1 hour ago, MattNischan said:

And, to the other part of the question, about morphing. Yes, sometimes, it's obvious, for sure. But this is also one of those problems that's really hard to wave a magic wand at.

There are a few limitations: one is, the data may be streaming in, either from disk or from online, and it might be behind, due to disk or internet conditions. Then, the data shows up, and suddenly smooth hills morph into cliffs. The second case is that the distance is such that you're hitting a better LOD ring by distance, and same thing, you notice the switch. What are some ways to combat this?

On the I/O side, you could just decide that if the data hasn't gotten to you by the time you enter the high LOD ring, just don't swap it out until it's out of view. That would avoid the morphing, but might also be frustrating; some users would certainly complain that on the first pass of an area it looked like word not allowed and when they turned around suddenly it was good. That feels like a bug too, to those people.

It's always great to read your insightful and level-headed posts, thanks a lot for contributing! People tend to ignore the vast complexity of the task of recreating the whole Earth from different types of data. Pointing to the fact that there's a lots of great looking add-ons of places, airports, cities, they miss the point that it isn't possible to expect the same level of accuracy and detail on the entire globe. 

I think there's little you can do when the morphing is due to data still loading from disk or internet. Also the morphing when a higher LOD gets loaded is hard to avoid - with a lot of GPU resources you can push the LOD farther out, still you can see it to some extent, as can be seen in pretty much any game out there today.

There's one thing though that makes the morphing ugly and really stand out: When the data for the different LODs are derived from different sources containing height maps that deviate a lot from each other at their LODs, mostly evident in areas with the "water bodies on pillars and plateaus" effect. Noticed this a lot in parts of Norway (fixed soon) and when crossing Greenland with a C172, some of these glitches are impressive by themselves, nevertheless they truly kill all the immersion. I see some ways to address this programmatically, but none of them is trivial and requires lots of adjustments for each data source. And if you're going to update the elevation data anyway, why bother. I for one hope Greenland (as a territory of Denmark) improves with the next world update, but wouldn't expect it to work wonders. 

Being involved with data processing myself (in an entirely different field), I totally understand that there are no simple solutions to many of the issues brought up in the forums. There are some low-hanging fruits though which can probably be addressed more easily than others. 

Edit: Concerning Jayne's role in all this, I wholeheartedly agree. She's hired because of her community managing skills, not for her knowledge in flight simulation, and while some may disagree, it was the right decision. She's come a long way learning about aviation in general and flight simulation in particular, and she's trying to get to the bottom of a problem in the official forums. 

 

Edited April 14, 20215 yr by pstrub

1 hour ago, MattNischan said:

All,

Also keep in mind the environment and context. Martial and Seb are great guys, and super smart engineers. I now get to talk with them a bunch. But the format is intimidating: they aren't native camera people, they're software engineers trying to understand and convey things as best they can in a second language.

Jayne is awesome too, but occasionally things get lost in translation because it might be some super deep vertical and she's a new simmer. Her skillset is community management.

I know, when you have a pet issue, it seems absolutely brain dead obvious to you what it is and it also seems completely glaring, like how could these guys not see this stuff. But even I was a bit lost on what the coastline bug was referring to, and I'm in the sim 6 hours a day or more. Now imagine if you're trying to parse that in a second language.

However @ChaoticBeauty, to answer it now, it's actually a really challenging problem. The resolution of the satellite images is generally way, way higher than the DEM data resolution. So ultimately you end up with areas where the image of the water doesn't precisely match the received DEM data, so in one area you might have 2cm/pixel visual accuracy but 10x worse DEM accuracy.  There are machine learning solutions to this that are quite complicated, but also introduce their own potential issues.

For example, this problem can be especially apparent with mountain river valleys. However, as we all know, rivers in the mountains lots of the time have a reasonable amount of elevation change to them. So if the algorithm is "find me water looking stuff and generate flat elevation data for that area" it isn't even immediately clear what elevation would be right. And then if things are running downhill, do you need to bake in some slope detection? How does it affect the surrounding terrain? Do you now, as a result, end up with areas of strange valley that have been "pulled" up or down to this new slope?

The same applies to coastlines, but maybe less in the extreme. It certainly isn't a super straightforward engineering or data problem to solve satisfactorily for all cases, as one algo that looks good for the coastline of Lake Michigan might look ridiculous for the Cliffs of Dover, and vice versa.

