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

FSD very far away due to regulations?

This site may earn commission on affiliate links.
Geofenced system is so easy. Smaller the area the easier it is. That's how everyone is using to "demo" their technology. Even a blind person could move unassisted inside his house or few blocks around the house. BTW Waymo cars has done what the car in the video did more than three year ago with a pod without a steering wheel.

I don’t get this narrative against mapped systems and demos. Geo-fencing will always be a-part of autonomous driving. The real question is, are your driving policy/path planning general enough to cascade to other environments and situations. If it is, all you have to do is add more map so you can scale quickly and you’re good.


G5xSDSY.png
 
I don’t get this narrative against mapped systems and demos. Geo-fencing will always be a-part of autonomous driving. The real question is, are your driving policy/path planning general enough to cascade to other environments and situations. If it is, all you have to do is add more map so you can scale quickly and you’re good.


G5xSDSY.png
Many places still don't have Google Streetview. I don't expect any mapping to be better,
 
This is pretty amazing for a manual effort by a total of 250 cars. Now imagine you had 1 million cars..and you could generate the maps with a push of a button in less than 24 hours.
500px-Google_Street_View_timelapse_additions.gif
For something like supercruise you'd need all the cars to have lidar in order to help with mapping.

For cars with lidar it may not be a problem... but there might be a bandwidth limitation.
 
For something like supercruise you'd need all the cars to have lidar in order to help with mapping.

For cars with lidar it may not be a problem... but there might be a bandwidth limitation.

MobilEye EyeQ4 already has vision-based mapping. Of course this requires great vision so that the map can then be made small and reliable for bandwidth etc reasons but this is already a production product.
 
Yes Tesla will put HW3 board in every new car shipped soon as supply is there. It does not save Tesla any money to continue using old chip from the third party supplier. HW3 may not be end of road for retrofits but it likely will be end of road for free retrofits.

I would agree HW3 is likely the end of road for free retrofits unless they wait for some next version before starting the retrofits at all.
 
For something like supercruise you'd need all the cars to have lidar in order to help with mapping.

For cars with lidar it may not be a problem... but there might be a bandwidth limitation.

The 25k KM Japan highway HD map was made with cameras and will be maintained with cameras. The entire map is 400 MB, way smaller than Tesla's ADAS map, yet with way more info (traffic signs, light poles, lane marks, round boundaries, drive-able trajectories, etc) and 10 cm accuracy.

Here's a look at Munich's map built from production sold cars so far (BMW X5). This density will rapidly increase when other cars with EyeQ4 are released this year (BMW 3 series, X7, Audi Etron, FCA cars, VW cars, etc).

If your driving policy is versatile and you have the map coverage then you have an opportunity to scale. We shouldn't look down on geo-fencing. Geo-fencing is done because you want to get something that works and is reliable before you scale it up.

owSpk0a.png
 
The 25k KM Japan highway HD map was made with cameras and will be maintained with cameras. The entire map is 400 MB, way smaller than Tesla's ADAS map, yet with way more info (traffic signs, light poles, lane marks, round boundaries, drive-able trajectories, etc) and 10 cm accuracy.

Here's a look at Munich's map built from production sold cars so far (BMW X5). This density will rapidly increase when other cars with EyeQ4 are released this year (BMW 3 series, X7, Audi Etron, FCA cars, VW cars, etc).

If your driving policy is versatile and you have the map coverage then you have an opportunity to scale. We shouldn't look down on geo-fencing. Geo-fencing is done because you want to get something that works and is reliable before you scale it up.

owSpk0a.png
That's the MobilEye solution... we were talking about some of the FSD demos with lidar.

Speaking with regards to limitations to pre-map and maintain a lidar based map.
 
The 25k KM Japan highway HD map was made with cameras and will be maintained with cameras. The entire map is 400 MB, way smaller than Tesla's ADAS map, yet with way more info (traffic signs, light poles, lane marks, round boundaries, drive-able trajectories, etc) and 10 cm accuracy.

How large is Tesla's ADAS map? And how do you know what Tesla's map includes?
 
  • Like
Reactions: outdoors
This is pretty amazing for a manual effort by a total of 250 cars. Now imagine you had 1 million cars..and you could generate the maps with a push of a button in less than 24 hours.
500px-Google_Street_View_timelapse_additions.gif

Seems that every route will eventually be a predetermined route. That seems the first scenario for achieving self driving cars.

If that is the case, is that FSD, is it level 5 ? How does Tesla distinguish a FSD "feature" from EAP? What is just a "FSD feature" and how is it distinguished from EAP?

Also, in the Brownlee video, the display of surrounding traffic was much richer and more detailed than what Tesla can now show. Could be for many reasons, but is it the lidar, will Tesla need lidar?
 
If that is the case, is that FSD, is it level 5 ? How does Tesla distinguish a FSD "feature" from EAP? What is just a "FSD feature" and how is it distinguished from EAP?
FSD will drive you around while EAP is glorified TACC with lane keeping assist (plus lane change and NoA).

