a year or two to get more data and compute power to execute their plans.

GSP
They solved auto pilot? Years ago? Lol…. My model 3 from 2019 on vision only now would like a word. Going from 83 mph down to 50 mph on a major highway just because the Webcams think heat mirages on a wide open road are objects or something mandating hard braking…
 
PS what you posted above is also incorrect. Tesla's snapshots do not include captured RAW at 30fps, ...
The way I understood him is he saying if Tesla did capture everything then the size would be xxx. Since Tesla doesn't capture that much data, then Tesla doesn't capture all the data and therefore doesn't capture billions of miles of all the data. Therefore one can question the assertion of Tesla billions of miles of data collected, since in reality the amount of data collected is very sparse.
 
Last edited:
The way I understood him is he saying if Tesla did capture everything then the size would be xxx. Since Tesla doesn't capture that much data, then Tesla doesn't capture all the data and therefore doesn't capture billions of miles of all the data. Therefore one can question the assertion of Tesla billions of miles of data collected, since in reality the amount of data collected is very sparse.
But the point is that's an easily defeated strawman (one of many commonly set up), no one has claimed Tesla is capturing every single bit of data the car generates in those billion miles, nor does it makes sense not to use compression where it is appropriate (as Tesla demonstratively does using h265). Also doesn't change most of what he posted is factually incorrect and thus highly misleading to those trying to understand Tesla's system and data gathering.
 
Last edited:
The way I understood him is he saying if Tesla did capture everything then the size would be xxx. Since Tesla doesn't capture that much data, then Tesla doesn't capture all the data and therefore doesn't capture billions of miles of all the data. Therefore one can question the assertion of Tesla billions of miles of data collected, since in reality the amount of data collected is very sparse.

Talking to these people is a complete waste of time. They are so full of blind tesla worship making them incapable of reason or any logic whatsoever.

I mean this same person that is calling you out and @GSP @diplomat33 for liking my post in which i stated 100% facts (The exact size of 10 seconds video from all 8 camera which is 2+GB, The fact that the new camera is 5x much larger and the fact that because of the huge size Tesla only collected 10 seconds from one camera (turns out to be two which is reasonable as i was going from memory) which amounts to 300-500mb. But the full 10 seconds from all 8 cameras would be 2+gb which is certainly considered "gigabytes".

Yet he himself liked KS post in which he stated absolute none-sense and completely incorrect information about the video data Tesla collects, that "Each camera captures at 30 megs of data per minute. 10 seconds of video, per camera, is only 5 MB. Times 8 cameras that's 40 MB." He actually liked it as "informative".

You can't make this stuff up. These fans love for Tesla has completely destroyed their ability to make judgment when Tesla is involved. They don't see anything but Tesla.

but even then your math is hilariously wrong

Each camera captures at 30 megs of data per minute

That means 10 seconds of video, per camera, is only 5 MB. Times 8 cameras that's 40 MB. Even if the interior cam is included that's 45MB for 10 seconds of video.

Spoiler: 45MB is much smaller than "Gigabyes of data"

Wow you are clearly moving goal posts and somehow claiming you are still right, when you made an incorrect statement and people are giving you thumbs up for it? @GSP @DanCar @diplomat33, I question what is the logic you guys see?

I hope everyone regardless of whether you believe Tesla is the best thing ever sees how illogical and beyond reason some Tesla fans are.
 
But the point is that's an easily defeated strawman (one of many commonly set up), no one has claimed Tesla is capturing every single bit of data the car generates in those billion miles, nor does it makes sense not to use compression where it is appropriate (as Tesla demonstratively does using h265). Also doesn't change most of what he posted is factually incorrect and thus highly misleading to those trying to understand Tesla's system and data gathering.

Everything I posted were 100% factual while you posted absolute nonsense (oh you said one camera instead of two...wtf?) and you even liked a completely factually incorrect post and called it informative. Instead of being ashamed for liking a post telling us that "10 seconds of video, per camera, is only 5 MB. Times 8 cameras that's 40 MB." You try to call others out for liking factually correct posts that stated 10 seconds from 8 cameras is 2.8 GB (gigabytes of data) and that Tesla collects 10 seconds from forward camera which is 300mb+.

This is the literal definition of: Your Brain on Tesla.
 
Expanded legal commercial autonomous operating zone in Beijing: from 60km² to 500km².