Once you get to these kinds of problems, I can say they can look from the outside to be conceptually simple. Of course, just adjust the elevation data so the water looking stuff doesn't ride up any coasts. But the devil is way way in the details with a lot of the issues that get forwarded in these Q/As; I could totally assign an engineer to a task like that and it would eat their entire next 2-3 months getting suitably decent.

Each on of these things is like that. ATC is an absolute time sink black hole of horrifying AI edge cases and deadlock conditions. A whole team of 10 could totally spend 6 months on it as a strike team and come up with something maybe only a little better than what exists. Many, many have tried, even commercially, in the real aviation world, with more resources. It's an incredibly difficult problem space.

I think something we could do better is communicating the depth of these engineering challenges. Lots of this is not low hanging fruit anymore that someone can spend a week or two on and fix. But it's really really hard to communicate these ideas on a good day, not in front of a camera, in my native language. So I definitely cut these guys a bit of slack for sure.

-Matt

Great points and I get it but I was the one asking the ATC question and that was a case where the question was way shorter then what i wrote.

There are challenging parts about atc improvement which I totally get. I fully understand why you get vectored into terrain and that this is a core issue with the data. And that is the part that got answered. I mean he told there is a roadmap so I Have some hope for the future. But I asked this question specifically because there are issue that are MUCH easier to fix/improve. 

Why can the ATC assign approaches automatically but not stars? The data is there. Why do I need to ask for vectors and not receive them automatically? The waypoints are there and the ATC knows where I am, since I get handed around all the time when I reach one of those. 

A simply system for emergencies, which simply lets you tell them which airport to get vectored too would be doable too imho. This could be greatly expanded later and I get why this is not a priority but nothing?

Telling ATC unable and some restrictions regarding flight plans should also be possible. The biggest problem I have is that there is no way to fly with atc without huge drawbacks at all. 

Pilot2ATC and Pf3 suck in regards of usability and can't communicate and control with live or offline traffic. That generic background chatter is not an option for me since this totally breaks my immersion if they talk to planes that are not even in the sim. So If I  would use that I would only have ATC speaking to me and me to them, but they would never communicate with other planes. 

Vatsim and co. and kind of interesting but no live ai traffic, often no coverage for when and were I fly and not all positions possible manned or always the same person. So while the concept is cool it has drawbacks too. 

Thats why I think the default ATC is so important and I have a feeling Asobo doesn't understand that. 

1 hour ago, pstrub said:

Replay functionality (minute 51:10): they want to go beyond normal replay, brain-stormed all kinds of ideas, such as recording the whole state of a plane, so that it'll be possible to enter the cockpit again at any point of the replay and re-fly a landing etc.

 

That's something I don't get, why not release a simple replay system, integrated in the sim, for the people who just want to watch how their landing went and also enjoy their payware airport from a different point of view.

And then, after a couple of month add functionality on top of it

When will their version be released, since it does not seem to be a priority and they have tons of work on other things? mid-late 2022?

 

4 minutes ago, bendead said:

 

That's something I don't get, why not release a simple replay system, integrated in the sim, for the people who just want to watch how their landing went and also enjoy their payware airport from a different point of view.

And then, after a couple of month add functionality on top of it

 

I too hope they're following this way, but still I think they have to consider it in the early design specification phase. If you don't, you end up doing a lot of work twice in software development that could've been avoided by proper planning.

Edited April 14, 20215 yr by pstrub

@MattNischan Genuine thanks for your insight on this. 

Between what you have said and the points @ChaoticBeauty has made, we can see the devil is in the detail, and sometimes people just don't have the same perception and even time to put into it, especially when you have a boss telling you what the business priorities are. 

This ambiguity or need for better clarity is actually the point that needs improving during the twitch feeds.  I could see Jorg at the end reading out the 'difficult' questions, and genuinely trying to get them answered, rather than avoiding them, and even Jayne was re-asking questions if she thought they hadn't been answered correctly (good on her, I was surprised at that!)

So it seems that for these sessions to be even better and more useful, Seb and Martial might need to take more time looking down the question list before the actual event, and if they really don't comprehend some of the comments or questions, ask for clarification before they give their answers.  It would be better for everyone and certainly people would appreciate it even more.

