Elon: "Feature complete for full self driving this year"

[S4WRXTTCS]

So I would be very surprised if Tesla suddenly made it an EAP feature.

It can't be made an EAP feature because traffic light detection (as it currently stands) requires HW3.

I don't believe any EAP only vehicle has HW3.

 

[banned-66611]

My understanding is that the information from V2I is via localised beacons (I believe 5G was the most likely candidate) and that global standardisations and security is/was (several years since I got any insider knowledge) a slow burner. I suspect part of it was initially down to coming up with a solution to what was not yet a widespread problem that needed solving. I do know that even 3 or so years ago, our traffic light controllers would have been pretty much 'feature complete', but there is a long way between that and a commercial partner taking that feature and building it into a real validated product that met global standards and the vehicle manufacturers then making use of that feature in their vehicles.

Even if the technology was ready today it would require every traffic light in the country to be replaced/upgraded to be of any use for self driving. Until that happens the car will have to be able to recognize lights as well as a human can by sight alone.

 

[banned-66611]

It can't be made an EAP feature because traffic light detection (as it currently stands) requires HW3.

I don't believe any EAP only vehicle has HW3.

What about ones that are owned an upgrade because they bought FSD?

At some point people are going to want that upgrade even before FSD is available so they can start getting some of the features they bought, especially since FSD is likely never going to be available for their vehicles. Tesla already lost one lawsuit over delayed features, upgrading them now and offering them a few tokens like traffic light detection might reduce their liabilities a little bit.

I guess it's a cost/benefit analysis, cost of upgrade to HW3 which will eventually need to be ripped out again for HW4/5/6 or just pay out the inevitable class action.

 

[Knightshade]

What about ones that are owned an upgrade because they bought FSD?

At some point people are going to want that upgrade even before FSD is available so they can start getting some of the features they bought, especially since FSD is likely never going to be available for their vehicles.

....not sure I understand your question?

First- for an EAP owner who bought FSD, there are no features to get right now (other than added visualizations, so nothing functional).... though the very first more-than-EAP feature of FSD is in early access right now, so that will be changing soonish for those in the US at least (still gonna be a while for non-US owners).... which brings us to-

Second- They've been doing HW3 upgrades for pre-HW3 model 3s for months now. Just make an appointment if you want it now.

 

[diplomat33]

Green is reporting that EAP will get FSD visualizations and the auto lane change to evade cones.

 

[diplomat33]

It seems the feature is mostly for data gathering, not really ready for public use yet as there are some quirks and risk of getting read ended (I was right!). I'm guessing the actual feature that goes wide will be different.

 

[willow_hiller]

It seems the feature is mostly for data gathering, not really ready for public use yet as there are some quirks and risk of getting read ended (I was right!). I'm guessing the actual feature that goes wide will be different.

Could be a good opportunity for Tesla to demonstrate how quickly they can train models. Elon is overly optimistic about everything, but it seems like he's saying they can go from a data collection branch on EAP to world-wide release in 2 quarters.

 

[Knightshade]

Green is reporting that EAP will get FSD visualizations and the auto lane change to evade cones.

I read it as they're getting JUST the cone visualizations... not the other stuff? (which makes sense, NoA is part of EAP and you'd want it to be able to deal with highway construction cones)

 

[roblab]

I like seeing the cones go by! My car sees them better than I do, and I'm usually an aware driver.

 

[S4WRXTTCS]

I like seeing the cones go by! My car sees them better than I do, and I'm usually an aware driver.

I wish mine didn't recognize EVERYTHING orange as a cone.

 

[mongo]

It seems the feature is mostly for data gathering, not really ready for public use yet as there are some quirks and risk of getting read ended (I was right!). I'm guessing the actual feature that goes wide will be different.

So it is not gating on maps then?

 

[banned-66611]

....not sure I understand your question?

If I had paid $5000 for FSD four years ago I'd want every little thing they can offer, even if it is only visualizations. Definitely get that hardware upgrade, the MCU is probably close to death from worn out flash memory anyway by now.

 

[diplomat33]

So it is not gating on maps then?

Apparently not. But green says that mapped lights give the system more advance notice. So it seems that if a traffic light is mapped, the car will respond sooner but if the traffic light is not mapped, the car will still respond just maybe a bit later when the vision sees the light. That's how I interpret what green tweeted.

 

[mongo]

Apparently not. But green says that mapped lights give the system more advance notice. So it seems that if a traffic light is mapped, the car will respond sooner but if the traffic light is not mapped, the car will still respond just maybe a bit later when the vision sees the light. That's how I interpret what green tweeted.

Yeah, that was my read also. If there is a mapped (expected) intersection, it will alart the driver to the upcoming decision ahead of time (before decelerating). However, if it detects a (high confidence?) traffic control device, it will alert when it does (since there is no way to see it earlier). Which may cause deceleration simultaneous with the warning.