"On Friday, authorities in Beijing made a policy decision to allow eligible companies to operate driverless taxis without safety supervisors on board and charge passengers for the rides, marking a definite step in commercialization of intelligent transportation.​
...​
Now, the Office of the Beijing High-Level Autonomous Driving Demonstration Zone plans to allow commercial operation of the driverless taxis in an area of 500 sq km. Interested companies are required to apply for commercial operation permits.​
According to the office, 116 self-driving taxis have been operating in Beijing on a trial basis for the past four months. They have completed over 1.5 million test trips, and the total length of the trips reached nearly 2 million km. More than 95 percent of passenger ratings have been favorable."​

 
As I understand today's regulations, Ohio (where I am) does permit autonomous vehicle testing today on any road but not driverless testing. Specifically:

All AVs tested in Ohio must have a designated operator responsible for the safe operation of the vehicle while in use and compliance with all traffic laws and regulations, among other requirements. The governor may pause the testing of AVs in Ohio if there is evidence that the technology is not safe.

Pennsylvania, Arizona, Nebraska, and Kansas appear to be similar per Motortrend. Florida, Georgia, Nevada, North Carolina, North Dakota, Utah, Texas, Tennessee, and West Virginia do not even require a licensed driver for AV's. So by my count at least 14 states (many with safety drivers required, several without) allow for AV testing today.

I'm starting to wonder "how much is enough" when it comes to allowing AV testing before Tesla starts higher SAE level testing - all 50 states? Obviously the rest of the world is another question but it will start here. Obviously they know where the car is and could enable/disable nags based upon geographic location. I'm starting to think the "when regulations permit" is a crutch.

But I can also see Elon choking on his Cheerios when he thinks of taking on financial liability - to me, that's the bigger question especially with the current state of Tesla's AV capabilities. It's not regs. And I'm not sure it ever was. So if desired, if Tesla felt FSD was capable of any higher SAE levels (which it is not today), there are many, many people ready and willing to test those features legally.

*EDIT* - up to date by state can be found here --> https://www.iihs.org/topics/advanced-driver-assistance/autonomous-vehicle-laws

Looks like many states have come a long way in the last couple of years on this topic.
 
Last edited:
I'm starting to wonder "how much is enough" when it comes to allowing AV testing before Tesla starts higher SAE level testing - all 50 states? Obviously the rest of the world is another question but it will start here. Obviously they know where the car is and could enable/disable nags based upon geographic location. I'm starting to think the "when regulations permit" is a crutch.

It was always a crutch because Tesla could have started L4 testing years ago. There were plenty of States that allow L4 testing. And it did not have to be driverless testing, Tesla could have done L4 testing with a safety driver and reported miles and disengagements like many other AV companies have done. But for years now, Tesla has used L2 as an excuse to be exempt from reporting AV miles and disengagements as required by the CA DMV (the only time they reported AV miles is when they did the demo for the 2016 video and for Autonomy Day).
 
But I can also see Elon choking on his Cheerios when he thinks of taking on financial liability - to me, that's the bigger question especially with the current state of Tesla's AV capabilities. It's not regs. And I'm not sure it ever was. So if desired, if Tesla felt FSD was capable of any higher SAE levels (which it is not today), there are many, many people ready and willing to test those features legally.
This may be why Tesla got into auto insurance; it'll ease Tesla into the liability game. Obviously, insurance is already a way to transfer financial liability to the insurance company, but if Tesla Insurance says that FSD's safety score is 92, then everyone with a safety score below that would have an incentive to use FSD. Tesla Insurance can go farther by saying that if there is an at-fault event while FSD is driving, then there are no consequences to the driver's insurance policy. They can also reduce rates for people who normally have a worse safety score. Then they keep bumping FSD's safety score.

My point is that instead of Tesla one day saying "We take liability for FSD accidents", they will use Tesla Insurance to create a transition - one that they can advance or back up as needed.
 
  • Like
Reactions: KArnold
This may be why Tesla got into auto insurance; it'll ease Tesla into the liability game. Obviously, insurance is already a way to transfer financial liability to the insurance company, but if Tesla Insurance says that FSD's safety score is 92, then everyone with a safety score below that would have an incentive to use FSD. Tesla Insurance can go farther by saying that if there is an at-fault event while FSD is driving, then there are no consequences to the driver's insurance policy. They can also reduce rates for people who normally have a worse safety score. Then they keep bumping FSD's safety score.

My point is that instead of Tesla one day saying "We take liability for FSD accidents", they will use Tesla Insurance to create a transition - one that they can advance or back up as needed.
liability for accidents and lawsuits caused by a corporation *greatly* exceed the insurance limits you see on a normal Tesla car insurance. Tesla also don't own an insurance company inhouse... they simply partner with lesser known insurance providers in various states... If Tesla is liable they need to either buy real insurance (LLYODS etc) or hold back hundreds of millions $ on their balance sheet for lawsuits
 
  • Like
Reactions: Doggydogworld
liability for accidents and lawsuits caused by a corporation *greatly* exceed the insurance limits you see on a normal Tesla car insurance. Tesla also don't own an insurance company inhouse... they simply partner with lesser known insurance providers in various states... If Tesla is liable they need to either buy real insurance (LLYODS etc) or hold back hundreds of millions $ on their balance sheet for lawsuits

They've been underwriting their own insurance for over a year, now: Tesla Insurance looks to expand in two more states, underwriting itself for the first time

"In these new states, the policies will be underwritten by Tesla General Insurance for the first time, as opposed to partner companies in other states."
 