Apart from the complexity of the sim, two factors that I think aren't helping are the different versions/branches of the sim, so they don't always see the same issues as the end user (this has been mentioned - they sometimes have their own version or build), and also, some questions are genuinely embarrassing for them and they don't have the answer at the time, such as the deleted logbook question - I was cringing and felt sorry for Martial having to try to answer that one, but at least we got a genuine on-the-spot reply rather than some pre-scripted corporate speak! 

Edited April 14, 20215 yr by bobcat999

21 minutes ago, kakihara123 said:

Why do I need to ask for vectors and not receive them automatically? The waypoints are there and the ATC knows where I am, since I get handed around all the time when I reach one of those. 

I am wondering about that as well. Is this standard procedure nowadays? What if the aircraft is not equipped with a GPS system, does the pilot ask for vectors all the time?

Flight Simulator X had the ability to vector you to your destination, even if not very proficiently all the time. Since the ATC system seems to be based on that with several tweaks, I hope they can bring that functionality back without a lot of work.

15 minutes ago, bobcat999 said:

So it seems that for these sessions to be even better and more useful, Seb and Martial might need to take more time looking down the question list before the actual event, and if they really don't comprehend some of the comments or questions, ask for clarification before they give their answers.  It would be better for everyone and certainly people would appreciate it even more.

Absolutely, if they study the questions, the one from the forum, before the Q&A, I am sure we would get much better answers

54 minutes ago, kakihara123 said:

Why can the ATC assign approaches automatically but not stars? The data is there.

Martial attempted to explain this a bit but was having trouble with the right words. Yes, the data is there. However, the flight plan system is deeply linked to the ATC system, such that getting everything sync'd up there throughout the various systems that depend on it is no easy feat. Approaches are a separate flight plan and thus it can just be swapped out. So, there's large scale fundamental work required there inherited from decades of the platform that need refactoring. And you have to not break everyone along the way, because people obvious hate when things just break. Additionally, it's somewhat rare to get a totally different arrival than filed, so the cost/benefit has to factor in to that feature.

1 hour ago, kakihara123 said:

A simply system for emergencies, which simply lets you tell them which airport to get vectored too would be doable too imho.

This sounds pretty easy on its face, but also has a number of tough engineering challenges. One such example, live traffic. Does it cease to be live and now gets vectored out of your way? When you land how does the live traffic sync back up to its live position? And again, emergencies are a non-normal scenario and thus surely there would be people complaining about other features they feel are more core not being worked on.

1 hour ago, kakihara123 said:

Telling ATC unable and some restrictions regarding flight plans should also be possible.

This is really and truly hard. The cases when you say the aircraft can refuse your instructions go from "lots" to "infinity". It's one thing to have ATC cancel you when you veer X NM from your predicted path, it's quite another for the system to invent a set of possible alternatives for any given instruction refusal scenario.

I think for these reasons those specific features are unlikely to be high on the priority list given the available resources, realistically.

On the automatic vectors question, I don't have the answer. FSX took the route of effectively always starting to vector you when you got to a certain distance from the airfield, but that's not that realistic either. It's certainly something I can bring up.

-Matt

2 hours ago, ChaoticBeauty said:

I appreciate your post, and as I mentioned in my question thread, I know that it is not just an easy fix. But the issue does not have to do with the data source, it's rather a global problem where the shapes of the mesh are not being retained at lower levels of detail, and as CptLucky8 explained in that thread, the LOD rings are broken as of now.

All the love in the world to Jean-Luc, but there's a tendency there to answer with a flippant absolute surety that belies the engineering challenge being described.

He is correct, sometimes mesh topology is not retained with great accuracy when reduced in detail throughout the processing pipeline. However, maintaining roughly equivalent looking mesh topology while automatically decreasing vertex detail is seriously a PhD whitepaper level issue that has been studied extensively for three+ decades without a globally usable answer. No technique works equally well for all mesh shapes, and you can get some truly bizarro results sometimes. Yes, he points out one such solution, but as any modeler will tell you, they've used them all, and they all are just swapping different evils. You do the best you can, but with a sample size the size of the whole world, I'm not sure there even is a good answer. Maybe some kind of regionally selective topology reducer, but then you get into all kinds of issues with blending seams at region boundaries etc etc etc. Not to downplay the results you're seeing, I agree that sometimes they can be ugly, but just to make it clear what the level of difficulty actually is.

As far as the LOD rings go, that seemed to be vegetation related, on the question thread, and not mesh related.

