FSDreamTeam forum
FS9 support => JFK for FS9 => Topic started by: cburna on September 20, 2008, 11:01:58 pm
-
Has anyone else noticed that the aircraft land to the side of the runway rather than on it during the Canarsie approach?
Is there a way of solving this problem?
-
Have the same exact problem. To help people out, I've attached a picture of an aircraft supposed to be landing on 13R.
[attachment deleted by admin]
-
It's a flight model problem.
Yesterday I've tried watching some AIs and, for example, a MD11 from FS Painter landed perfectly on the centerline, a CRJ from CDAI landed perfectly on the centerline as well. A default MS 737 couldn't.
Since the approach file is obviously the same, it's clear that the only thing that changes are the different flight models. AI designers not always test their models to be able to perform a curved approach. The issue is that FS tries to figure out when initiating the turn depending on the flight model. If you follow different AI models, they don't start their turns over the approach waypoints at exactly the same spot. Some turn earlier, others turn later, but the approach has the waypoints always in the same places, which of course are the real-world positions.
-
Same here
I have uniform FDE (based on default FDE) created by myself for all of my AI commercial jets . This approach works flawlessly for AI with other specific appoaches such as VHHX IGS13 and KDCA 19 River visual. It's not a case with KJFK unfortunately. Let's hope that Jim Ville is reading us.
-
Hello Dimon
There are 2 objectives for any airport.
1. Make it look as good as or better then the default FS9/FSX airport
2. Make it work as good as or better then the default FS9/FSX airport.
With that said there are also 2 types of scenery in FS9/FSX
1. The visible scenery
2. The invisible scenery
In all cases Scenery Designers work with the number 1's above
In all cases I work with the number 2's above.
I have never claimed too be a scenery designer and leave that to the experts like FSDT. I have studied the invisible scenery as long as the Scenery Designers have studied the visual scenery. I have always given credit where credit is due along with the respect that a Scenery Designer deserves for their knowledge and hard work.
Many of the techniques that I have introduced over the years such as the Crosswind Runway Technique, non-related weather type curved, circle to land, LDA, IGS approaches, 3 or more multiple parallel runway usage based on Empty Weight of both AI and User Airplanes as per ATC, STAR arrivals used by AI Planes, etc. did not occur overnight. I also have earned the respect from some but not all the highend Scenery Designers for all the invisible scenery I work with and have been given in some cases credit where credit was due.
Scenery Design is very competitive because more then one Freeware/Payware development group exist. That helps to keep honesty and integrity amoung all the designers so the User can see the end result of the visible scenery. On the flip side of that coin where is my competition. If I post that something is not a limitation with FS9 or FSX and a scenery designer disputes what I say then where do I go from there. I learned my lesson back with Cloud9 and EHAM and it was a no win situation. If a scenery designer delves into the world of invisible scenery then it is up to the end user to become more knowledgable and dispute what might be said or where fingers are being pointed.
As noted it is very important that AI models have a good set of FD's so they do not fight the approach to the runway texture. You know that some independent people have in fact designed .air and .cfg's that test their AI Models against both my approaches and the default FS9/FSX curved approaches. It is worthy to point out that all AI models may not be able to fly a curved approach accuratly. However there is a lot more to getting the approach to work besides what AI Model is flying the approach.
I would like to say that AI Planes do not fly a AFCAD style enhanced airport ILS but only a User Plane flys a ILS based on a transmitter location.
The AI Plane knows nothing about a VOR, NDB, Waypoint or Terminal_Waypoint but only understands headings within the visual airport zone and runways listed in the bgl. When I say runway I do not mean the black runway link line but the default texture of the runway regardless if it is covered by a 3rd party runway texture.
The User Plane understands some of the Navaids on approach if and only if the Pilot uses the GPS receiver. The AI Plane works from an invisible target which once again is my area of expertise in knowing where these invisible targets can exist that is part of the AI Engine.
If I work with a design group to add an enhanced type approach I am also the one that does the AFCAD (FS9) or ADE (FSX). The 2 must work in unison with each other for any type approach code written. AI Planes unlike the User plane must use code to Transition from flying characteristics to ground behavior controlled by the ATC engine. Very large airports unlike smaller airports use what is called a throttle valve which throttle backs certain AI Plane usage regardless of the slider position percent setting for AI Planes. In FS9 it is hidden code but is exposed in FSX. When FS couples that to the segments of the taxiway network then and only then can ATC be precise in how it controls the AI Plane behavior.
