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

Will HW3 REALLY deliver FSD ? some questions

This site may earn commission on affiliate links.

Tes La Ferrari

Active Member
Jun 1, 2018
1,343
10,208
Canada
Let me start by saying I hope the answer is yes - and I'm rooting for Tesla's success.

I have paid for FSD already, in that spirit.

I have been reading through a lot of posts / threads on AP3 computer and FSD, and it brought some questions to mind.

1. I seem to recall hearing somewhere (can't recall where that they are working on HW4. Why would they be upgrading people to HW3 if HW 4 is in the pipeline, and FSD isn't working yet anyway? Would this not result in double the costs to swap HW2 & 2.5 to 3 then swap all the HW3 to HW4 ?

2. Why are they working on a new chip / hardware HW4 if HW3 is plenty for FSD ? / Why not focus those resources on the AI / solving vision side for FSD to ensure maximum "bang for the buck "?

Thoughts from the TMC intelligencia ?
 
  • Funny
Reactions: cwerdna
They talked about HW4 in the autonomy day presentation I believe. Tesla is always looking forward and changes thing frequently, so it isn’t too surprising they are already developing HW4.

HW2 then HW 2.5 was supposed to be enough for FSD when I bought my car, so I am not surprised there will be a HW4. HW4 might also come with a new sensor suite like AP1 to AP2, who knows.


As far as FSD upgrades go, I believe HW3 will be it for the current gen. Tesla won’t upgrade us to HW 4, that will require a new car purchase I am willing to bet. I believed they noted in the presentation that HW3 development was somewhat constrained by the need to make it retrofittable. That implies HW 4 will not be constrained in the same way.
 
They talked about HW4 in the autonomy day presentation I believe. Tesla is always looking forward and changes thing frequently, so it isn’t too surprising they are already developing HW4.

HW2 then HW 2.5 was supposed to be enough for FSD when I bought my car, so I am not surprised there will be a HW4. HW4 might also come with a new sensor suite like AP1 to AP2, who knows.


As far as FSD upgrades go, I believe HW3 will be it for the current gen. Tesla won’t upgrade us to HW 4, that will require a new car purchase I am willing to bet. I believed they noted in the presentation that HW3 development was somewhat constrained by the need to make it retrofittable. That implies HW 4 will not be constrained in the same way.

But if HW3 is fully capable of FSD and the Robotaxi network, what added value would HW4 have?
 
It depends on what level of autonomous driving you mean by FSD... And Tesla never committed on this.
If FSD is just the ability to drive on any road and street while you keep your hands on the wheel, and stay responsible for everything your car does, then HW3 may do it in a short future as it does not need to be reliable. The current FSD option doesn’t sell anything more, so Tesla does not not have to deliver more than this.
If FSD means you can take a nap or even not be on the driver seat (level 4+), feature it did not sell to anyone, then it’s pure speculation whether HW3+ can do it. Nobody can tell if it’s possible without a LIDAR currently.
 
  • Like
Reactions: MrKool
...HW 4 is in the pipeline...

It's still being developed.

They could withhold HW2.0, HW2.5, HW3.0 and wait for HW4 or HW5 too!

But I still need something while waiting!

It's the same thing with cell phone camera. The first-generation cell phone camera picture was terrible. They could skip the camera and wait until it is more decent to 108 MegaPixels today. But I still need something to take a picture for all these years.
 
L5 FSD (Full driverless autonomy) is currently still a pipe dream IMO. Maybe two decades from now.

I realized this while sitting third in queue behind a night-time wreck with emergency responders having to direct traffic around the scene, taking a highly unusual pattern that involved turning 90 degrees to the normal traffic flow, crossing the centerline and briefly driving on the wrong side of the road before clearing the scene and returning to normal positioning on the road, traffic speeds, etc.

Now, if instead of me being the loose nut behind the wheel and third in queue, imagine an L5 car and it being first in queue with no one to follow and having to take the same actions - then add having to negotiate a debris field from the wreck. Can’t get a flat tire from busted headlight bits because your six-year-old daughter in the back seat (the car was driving her home from school) isn’t able to change a tire even if the car were so equipped.

Extreme case, sure. But quite easy for a human driver to do, not so easy for an algorithm due to all the “rule breaking” required to get through the scene.

Then there is the conundrum of what to do when sitting and waiting. Preserve battery and let the cabin get cold, or keep the child warm and risk battery depletion and having to divert to a charger (which maybe the passenger can manage to get plugged in).

This all has to tested and proven absolutely reliable before L5/“robotaxi” can happen.

So it all depends on what Tesla defines as FSD.
 
Let me start by saying I hope the answer is yes - and I'm rooting for Tesla's success.

I have paid for FSD already, in that spirit.

I have been reading through a lot of posts / threads on AP3 computer and FSD, and it brought some questions to mind.

1. I seem to recall hearing somewhere (can't recall where that they are working on HW4. Why would they be upgrading people to HW3 if HW 4 is in the pipeline, and FSD isn't working yet anyway? Would this not result in double the costs to swap HW2 & 2.5 to 3 then swap all the HW3 to HW4 ?

2. Why are they working on a new chip / hardware HW4 if HW3 is plenty for FSD ? / Why not focus those resources on the AI / solving vision side for FSD to ensure maximum "bang for the buck "?

Thoughts from the TMC intelligencia ?
The answer to your question is which FSD you paid for. If it was the pre-2019 version, not likely you're going to get it, even less if you have MCU1. If it's the post some time in 2019 when Tesla redefined FSD to be what it is today, which consists mostly of features which used to be part of EAP, then probably, at some point in the future you will get it (also, if you have MCU1, less likely or at least longer delay).
 