Everything I posted were 100% factual while you posted absolute nonsense (oh you said one camera instead of two...wtf?) and you even liked a completely factually incorrect post and called it informative. Instead of being ashamed for liking a post telling us that "10 seconds of video, per camera, is only 5 MB. Times 8 cameras that's 40 MB."
His post overall is more accurate than yours given it correctly points out that much of the useful data is non-video data that takes much less space and can be uploaded with very low bandwidth requirements. Something you fail to acknowledge by continually focusing on a strawman of uploading 8 cameras of 30 fps uncompressed RAW which makes zero sense and no one has ever suggest Tesla is doing.

While he uses the dash cam estimates for the video, which are more highly compressed, he is at least correct that much of the video data is compressed (except it is in h265 frames, not the dashcam h264). If the situation calls for it nothing is stopping Tesla from using h264 dashcam frames either for data gathering. Point is uploading all RAW at 30 fps with zero compression (not even zip which as below RAW is easily losslessly compressible with zip) is completely unnecessary.
You try to call others out for liking factually correct posts that stated 10 seconds from 8 cameras is 2.8 GB (gigabytes of data) and that Tesla collects 10 seconds from forward camera which is 300mb+.

This is the literal definition of: Your Brain on Tesla.
LOL you just double down on this? To make it more clear

300 MB+ of this:

"10 seconds .h265 at 30fps from main and narrow. And 10 seconds RAW at 1 fps from all 7 B&W cameras."

Does not equal

"10 seconds from forward camera which is 300mb+"

To take one example snapshot, the whole snap shot is 251MB (including 10 seconds of h265 video from 2 cameras, 10 seconds of 1fps RAW from 7 cameras, radar data, CAN data).

Extracting forward camera, uncompressed from zip, 10 seconds of 1fps RAW is 23.9 MB (losslessly compresses with zip to 16.2 MB), 10 seconds of 30 fps H.265 is 72 MB.
 
Last edited:
  • Like
Reactions: Knightshade
Here's an article with an attempt to contact CPUC on why (they didn't comment) and also got comments from various parties (both supportive and against).
Cruise and Waymo’s 24/7 San Francisco Robotaxi Operations Delayed Again

From experience with what happened with NEM, delaying is a sign CPUC has doubts about their original direction. A lot of people dismissed the concerns raised by SF (about driverless operation specifically in case people try to bring up SF's prior approval about safety driver operation) and the effectiveness of the protest, but I think two delays shows there are serious questions about just rubber stamping another driverless expansion. Obviously things have changed from half a year ago.
 
I'm starting to wonder "how much is enough" when it comes to allowing AV testing before Tesla starts higher SAE level testing - all 50 states? Obviously the rest of the world is another question but it will start here. Obviously they know where the car is and could enable/disable nags based upon geographic location. I'm starting to think the "when regulations permit" is a crutch.

1689087423671.png
 
Here's an article with an attempt to contact CPUC on why (they didn't comment) and also got comments from various parties (both supportive and against).
Cruise and Waymo’s 24/7 San Francisco Robotaxi Operations Delayed Again

From experience with what happened with NEM, delaying is a sign CPUC has doubts about their original direction. A lot of people dismissed the concerns raised by SF (about driverless operation specifically in case people try to bring up SF's prior approval about safety driver operation) and the effectiveness of the protest, but I think two delays shows there are serious questions about just rubber stamping another driverless expansion. Obviously things have changed from half a year ago.

I disagree with the term "rubber stamp". That term means approving something automatically without proper consideration. The CPUC permit follows a lengthy application process. And the CPUC carefully considered all the facts for a long time. And Waymo and Cruise met all their criteria for a permit. Furthermore, the CPUC permit process follows a lengthy permit process with the CA DMV. So it is certainly not a rubber stamp. The SFMTA just resented not being a part of that process but it was not a rubber stamp.

And by the way, Waymo is addressing the concerns. They work closely with first responders and provide training. They also make it possible for first responders to communicate directly with Waymo via car speaker and mic or by calling a number. First responders can also disengage the Waymo and move it in an emergency. So I think that addresses the concerns of first responders who say they can't do anything when a driverless car gets stuck.