EAP can't take you from your driveway to the grocery store. EAP can't go find itself a parking spot. etc
 
Yes Tesla will put HW3 board in every new car shipped soon as supply is there. It does not save Tesla any money to continue using old chip from the third party supplier. HW3 may not be end of road for retrofits but it likely will be end of road for free retrofits.

Nope- they've said repeatedly anybody who paid for FSD gets all HW upgrades needed for FSD. No expiration date.

That's one of the primary reasons some folks paid for FSD in the first place- to future-proof that aspect of the cars hardware.

If there does end up being more HW changes needed start looking for the post-purchase cost of FSD to become tiered based on how much HW needs to be upgraded on older cars.

That's not what I meant. I meant HW3 equipped cars will not get free retrofit when HW4, 5 6... comes out in the future.

I don't know... I feel a lawsuit if they can't provide FSD still even with HW 3 unless there's a free upgrade route.

That would be an exception although there is no reason to think Tesla will not make FSD to work with HW3. It might not work perfectly but it will work. Kind of like the first iPhone which already could surf the web and run apps. Later generations just do those same things better. That will be the way for the progress of FSD too.
 
Last edited:
Tesla will probably take HW2/2.5 with HW3 retrofits as far as it can go and the basically use the regulator caveat to explain away the rest when it will never be ”Level 5”, possibly never beyond Level 2. There may be some more lawsuits eventually but Tesla will be big enough to weather them at that stage.

When HW3 launches it would not surprise me if Tesla repackages the EAP/FSD setup in some way that does not create so much liability of past promises and all that Level 5 capable hardware talk from October 2016 slowly melts into history. It helped Tesla weather the MobilEye break-up at the time, was mostly just hot air but it did its trick.

I mean it is not like this has not played out exactly the same way before. P85D performance with Ludicrous and software updates, right? The same with P90D... AP1 promises that never came and so on (software 8.1 with navigation-based exit taking in December 2016 for AP1 for example). The history of Tesla is littered with broken promises that have kept products selling at the time and are conveniently forgotten in the future.

I think people expecting retrofits beyond HW3 will be disappointed, assuming that is the stage at which Tesla begins the HW2/HW2.5 retrofits.
 
I think people expecting retrofits beyond HW3 will be disappointed, assuming that is the stage at which Tesla begins the HW2/HW2.5 retrofits.

You may still be able to retrofit HW3 with even newer one. You will just need to pay for it. In that sense it will still be better than the iPhone analogy.

The history of Tesla is littered with broken promises that have kept products selling at the time and are conveniently forgotten in the future.

Promises (not contract obligations) are meant to be broken. Everyone does that all the time. At least Elon does not lie like many others do. Regardless of that most people only care how their cars work not how that compares to some expectations. Only few of us on the forum do.
 
Last edited:
You may still be able to retrofit HW3 with even newer one. You will just need to pay for it. In that sense it will still be better than the iPhone analogy.

That is possible. Tesla has a somewhat checkered but not non-existent history of allowing technology retrofits at cost.

That said what I also find a possibility is Tesla may at some point simply abandon the HW2/2.5 sensor suite together with HW3 hardware similar to AP1 and move on to a next level sensor suite with HW3.5/HW4 where development of new features continues. At that stage the HW2/2.5/3 system would be relegated to maintenance updates similar to AP1. The NNs may well improve and all that but new features would require the newer sensor suites.

In that scenario retrofitting a HW3.5/HW4 computer to HW2/2.5/3 cars that may be part of a wholly different system may not be possible or offer any benefit and Tesla may not go to the trouble of making it retrofittable anyway, as the new sensor suite is what is making the difference then for example for the redundancy needed to go up a SAE Level or two.
 
We are only talking about the computer board. Sensor suit certainly is not retrofitable. Upgrades will come when necessary and changes may not need to be transparent. Although as I always believed (and I think Tesla too) it's the brian not eyes that needs more work.
 
Last edited:
  • Informative
Reactions: pilotSteve
We are only talking about the computer board. Sensor suit certainly is not retrofitable. Upgrades will come when necessary and changes may not need to be transparent. Although as I always believed (and I think Tesla too) it's the brian not eyes that needs more work.

Of course but these things can go hand in hand. If sensor suite gets upgraded (and I am sure it will at some point) it only makes sense future software features might require it.

So while ”FSD” will continue to get better on HW2/2.5/3 at some point a new sensor suite might mean no more new features come to that old setup as Tesla focuses development on the new sensor suite.

The analogy to AP1 (or even AP2 already with the dashcam) naturally does not fully fit but can offer some insights. Tesla would maintain ”FSD” on HW2/2.5/3 indefinitely but at some stage new capabilities will require new hardware that is not retrofittable.

One such step could be that a move from Level 2 to Level 3-4 could require more redundancy in sensors at which stage the old suite could stagnate.