  • Like
Reactions: OPRCE
IMHO FSD is not going to be possible until ALL of the cars on the road can talk directly to each other via multiple radio channels. It has to be direct, not through a server 3000 miles away. I do not expect that to happen in my lifetime because it means most of the existing cars will have to have been scrapped.

A more common problem to be solved is 4 cars arriving simultaneously from 4 directions at a 4-way stop sign.
The cars will vote to decide which one goes first.
 
I know I've been saying it for a while now, but we will not see FSD on our cars. It's a hardware problem. It's a software problem. But mostly, it's an infrastructure problem. It's not something that can be "fixed" with over-the-air updates or faster image processing chips.

The idea that we'll see FSD for "city streets" is... well... I'm sure they'll get something that can drive down a perfectly marked, wide and clear American Main Street. Anywhere else in the world? Not so much.
 
IMHO FSD is not going to be possible until ALL of the cars on the road can talk directly to each other via multiple radio channels. It has to be direct, not through a server 3000 miles away. I do not expect that to happen in my lifetime because it means most of the existing cars will have to have been scrapped.

A more common problem to be solved is 4 cars arriving simultaneously from 4 directions at a 4-way stop sign.
The cars will vote to decide which one goes first.

Now I have an image in my head of the cars forming a cluster and how to determine a quorum.
 
But if HW3 is fully capable of FSD and the Robotaxi network, what added value would HW4 have?

In the unlikely event that HW3 is capable of driving the car as a robotaxi, I would imagine HW4 may provide better performance for said robotaxi - maybe less hesitation in traffic, etc. Maybe it will be able to Smart Summon at 10mph vs 5, stuff like that. Probably akin to the AP performance differences between AP2 and AP2.5. Both have the same capabilities, but I have seen it reported by owners with both that AP2.5 does better overall.
 
IMHO FSD is not going to be possible until ALL of the cars on the road can talk directly to each other via multiple radio channels. It has to be direct, not through a server 3000 miles away. I do not expect that to happen in my lifetime because it means most of the existing cars will have to have been scrapped.

there is a standard for that, known as v2x (vehicle to 'anything', ie, 'v-2-v' as the most obvious use-case).

Vehicle-to-everything - Wikipedia

I've been told that there are thousands of messages being seen at the radio, per second. this may need a whole cpu subsystem to process and filter these out. I'm hopeful that this will add value, over time, but its going to be a 'learning experience' while we all implement and debug this..
 
Now I have an image in my head of the cars forming a cluster and how to determine a quorum.

speaking of quorum, I'm hoping that we see more redundancy in the internal designs of the new cars. ASIL-D will require fault tolerance for many things, and the new breed of car cpu chip does have a multi lock-step core architecture. but I have not seen a 3 way 'vote' before in automotive; I would like to see it, in fact, but they always reject it due to COST.
 
  • Like
Reactions: Banksy888
I'm sure they'll get something that can drive down a perfectly marked, wide and clear American Main Street. Anywhere else in the world? Not so much

While I certainly don't contest the fact that roads here are well marked and I have no idea what roads in most other countries look like, there are still a good number of roads here that look like the lane markings were painted on by a blind guy who was drinking before his shift :)

I'd be curious to know how it would handle an unmarked dirt road in a rural area.
 
It depends on what level of autonomous driving you mean by FSD... And Tesla never committed on this.
If FSD is just the ability to drive on any road and street while you keep your hands on the wheel, and stay responsible for everything your car does, then HW3 may do it in a short future as it does not need to be reliable. The current FSD option doesn’t sell anything more, so Tesla does not not have to deliver more than this.
If FSD means you can take a nap or even not be on the driver seat (level 4+), feature it did not sell to anyone, then it’s pure speculation whether HW3+ can do it. Nobody can tell if it’s possible without a LIDAR currently.
There are different levels to Full Self Driving? Can you clarify which levels those are?
 
btw, has tesla ever said anything about their plans (or non plans) re: v2x?

(maybe they have the radio hardware already there and just not enabled?) don't know. anyone know what their stance is on v2x?

They want to solve self driving without any crutches such as that. Also you would have to wait for the entire infrastructure to be updated.
 
you don't have to wait; once cars start 'advertising' (pinging each other), it will be an evolution over time. no one expects it to just be enabled one day, all of a sudden. its a work in progress, clearly, and will take lots of refining before its 'done'.

but the longer you avoid getting into it, the more behind you become (as a company).

I would hope that there's a skunkworks project that is at least looking into v2x for the next gen tesla architecture.
 
Will HW3 REALLY deliver FSD ?

Yes.

But to be fair "Full" will likely only be in certain circumstances and conditions, initially well marked highways in good weather for example, then progessively extending. In more problematic scenarios the driver will still be required to take over.

More importantly Tesla strategy might currently only be contesting "best driver assistance" status, but other companies technologies will simply run into a wall and only ever be able to get to 99%; Tesla strategy fleet based deep learning will ultimately get them to 99.999% which is what is truly required to get to FSD and at that point Tesla will simply extend its lead in a way that will just leave the rest floundering.

Quite conceivably Tesla will then start offering FSD to other auto-manufacturers to the point of becoming maybe one of two default choices for FSD solutions from which the revenue stream could be simply astounding. You only have to look at how much Intel paid for Mobileye ( iro $22bn iirc) to realise just how much this technology is worth.