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.
#7,381
daktari said:
I think
I understand the levels, but you are trying a semantic trick "within seconds" to discredit the value of level 3.

It is obvious that one need several/many seconds to hand over driving task safely if the driver is not intended to pay attention.
The car manufacturer will of course never deploy a system where the system, in the transition phase, let the car crash because of a principle. Such a system is worthless (here we might agree?) and will put huge liabilities on the car manufacturer.

The only way to interpret L3 is that the system safely stop the car if the driver doesn't respond within a reasonable interval. That interval is certainly closer to 10 seconds than "within seconds". The stopping procedure will probably then take another 10-20 seconds.
Click to expand...
It's not a semantics trick, that literally is what the standard says!
This was quoted by @diplomat33
"At Level 3, an ADS is capable of continuing to perform the DDT for at least several seconds after providing the fallback-ready user with a request to intervene. (Note 3, page 11)"

The standard does not guarantee the car can stop safely if the driver does not respond within the allotted time (that is why it's not L4!) After that allowed time, the manufacturer also is no longer liable for anything that happens, given it would have deactivated.

I'm simply saying sleeping is not compatible with L3, given "several seconds" is not necessarily enough time to get someone ready to take over a car after being asleep. You can ask Honda and Mercedes if it is and I bet they will respond the same that it is not. I linked Honda's warning already.

“At Level 3 of automation, the system monitors the driving environment surrounding the vehicle and takes over driving operations under certain conditions. When any of operable driving environment conditions become unsatisfactory, the system will issue a warning, and the driver must take over the driving immediately.”
Honda Global | November 11, 2020 Honda Receives Type Designation for Level 3 Automated Driving in Japan

Here's Mercedes L3 system:
"However, the driver must always be able to take over control of the vehicle. Meaning that the driver, for example, is not allowed to sleep, continuously face backwards, or leave the driver’s seat."

Conditionally automated driving with the DRIVE PILOT | Mercedes-Benz Group

If you can link to a L3 system that allows the driver to sleep, please do so, but it's obvious to me L3 is not compatible with sleeping.
 
#7,382
Nvidia is offering HD maps of over 300,000 miles of roadway in North America, Europe and Asia by 2024:

Nvidia has launched a new mapping platform that will provide the autonomous vehicle industry with ground truth mapping coverage of over 300,000 miles of roadway in North America, Europe and Asia by 2024, founder and CEO Jensen Huang said at the company’s GTC event on Tuesday.

The platform, dubbed Drive Map, is geared toward enabling high levels of autonomous driving. Drive Map isn’t only open to existing Nvidia customers, but it does augment the company’s existing solutions for the AV industry.
Click to expand...

techcrunch.com

Nvidia launched a mapping product for the AVs

Drive Map is geared toward enabling high levels of autonomous driving.
techcrunch.com techcrunch.com
 
Last edited:
  • Informative
Reactions: EVNow
#7,384
EVNow said:
They claim 1 cm accuracy vs 10 inch (?) of MobilEye REM.

So, how does this work ? How much data do cars with NVidia upload ?
Click to expand...

The article does not say how data is uploaded but it does give us some info on how the maps work:

Basically, it works by combining survey mapping with crowdsourced data from vehicles with the Nvidia Hyperion architecture. The mapping uses 3 layers: camera, lidar and radar. The crowdsourced data from Nvidia customers is uploaded constantly to the cloud as the vehicles drive. Nvidia's Omniverse uses automated content generaion tools to build the maps.

The tool provides centimeter-level accuracy by combining DeepMap’s accurate survey mapping with anonymous mapping data that’s been crowdsourced from all the vehicles that use Nvidia’s Hyperion architecture. The mapping tool features three localization layers — camera, lidar and radar — to provide the redundancy needed for autonomy.

All of the data pulled from Nvidia customers is being constantly uploaded to the cloud as vehicles drive. It’s then aggregated and loaded onto Nvidia’s Omniverse, the company’s open platform built for virtual collaboration and real-time physically accurate simulation, and used to update the map so vehicles can achieve proper localization. In the process, Nvidia is able to more quickly scale its mapping footprint.