-Matt

1 hour ago, kakihara123 said:

 

Pilot2ATC and Pf3 suck in regards of usability and can't communicate and control with live or offline traffic. That generic background chatter is not an option for me since this totally breaks my immersion if they talk to planes that are not even in the sim. So If I  would use that I would only have ATC speaking to me and me to them, but they would never communicate with other planes. 

 

You are totally incorrect regarding your statement about PF3. PF3 looks at the traffic flying in the sim, gives traffic advisories to the both pilots when they are nearing each other in distance and altitude, naming the heading, N number, and altitude of the AI traffic. Now I believe Pilot2 ATC just uses recordings of ATC facilities which have zero relationship to what is going on the sim, which I find silly frankly. . 

16 minutes ago, Bobsk8 said:

You are totally incorrect regarding your statement about PF3. PF3 looks at the traffic flying in the sim, gives traffic advisories to the both pilots when they are nearing each other in distance and altitude, naming the heading, N number, and altitude of the AI traffic. Now I believe Pilot2 ATC just uses recordings of ATC facilities which have zero relationship to what is going on the sim, which I find silly frankly. . 

That is not what I meant. I know it PF3 can see other traffic, but it can't direct it. It cannot change the approach of an AI plane. 

Also PF3 has a horrible hotkey system  and doesn't support Navigraph. It doesn't even assign approaches and stars so in this regard it is worse then the default ATC. There is also no plans to ever support VR since the developer doesn't use it. I don't blame the dev for not directing ai traffic since I know this is something that can't be done but thats still a drawback. 

Edited April 14, 20215 yr by kakihara123

4 minutes ago, kakihara123 said:

That is not what I meant. I know it PF3 can see other traffic, but it can't direct it. It cannot change the approach of an AI plane. 

Also PF3 has a horrible hotkey system  and doesn't support Navigraph. It doesn't even assign approaches and stars so in this regard it is worse then the default ATC. There is also no plans to ever support VR since the developer doesn't use it. I don't blame the dev for not directing ai traffic since I know this is something that can't be done but thats still a drawback. 

All I have to do is remember one hotkey, which isn't really hard, and I can see all relevant hotkeys on the frequency I am on. The three hotkeys I use all the time are assigned to my yoke buttons. PF3 does use whatever is in the sim. I just updated it after installing the MSFS update. 

26 minutes ago, MattNischan said:

All the love in the world to Jean-Luc, but there's a tendency there to answer with a flippant absolute surety that belies the engineering challenge being described.

I presume Jean-Luc is CptLucky8? I don't think he was saying anything about the vegetation in that thread, his claim about the broken LOD rings comes right after mentioning the new threaded tessellator. His exact quote is:

Quote

I believe Sebastian was mentioning the new tessellator around the time of the USA update indeed.

However I don’t remind any FS2020 version preserving LOD shape. The problem has always been there. The difference is that it is most likely more visible now because of the more recent additional bug (I believe it is bug) where it is using even shorter LOD ring distances. Before it was a little longer yet still shorter than you’d think

 

28 minutes ago, MattNischan said:

However, maintaining roughly equivalent looking mesh topology while automatically decreasing vertex detail is seriously a PhD whitepaper level issue that has been studied extensively for three+ decades without a globally usable answer. No technique works equally well for all mesh shapes, and you can get some truly bizarro results sometimes.

I understand that it's a hell of an undertaking, and I'm not suggesting it is something they can do in a few months and without lots of experimentation. I'm just saying, whatever system they had before the USA World Update worked a lot better than it does now, so it's not like they have not already achieved a good result that somehow went away, and I'm impressed that they don't seem to be aware of that regression. But I guess what's done is done, there is nothing else we can do to get an answer for this.

Why would you need to remember hotkeys for PF3 when you have MCE to interface with voice recognition? Is it not the most realistic way to interact with (any) ATC??

Edited April 14, 20215 yr by Jean-Claude

2 minutes ago, Jean-Claude said:

Why would you need to remember hotkeys for PF3 when you have MCE to interface with voice recognition? Is it not the most realistic way to interact with (any) ATC??

When I got my pilots license I had to learn a great deal of terminology to use when talking with ATC. Learning 3 or 4 hotkeys, in comparison, is a walk in the park. If you want to use MCE with PF3, it works very well. I hate wearing headphones when flying a sim, same goes for VR headsets.