Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
I was watching the demo video and it showed at least 1 front black and white camera view.

I don't know if it showed the other 2 front cameras, however non of the cameras were in color except for the go-pro that was being used behind the driver.
We know the current rear camera is color. As far as I could tell they just showed the side cameras and the medium range front camera, They didn't show the other front cams.
 
As far as I can tell, the video only really definitely demonstrated level 2 other than the parking sequence.

The driving mode-specic execution by one or more driver assistance systems of both steering and acceleration/ deceleration using information about the driving environment and with the expectation that the human driver perform all remaining aspects of the dynamic driving task

The car was definitely doing one or more of the driving tasks, (acceleration and steering) and we really could not tell if the car or driver was doing things like navigation and accident avoidance. ( it may have been, but it didn't look definitive.)

I think that level Is a fair assumption if we assume that the car really was doing everything (implied, but not proven)


(Level 3 requires that the driver be available to respond to a request for intervention. Which they certainly were)

From what I can tell, level 4 requires that the car can drive on its own Most of the time, with a fall back option if the driver fails to respond. (So, if the car has to drop off the kids and is confused by the school drop off zone, the car simply pulls over and sulks until someone helps.

Level 5 then means that the car can handle anything that is thrown at it. (That a human could handle)

I didn't see anything that said that there could NOT be a driver in a level 5 car, just that they would be unneeded. (To be fair, I don't have access to the full SAE document though).

So... the video doesn't prove level 4 or 5 since the drive is very limited and there was no chance to show much in the way of complexity.

I guess the parking sequence would probably count as level 4 (with no one there to take over...) but it was short and controlled enough environment that (to me) I would have trouble counting it as such.

Now don't get me wrong, it was a cool video, and it definitely showed that progress is being made and that some things are possible... but it didn't show a level 5 autonomous system (IMHO)
 
As far as I can tell, the video only really definitely demonstrated level 2 other than the parking sequence.

The driving mode-specic execution by one or more driver assistance systems of both steering and acceleration/ deceleration using information about the driving environment and with the expectation that the human driver perform all remaining aspects of the dynamic driving task

The car was definitely doing one or more of the driving tasks, (acceleration and steering) and we really could not tell if the car or driver was doing things like navigation and accident avoidance. ( it may have been, but it didn't look definitive.)

I think that level Is a fair assumption if we assume that the car really was doing everything (implied, but not proven)


(Level 3 requires that the driver be available to respond to a request for intervention. Which they certainly were)

From what I can tell, level 4 requires that the car can drive on its own Most of the time, with a fall back option if the driver fails to respond. (So, if the car has to drop off the kids and is confused by the school drop off zone, the car simply pulls over and sulks until someone helps.

Level 5 then means that the car can handle anything that is thrown at it. (That a human could handle)

I didn't see anything that said that there could NOT be a driver in a level 5 car, just that they would be unneeded. (To be fair, I don't have access to the full SAE document though).

So... the video doesn't prove level 4 or 5 since the drive is very limited and there was no chance to show much in the way of complexity.

I guess the parking sequence would probably count as level 4 (with no one there to take over...) but it was short and controlled enough environment that (to me) I would have trouble counting it as such.

Now don't get me wrong, it was a cool video, and it definitely showed that progress is being made and that some things are possible... but it didn't show a level 5 autonomous system (IMHO)

No, level 3/4. With SAE level 2 the human is still doing a lot of the work see this

The driver was there in case of intervention but did not fiddle with anything (supposedly) the entire time.

We will only know for sure when they show us an uncut video real time video.
 
Oops, I missed a character in my post... I meant to say "I think that level 3 Is a fair assumption if we assume that the car really was doing everything (implied, but not proven)" but I guess my fingers were not working at Level 5 yet <grin>.

I still think that all that was proven in the video was level 2 (other than the parking sequence, and as long as we can assume it was not totally faked) From my read of the SAE document the difference between level 2 and level 3 is that the human is doing at least one part of the dynamic driving task, which, from the SAE definitions, includes operational tasks like "the steering, braking, accelerating, monitoring the vehicle and roadway, and tactical tasks such as responding to events, determining when to change lanes, turn, use signals, etc." and the human in this case sure looked like they were monitoring the vehicle and roadway, and from other posts it did not look like navigation was enabled, meaning that they probably were also determining when to turn etc.. The human does not have to do a lot of the work, just "all remaining aspects", whatever those are.

However, as you said, without an uncut video of the drive in real time, showing him get in, tell the car where to go, and showing feet not touching the pedals as well, we really don't know either way.

To show level 4, there would have to be a point where the car asked for aid and the human driver did not respond, and the car handled it somehow. Also, to show level 4 OR 5 (to me at least), I would have had to see lots more scenarios covering more aspects of what the SAE is calling "driving modes", with much more complex and problematic cases. (which doesn't mean the car can't do those already, just that a single short film that was pretty obviously cut together doesn't prove it.)

Again, though, I am not saying that the video is "lying" or "faked" as some people in the media seem to be saying. I think it is showing autonomous driving at some level, and it is a reasonable next step from where we were and I think it shows a reasonable expectation of where we might be soon.
 
  • Like
Reactions: JeffK
I am not saying that the video is "lying" or "faked" as some people in the media seem to be saying
Totally agree, I'd maybe say that the video portrays the expected future behavior. I think the DNN models are still going to be trained extensively between now and then using data gathered from new vehicles with new hardware.

It's unrealistic to think that the car can do that behavior in any location at this time before the massive data collection.
 
Totally agree, I'd maybe say that the video portrays the expected future behavior. I think the DNN models are still going to be trained extensively between now and then using data gathered from new vehicles with new hardware.