In addition, Omniverse uses automated content generation tools to build a detailed map, which is then converted into a drivable simulation environment that can be used with Nvidia Drive Sim, an end-to-end simulation platform for autonomous vehicles.
Click to expand...
 
#7,385
diplomat33 said:
The article does not say how data is uploaded but it does give us some info on how the maps work:

Basically, it works by combining survey mapping with crowdsourced data from vehicles with the Nvidia Hyperion architecture. The mapping uses 3 layers: camera, lidar and radar. The crowdsourced data from Nvidia customers is uploaded constantly to the cloud as the vehicles drive. Nvidia's Omniverse uses automated content generaion tools to build the maps.
Click to expand...
I read the report too.

But the report “begs” the questions I posted …
 
  • Like
Reactions: DrGriz
#7,389
Daniel in SD said:
Go back and read it again. What I said is that FSD Beta is the beta of FSD and they will never release it widely. My prediction is that they will release something ore limited in functionality than FSD Beta as "autosteer on city streets." Obviously I could be wrong! But not about FSD Beta, it's the beta version of FSD.
Click to expand...
What functionality will need to be limited? They can release a system that does everything FSD Beta is doing now, and as long as it requires a driver to monitor it, it'll always be considered a L2 system.
Wide release would simply be removing the safety score limitations. I don't see them going back and adding new limitations on functionality to it before doing so.
Daniel in SD said:
It seems like you and @EVNow have some insider info on the new Master Plan...
Though even if Elon tweeted out that FSD Beta is the beta version of FSD I don't think you two would believe it. :p
Click to expand...
I was beat to it, but this is not insider info, this is based on what they said to CA DMV and is all public. What they say to CA DMV is all vetted by lawyers, so I trust those statements are made much more carefully than any marketing statements that may have more embellishment or may use looser definitions of the various levels.
 
#7,391
EVNow said:
Its always compressed before uploading - do you know how large "raw" video is ! A highly compressed 4k BluRay of 2 hours is over 60GB so - about 500MB per camera per minute.
Click to expand...

Like I said, I imagine it is probably a lot of data. I see several possibilities: either they found a way to compress it even further, maybe the cars have 5G that allows pretty fast upload speeds or the cars only send abstract information and not raw camera, lidar and radar data. What I mean is, maybe the car's cameras detect say lane lines and then convert the lane lines into vectors and just send the vectors to build the lane lines on the map. That would drastically reduce the amount of data.

Mobileye says that they upload small data packets of only 10kb/km. So Mobileye only needs very small data to build their REM maps. Perhaps, Nvidia is able to do something similar?

Again, I don't have all the answers. But I am sure Nvidia engineers have figured it out since their mapping does work.
 
#7,392
diplomat33 said:
either they found a way to compress it even further,
Click to expand...
If so - we'd hear a lot about it. Video compression is big business.

Mobileye says that they upload small data packets of only 10kb/km. So Mobileye only needs very small data to build their REM maps. Perhaps, Nvidia is able to do something similar?
Click to expand...
Right - thats why the question of how come NVidia claims 1 cm accuracy and REM is like 10 inches.
 
#7,393
EVNow said:
If so - we'd hear a lot about it. Video compression is big business.


Right - thats why the question of how come NVidia claims 1 cm accuracy and REM is like 10 inches.
Click to expand...
Isn't it 10cm (around 4 inches) for REM? Where did you get 10 inch? From what I can find, every new Hyperion 8 vehicle will have 12 cameras, 9 radar, and 1 lidar, so that likely brings the lowest common denominator higher in terms of accuracy. I don't believe the hardware requirements for REM is as high.
techcrunch.com

Nvidia's next-generation self-driving toolkit is available for 2024 vehicle models | TechCrunch

Automakers, suppliers and startups aiming to roll out automated driving features, robotaxis and self-driving trucks are a little closer to accessing
techcrunch.com techcrunch.com
 
#7,394
Daniel in SD said:
Though even if Elon tweeted out that FSD Beta is the beta version of FSD I don't think you two would believe it. :p
Click to expand...
I would believe he said it, but it would not be true. A beta is a near-release version of a product. Primary debugging is done, it's almost good enough for production use, though some exposure to and feedback from customers is needed. It happens when internal QA is not finding enough bugs to keep the team busy. It is not planned to do major revisions at this point, ideally just minor bugfixes.