 

[Knightshade]

If I had paid $5000 for FSD four years ago I'd want every little thing they can offer, even if it is only visualizations. Definitely get that hardware upgrade, the MCU is probably close to death from worn out flash memory anyway by now.

FWIW if purchased with the car it was $3000 4 years ago (and even 1.2 years ago- it only went way up Feb/March 2019 when EAP went away)

Also upgrading the driving computer doesn't upgrade the MCU (though MCU1 owners have the option to do that at additional cost at the same time)

 

[paaltio]

Green is reporting that EAP will get FSD visualizations and the auto lane change to evade cones.

That might still be true, but I assume he's talking about Early Access Program here, not Enhanced Autopilot, as in his other tweets he's talking about the new stopping feature.

 

[diplomat33]

That might still be true, but I assume he's talking about Early Access Program here, not Enhanced Autopilot, as in his other tweets he's talking about the new stopping feature.

Yes, he is talking about Early Access but this tweet is not talking about stopping for red lights. In the tweet, he specially mentions visualizations to see cones and auto lane change to see cones. He says this is coming to HW2.x. So he is saying that seeing cones and auto lane change for cones, which up to now was HW3 only, is now coming to HW2 and HW2.5.

 

[nepenthe]

Apparently not. But green says that mapped lights give the system more advance notice. So it seems that if a traffic light is mapped, the car will respond sooner but if the traffic light is not mapped, the car will still respond just maybe a bit later when the vision sees the light. That's how I interpret what green tweeted.

So... about those HD maps. Can we continue discussions from the agreed upon understanding, if you need to operate on the assumption map data could be inaccurate/out-of-date you can functionally operate without them.

Waymo and Cruize are not consumer vehicles, while I can appreciate the comparisons on a technical curiosity level, any system that reduces the rates of accidents, even if just while on the forward path to "FSD" claims, is in fact good enough.

 

[diplomat33]

So... about those HD maps. Can we continue discussions from the agreed upon understanding, if you need to operate on the assumption map data could be inaccurate/out-of-date you can functionally operate without them.

I am not sure we can say that. The whole point of HD maps is to provide the car with information that it cannot get from cameras. So even if your camera vision is so accurate and reliable that you don't need HD maps for redundancy, you will still need HD maps to get information that you physically can't get with cameras or that is difficult to get from cameras:

"A HD map is therefore necessary as a reliable off-board sensor containing processed a-priori information to "detect" features that are not easily detectable by on-board sensors or to provide a redundant source of information for on-board sensors..."

Source: "Safety First for Automated Driving"

There is also the issue of how reliable your HD maps are versus how reliable your camera vision. Obviously, if your HD maps suck and your camer vision is great, then you won't rely on your HD maps. But if your HD maps are accurate and can be updated easily and efficiently, then it would be silly to toss them, just because you also have good camera vision.

Personally, I think the argument that HD maps will get outdated too quickly to be useful to be a false argument. Most of the a-priori information that you can't get from cameras won't change very often. How often do they change a one way street into a two way street for example? Not very often. And if say road construction happens, then you can update your maps for that area. In my town, roads get worked on every few years. New traffic lights or new stop sign locations don't get added that often. It should not be too hard to update HD maps for a town every 2-3 years for example.

Also, remember that you can update HD maps with any car in your fleet driving that area. So every car in your fleet can be continuously updating the HD maps for the fleet. Toyota also has a new method of updating HD maps via commercial satellites. Nvidia has end-to-end creation and updating of HD maps. So keeping HD maps updated is not as big a problem as people seem to think, IMO.

But if your vision is good enough to easily update maps, then why not just use camera vision exclusively? Again, you still need HD maps for information that you can't get from cameras. And it is still helpful to have HD maps as a redundant source for the rest of the fleet that visits that area again after the first pass that updates the HD maps.

Waymo and Cruize are not consumer vehicles, while I can appreciate the comparisons on a technical curiosity level, any system that reduces the rates of accidents, even if just while on the forward path to "FSD" claims, is in fact good enough.

I agree that reducing the rate of accidents is a very good thing. But I don't think we should be happy with just slightly safer than human drivers. If we truly want millions of L5 autonomous cars that can safely drive us around, then we want the highest safety possible. Anything that can further improve safety should be looked into.

 

[ZeApelido]

Just an interesting note, I noticed stop lights in Redwood City have been updated with yellow borders like this:

Retroreflective Borders on Traffic Signal Backplates – A South Carolina Success Story - Safety | Federal Highway Administration

But some of them have an alternating black/yellow dash-like border. Could be useful in helping autonomous vehicles detecting stop lights.