It's unrealistic to think that the car can do that behavior in any location at this time before the massive data collection.

Yup, what he said!
 
  • Like
Reactions: JeffK
According to my understanding of the NHTSA definitions, there is no difference in the software/hardware capability between Levels 4 and 5.
Therefore, Level 4 and 5 would operate the same way. Since Tesla does not currently offer cars without steering wheels, brake pedals, nor accelerators, the Model X in the video was operating as a Level 4.

The first mistake is you mistook the NHTSA definitions with the definitions of SAE. There is no such thing as Level 5 within the NHTSA levels of automation.

The second mistake you made is assuming what Google is aiming for defines level 5 on the SAE's levels of self-driving, but it doesn't. Can you imagine Star Trek without the ability of the crew to manually override the controls?
 
Last edited:
  • Like
Reactions: MP3Mike
The first error is you mistook the NHTSA definitions with the definitions of SAE. There is no such thing as Level 5 within the NHTSA levels of automation.

The second error you made is assuming what Google is aiming defines level 5 on the SAE's levels of self-driving, but it doesn't. Can you imagine Star Trek without the ability of the crew to manually override the controls?
Errors?
 
I hope Elon and Nvidia don't get a divorce like Elon did with MobilEye. That would totally change the hardware again.

They are very different things.

With the MobileEye they were able to leverage an existing system that was proven out by MobileEye for various functions especially concerning active safety like FCW, etc. Before MobileEye even landed in AP 1.0 there were a fair number of Model S customers who bought retail MobileEye systems to put in their cars to do things like FCW, and pedestrian detection. With MobileEye's system Tesla didn't have to independently test/validate over millions of miles all the various elements of the MobileEye system.

With NVidia it's a very different route. Where it's completely in-house Tesla software. NVidia does provide a base to build on like libraries to accelerate DNN's, etc. They have their Visionworks software for image processing. But, ultimately it's up to Tesla to create the software to run on the NVidia hardware.

NVidia's reputation isn't going to be impacted like Mobile Eye's reputation was. If something happens all NVidia has to do is make it clear that they're not responsible for the software Tesla runs on it. Even if NVidia created examples that Tesla just re-used. It's up to Tesla to validate all the code, and to take 100% responsibility for it.

The only possibility for NVidia to be harmed is if the hardware itself fails. So while that can happen it likely won't. It did happen though back in 2007/2008 though when NVidia sold a bunch of graphic chips to notebook makers that failed. If that happened again it could end up in a nasty divorce, but since it's custom software it can be retooled to work on a different CPU/GPU offering from another company.

With the Tesla Vision software it enabled Tesla to be less tied to the hardware maker. They also hired a bunch of IC people so they could be developing their own chip.

The other cool thing about AP 2.0 is it's at least plausible the CPU/GPU could be upgraded. The primary reason we can't upgrade AP 1.0 cars is all the extra cameras, and the improved ultrasonic sensors.
 
Last edited:
  • Like
Reactions: M0DEL³ and JeffK
They are very different things.

With the MobileEye they were able to leverage an existing system that was proven out by MobileEye for various functions especially concerning active safety like FCW, etc. Before MobileEye even landed in AP 1.0 there were a fair number of Model S customers who bought retail MobileEye systems to put in their cars to do things like FCW, and pedestrian detection. With MobileEye's system Tesla didn't have to independently test/validate over millions of miles all the various elements of the MobileEye system.

With NVidia it's a very different route. Where it's completely in-house Tesla software. NVidia does provide a base to build on like libraries to accelerate DNN's, etc. They have their Visionworks software for image processing. But, ultimately it's up to Tesla to create the software to run on the NVidia hardware.

NVidia's reputation isn't going to be impacted like Mobile Eye's reputation was. If something happens all NVidia has to do is make it clear that they're not responsible for the software Tesla runs on it. Even if NVidia created examples that Tesla just re-used. It's up to Tesla to validate all the code, and to take 100% responsibility for it.

The only possibility for NVidia to be harmed is if the hardware itself fails. So while that can happen it likely won't. It did happen though back in 2007/2008 though when NVidia sold a bunch of graphic chips to notebook makers that failed. If that happened again it could end up in a nasty divorce, but since it's custom software it can be retooled to work on a different CPU/GPU offering from another company.

With the Tesla Vision software it enabled Tesla to be less tied to the hardware maker. They also hired a bunch of IC people so they could be developing their own chip.

The other cool thing about AP 2.0 is it's at least plausible the CPU/GPU could be upgraded. The primary reason we can't upgrade AP 1.0 cars is all the extra cameras, and the improved ultrasonic sensors.
To add to this response I'd mention that Tesla did try their software with AMD, Intel, and Nvidia. They chose to go with Nvidia because they felt the hardware was better.

Make no mistake, the Tesla Vision software can run on other platforms.
 
  • Like
Reactions: S4WRXTTCS
Wow, that house is expensive. They could use my house if they want to for their next video.

I would only charge them...lets see......how much is AP 2.0 again?
That house is also conditionally Sold. Over $1,000/sq.ft! Shows the value of land in that area is very high as the construction costs are relatively similar in most areas of the country. Gives credence to the old Mark Twain (?) quote, "Buy land, they aren't making any more of it." (I may be misquoting for which I apologize)
 
If you have nothing else to do....Here is an hour video of Nvidia's CEO ( and company ) explaining their new autonomous super computer. I know they promote Volvo at the end as the first collaboration partner, however there is a Tesla version of the hardware mentioned.




Here is Elon talking to the Nvidia CEO