The AFCAD designer for both FS9 and FSX will in most cases try to out think the code that is embedded into the AI/ATC Engines and remove parts of a path network that is very important for how a User can change certain features based on the ATC selection of taxiways and active runways. Any non-compliant changes made then reverberate all the way back to the spawning point of the AI Plane which in some cases is 108NM's from the airport or actually the User Planes position. Some of the different FS Engines and code that has to be considered when writing approaches are
winds
base vs recip end
parallel runways vs non-parallel
CW runway technique in use to simulate a parallel group within 7.9 degrees
weather visibility
Start Location on or off the runway texture
Hold Short Node at takeoff end
Main Terminal placement between parallel runways or to the side of parallel runways
parking spot Lat/Lon vs runway center line
runway length based on type aircraft (6 default type aircraft in the .air vs EW in the .cfg)
An example is a AI Plane may sometimes go missed especially in a curved approach because the Start Location is off the runway texture and located on a taxiway. Why in the world would a AFCAD Designer try to out think the AI Flightplan code is beyound my comprehension. Start Locations are also Ending Locations for any FP that is generated with the TDBB or TTools. The curved approach only gives the AI plane a few seconds to find the runway (does not know where it is at) after the turn so ATC can calculate the contact points of the main gear which holds certain key coded ATC commands.
Could the curve approaches to runway 13L and 13R be more exact? Yes. But some of the Scenery could also be tweaked. I suppose that is what updates are for and FSDT seems to be on top of that based on Users input.
Jim
-
Hi Jim,
A very nice text from maybe the person who knows more about AIs in the world of flight simulation. I know your work quite well. I also asked some things to you and I am sure that the world of FS want, need and appreciate your work as well, but our question stays without answer: when a 747 flyes the canarsie approach and lands inside buildings, should we change the flight model or should we modify the approach path? For example for the 747 there is only one type of model... the one from AI aardvark. Should the starting point be moved to the runway? What should be done? We wait for your solutions. :)
harpsi
-
I would like to point out that we certainly believe AFCAD, AI and Approaches are important but, AS ALWAYS in scenery developement, everything is made of choices and compromises and trying to figuring out what would accomodate most of the users.
Many times, doing one thing "right", breaks something else. Since everybody have a personal view about how Flight Sim should be used, there are a lot of often incompatible things that we need to keep together, and ultimately the only judge is, I'm afraid, the market.
So, for example, let's recall the long threads about the infamous "bridged taxiways". We are strongly opposed to that, even if we are perfectly capable of creating it, but that would have a far too big impact on the behaviour of the airport because other designers that made them, had to delete the taxiways in the AFCAD, in order not to create a mess with AI, which wouldn't go over the bridged taxiways anyway, impacting probably even the whole airport operation, the choice of runways (a runway that has a bridged taxiway as the only connection with the rest of the airport, would probably been cut off forever).
In this case, we made a choice in favor of the "invisible" airport, at the expense of the "visible". Functionality in this case has to win over eye-candy. Of course, here and there, I get emails or PM or even public message by users not agreeing with this choice, with the reasoning that "nobody on VATSIM use AI anyway"...so there's a part of user obviously not interested at all with anything related with AI, AFCAD, Approaches, etc. They only fly online, they have human controllers, they use their own charts and couldn't care less if an airport has an AFCAD in the first place, let alone one that works well.
Jim Vile rightly says that the default textured runway placement has an impact over the AI/ATC behaviour. Ok, we agree. This means we need to place a default runway, which is something that many users hates in a payware scenery. They expect a payware scenery not having any *visible* default element. So, in Zurich we had fully customized runways without default textures, which looked brilliant and close to the real one, but of course had other side effects, like problems with AI models made for FS9 when used in FSX. We favored the visual aspect here. Was it a good choice ? Lookin at sales, yes... and the customer is always right, they say.
At O'Hare and JFK, we went back trying to create a more compliant airport, but then we need to take a lot of steps because, in order to make both visualists and behaviourists happy, we need to put default runways to make the AFCAD (and AI engine, and ATC) happier, satisfying the behaviourists, but trying to *conceal* those runway textures, to make the visualists happy, because they scream each time they see something that looks like default.
About the approaches at JFK. The approach file is made using correct real world positions and headings. It might be probably tweaked but, right now, real world positions and headings are there, waypoints needed for the approach are there, and it draws correctly on the gps. Further tweaks are probably either to be done in the main AFCAD.