Now, over time, people have played fast and loose with what a Beta is, but never has it referred to what Tesla FSD is, which is an early prototype, unable to perform the production function, with many complete rewrites of sections of the system planned, in which bugs appear so frequently that the dev team couldn't possibly keep up even with internal QA and alpha testing.
 
#7,395
Bladerskb said:
That's not SF, that's majority mountain view. You do realize that CA isn't SF right?
Local reports before Waymo large scale testing was week(s) without seeing a waymo car. After their large scale testing began (late 2020 and fullly realized early 2021) there are reports of people seeing a waymo car every few mins.

TLDR: Before Waymo's large scale testing in SF, spotting a waymo in SF was a rarity, after...spotting a Waymo was common every few minutes.

https://twitter.com/x/status/989909879975378944

https://twitter.com/x/status/1316817655525859328

https://twitter.com/x/status/1491803705133322258


"To give you a sense of how much they have been testing in SF of late, it is now unusual if on a 30-minute bike ride I do not pass at least two
@Waymo vans. They are everywhere."

"Are there actually way mo WayMo cars on the streets since SF has been a ghost town, or am I just noticing them more? I swear I see one go by my place 3-4 times a day."


"It used to be that every bike ride through SF would include seeing 1-2 Cruise self-driving cars on the road. Now it's 1-2 Waymo minivans. Lots of testing citywide."


“There are some days where it can be up to 50. It’s literally every five minutes. And we’re all working from home, so this is what we hear,” King noted.

sanfrancisco.cbslocal.com

Dead-End SF Street Plagued With Confused Waymo Cars Trying To Turn Around 'Every 5 Minutes'

A normally quiet neighborhood in San Francisco is buzzing about a sudden explosion of traffic. Neighbors say their Richmond District dead-end street has suddenly become crowded with Waymo vehicles.
sanfrancisco.cbslocal.com sanfrancisco.cbslocal.com
Click to expand...
Lol you just proved my point to @diplomat33. People were spotting Waymo minivans and vans in SF in scale in 2020 (just like I did personally). My point is the ramp-up in SF did not start up with the I-Paces in 2021, so leaving out the Pacificas is misleading. And they were testing in SF already even before then with the Pacificas in smaller scale (which should not be left out of the timeline).

PS: Mountain View is still in the Bay Area, read what I wrote. There is no evidence that they have done much of their testing in CA out of the Bay Area, so a bulk of their tests if not almost all of it is likely inside the Bay Area.
 
Last edited:
#7,397
bradtem said:
I would believe he said it, but it would not be true. A beta is a near-release version of a product. Primary debugging is done, it's almost good enough for production use, though some exposure to and feedback from customers is needed. It happens when internal QA is not finding enough bugs to keep the team busy. It is not planned to do major revisions at this point, ideally just minor bugfixes.

Now, over time, people have played fast and loose with what a Beta is, but never has it referred to what Tesla FSD is, which is an early prototype, unable to perform the production function, with many complete rewrites of sections of the system planned, in which bugs appear so frequently that the dev team couldn't possibly keep up even with internal QA and alpha testing.
Click to expand...
I guess I should update my statement, FSD Beta is a prototype of FSD.
stopcrazypp said:
What functionality will need to be limited? They can release a system that does everything FSD Beta is doing now, and as long as it requires a driver to monitor it, it'll always be considered a L2 system.
Wide release would simply be removing the safety score limitations. I don't see them going back and adding new limitations on functionality to it before doing so.
Click to expand...
All AV prototypes require a driver to monitor them...
Why didn't they release FSD Beta to everyone a year and a half ago?
Irrespective of FSD I'm curious if you and @EVNow think autonomous vehicle testing is potentially dangerous and if so, why?