I want driverless cars to be as safe and reliable as possible. So if the delays means Cruise and Waymo have time to address issues and we get better driverless cars, then great. But the SFMTA should not get veto power over the deployment of driverless cars. I don't think it will happen but if Waymo and Cruise do put safety drivers back in, there needs to be clear goals and standards for when they can remove safety drivers again. If it is needed, I would be ok with putting safety drivers back in temporarily while they sort out some issues and make the autonomous driving more reliable. But I am not ok if this is a back handed way to just ban driverless cars completely. And we know there are some groups that have an anti-driverless agenda. They are not interested in Waymo and Cruise improving their software and going driverless, they want to ban driverless for good. That, I am not ok with.
 
  • Like
Reactions: texas_star_TM3
super convenient crutch though.... when you purchase FSD for a whopping $15k ... Tesla states this:

"The currently enabled features require active driver supervision and do not make the vehicle autonomous. The activation and use of these features are dependent on achieving reliability far in excess of human drivers as demonstrated by billions of miles of experience, as well as regulatory approval, which may take longer in some jurisdictions. As these self-driving features evolve, your car will be continuously upgraded through over-the-air software updates."

that's a lot of if's 😅 which gets me back to the point someone made earlier that Tesla could just roll out L4 at their own discretion later this year. They currently clearly do *NOT* agree with that approach and conventiently call out "regulatory approval" as a dependency. so what is it?
 
I disagree with the term "rubber stamp". That term means approving something automatically without proper consideration. The CPUC permit follows a lengthy application process. And the CPUC carefully considered all the facts for a long time. And Waymo and Cruise met all their criteria for a permit. Furthermore, the CPUC permit process follows a lengthy permit process with the CA DMV. So it is certainly not a rubber stamp. The SFMTA just resented not being a part of that process but it was not a rubber stamp.
The argument is the CPUC permit process and criteria was set long ago based on safety driver operation, while they didn't wait for driverless operation statistics to come out before approving back in January. As such it was not "proper consideration".

As discussed previously.
"a decision that the Commission adopted long before there were any driverless AV operations on San Francisco streets and before the driverless readiness problems the City has documented were apparent.5 Since that time, Waymo driverless AVs have committed numerous violations that would preclude any teenager from getting a California Driver’s License."

Autonomous Car Progress

From the SF paper, updated link here. Back when Waymo was approved in January, their cumulative reported incidents were in single digits, after driverless started in mass, it skyrocketed to 30 a month.

Cruise was averaging around 20 a month previously, but it also increased to around 60 a month.

Pausing the approval allows for "proper consideration". If they just approved it with a vote, while dismissing concerns brought up, as they did previously, that would be "rubber stamp".
And by the way, Waymo is addressing the concerns. They work closely with first responders and provide training. They also make it possible for first responders to communicate directly with Waymo via car speaker and mic or by calling a number. First responders can also disengage the Waymo and move it in an emergency. So I think that addresses the concerns of first responders who say they can't do anything when a driverless car gets stuck.

I'm glad they are finally doing this, but this still leaves Cruise and notice this is being done only after their expansion approval was clearly blocked, six months after the concerns were raised.
I want driverless cars to be as safe and reliable as possible. So if the delays means Cruise and Waymo have time to address issues and we get better driverless cars, then great. But the SFMTA should not get veto power over the deployment of driverless cars. I don't think it will happen but if Waymo and Cruise do put safety drivers back in, there needs to be clear goals and standards for when they can remove safety drivers again. If it is needed, I would be ok with putting safety drivers back in temporarily while they sort out some issues and make the autonomous driving more reliable. But I am not ok if this is a back handed way to just ban driverless cars completely. And we know there are some groups that have an anti-driverless agenda. They are not interested in Waymo and Cruise improving their software and going driverless, they want to ban driverless for good. That, I am not ok with.
To be clear, it was never suggested that SF was trying to ban driverless cars! In fact they are very supportive. All they want addressed is that the cars don't randomly halt in the middle of the roads blocking traffic and at minimum that emergency responders have ways to quickly move them if they do. It is great there is finally some movement in that front, as that's what they have been calling for the whole time!

Even the protesters aren't using a method to ban driverless cars! A lot of you dismissed the protest method, but it's actually brilliant if you think about it. They don't put cones on cars that have passengers or a safety driver already in it. For cars with a safety driver, it's a non issue even if they did (safety driver can just take it off). Or if operators have a rapid response team, that team also can quickly remove it (and the same team can respond to the incidents being complained about). It doesn't damage the vehicles, does not put it in danger, and has practically no liability to the protester (no police officer will bother writing a ticket for a person that put a cone on a stationary car). Nothing about it bans driverless cars, while it gives the operators an incentive to have a way to quickly respond to halts.