stopcrazypp said:
I was beat to it, but this is not insider info, this is based on what they said to CA DMV and is all public. What they say to CA DMV is all vetted by lawyers, so I trust those statements are made much more carefully than any marketing statements that may have more embellishment or may use looser definitions of the various levels.
Click to expand...
I think the lawyers were hired to represent their client and the client told them to come up with an argument as to why they shouldn't subject to autonomous vehicle testing regulations.
 
  • Like
Reactions: Doggydogworld
#7,398
Daniel in SD said:
Irrespective of FSD I'm curious if you and @EVNow think autonomous vehicle testing is potentially dangerous and if so, why?
Click to expand...
It’s the other way round.

It is dangerous not to test AVs.

I firmly believe AVs of any “level” should be tested for many million miles in real world with attentive drivers and error rate determined statistically before unleashing them on unsuspecting public. 5 minutes of testing isn’t enough. Just simulation isn’t enough.
 
#7,399
Daniel in SD said:
I guess I should update my statement, FSD Beta is a prototype of FSD.

All AV prototypes require a driver to monitor them...
Click to expand...
No they don't, Waymo has tested without a driver for a while in AZ, and they are doing so also in California.
Daniel in SD said:
Why didn't they release FSD Beta to everyone a year and a half ago?
Click to expand...
Did you not see FSD Beta a year and a half ago? It fell flat on its face for even relatively simple traffic maneuvers. And although it has improved a lot since then, even today it can't reliably make the certain maneuvers (like the unprotected left that has continually been tested).

You also dodged the question. What functionality do you foresee Tesla removing from FSD Beta if they launch "Autosteer from City Streets" without the safety score requirement? If they don't have to remove any functionality, why is it out of possibility to just release FSD Beta as a L2 feature?
Daniel in SD said:
Irrespective of FSD I'm curious if you and @EVNow think autonomous vehicle testing is potentially dangerous and if so, why?
Click to expand...
Just driving in general is potentially dangerous, I'm not sure what point you are trying to make here?
Daniel in SD said:
I think the lawyers were hired to represent their client and the client told them to come up with an argument as to why they shouldn't subject to autonomous vehicle testing regulations.
Click to expand...
So you think Tesla is lying about working on a door-to-door L2 feature? Or is it because you believe that there is no such thing as door-to-door L2?

Note Tesla is not the only one working on this, here's GM's effort:
GM Ultra Cruise Reveals The Brain Of Its Hands-Free Door-To-Door Driving Tech
Here's Geely/Zeeker's based on Mobileye's Supervision:
https://www.eetimes.com/chinas-geely-to-deploy-mobileyes-hands-free-adas/

Are you suggesting those be treated as L3+ systems in testing, even though they are clearly L2?
 
#7,400
  • Like
Reactions: Terminator857 and diplomat33
You must log in or register to reply here.

Similar threads

Watts_Up
I almost get run over by an autonomous car when crossing a street.
Replies
10
Views
857
AI, Autopilot, & Autonomous/FSD
JulienW
JulienW
diplomat33
  • Article
Glossary for AI, Machine Learning and Autonomous Driving
Replies
4
Views
2K
AI, Autopilot, & Autonomous/FSD
Larry Hutchinson
L
EVer Hopeful
Your Full Self-Driving (Supervised) Trial starts now!
Replies
37
Views
2K
AI, Autopilot, & Autonomous/FSD
GData
G
A
Revisiting the Indicator light needed for Autonomous driving! We need an external light!
Replies
41
Views
2K
AI, Autopilot, & Autonomous/FSD
Dan D.
D
A
separate commercial and personal use FSD pricing?
Replies
6
Views
1K
AI, Autopilot, & Autonomous/FSD
tevman
tevman
Top
Back
Blog
New
Forum list
Marketplace
Menu