A LOT of AI I have tested are perfectly able to perform the Canarsie approach without any problems and land exactly on the centerline. Some of them fails, yes. The default 737 fails, for example. Again, a matter of identifying user demographics. How many users still fly with default AI in FS9 ? I don't know exactly, but probably not many of our users. Should we worry if the default 737 can't perform the Canarsie approach ? OR, should we worry about the approach file (or the main AFCAD) being wrong but then, why many AI don't exhibit the problem ? We should ideally try to tweak the file to be able to perform with the maximum number of AI models as possible but, I'm sure will never find a way to make it right for everyone.
For example, the suggestion about the start positions looks interesting. I don't know who started it, but there's a new trend for scenery designers to put starting positions for a runway over the hold short zones. Which is convenient for the user, if you think about it, but since FS used the runway numbers (more or less) as starting positions since FS1, perhaps it's not a good idea to deviate from this assumption...
Without forgetting, as I've said before, that a good percentage of users (online flyers) couldn't care less of what we are talking about here, and might instead prefer we'd relase another scenery quickly, instead of focusing too much on the AI/ATC engine issue. And they are customers too...
If you want my *personal* opinion, I would like to have the AI/ATC done right, because I never fly online, I like the default ATC regardless of how many defects it has, and I like watching AI, because I woulnd't have too much time for flying anyway...so, you can be sure I'm listening to these issues.
-
A LOT of AI I have tested are perfectly able to perform the Canarsie approach without any problems and land exactly on the centerline. Some of them fails, yes. The default 737 fails, for example. Again, a matter of identifying user demographics. How many users still fly with default AI in FS9 ? I don't know exactly, but probably not many of our users. Should we worry if the default 737 can't perform the Canarsie approach ? OR, should we worry about the approach file (or the main AFCAD) being wrong but then, why many AI don't exhibit the problem ? We should ideally try to tweak the file to be able to perform with the maximum number of AI models as possible but, I'm sure will never find a way to make it right for everyone.
What AI models were able to correctly fly the approach? I have Aardvark, Fruit Stand, Alpha-India, EvolveAi/DJC, and AI Malcontent models flying into my JFK and none of them except for the Aardvark 747-400 even landed on the runway.
-
What AI models were able to correctly fly the approach? I have Aardvark, Fruit Stand, Alpha-India, EvolveAi/DJC, and AI Malcontent models flying into my JFK and none of them except for the Aardvark 747-400 even landed on the runway.
Yeah...I'm wondering that too. I've been observing the Canarsie approaches as I'm typing this (have been for close to an hour) and I'm yet to see a single successful landing. Most end up either in the water (13R) or on the taxiway (13L), regardless of the AI model.
-
What AI models were able to correctly fly the approach? I have Aardvark, Fruit Stand, Alpha-India, EvolveAi/DJC, and AI Malcontent models flying into my JFK and none of them except for the Aardvark 747-400 even landed on the runway.
This is what I've found with some testing:
AiArdwark 747 lands on the runway
CDAI CRJ lands on the runway
FSP MD11 lands on the runway
Evolve AI A319 lands on the runway
PAI Dash8-Q200 lands on the runway
WoAI 777-200 lands on the runway
ERJ 135 AI Malcontent lands on the runway
AiArdwark 737 lands on the side of the runway
PAI CRJ-190 lands on the side of the runway
Well, at least I've found some that does't land correctly, so we can start tweaking, hopefully, it will not break those that works...
Note that, to test correctly, you should probably put a straight headwind. If there's a crosswind, there's another variable in play you are testing, which is the ability for an AI flight model to perform a crosswind landing
-
What AI models were able to correctly fly the approach? I have Aardvark, Fruit Stand, Alpha-India, EvolveAi/DJC, and AI Malcontent models flying into my JFK and none of them except for the Aardvark 747-400 even landed on the runway.
I have exactly the opposite. The 747 is exactly the worst aircraft in terms of landings. The 757 from AI aardvark flies good. The ERJ 145 resonable as well. The rest land outr of the runway.
harpsi
-
Now, it would be better if we don't dive into *release dates* of these respective .AIR files, because it might become simply too complex to test.
-
That approach seems to have a sharper curve than the Canarsie approach so it's interesting that AI Traffic flies that approach so well but can't fly the Canarsie approach correctly.
Yes, but at Kai Tak, the waypoints that made the approach are more spaced away from each other. At JFK, the waypoints are closer, perhaps this might be a factor (Jim, do you think this might be an issue ? ).
We started with the most obvious method: putting the waypoints at their real world places, with real world distances and headings. This results in the approach patch drawn on the GPS exactly like the charts.
-
In FlyTampa's Kai Tak, AI Traffic always lands on the runway when flying the IGS approach regardless of the model. That approach seems to have a sharper curve than the Canarsie approach so it's interesting that AI Traffic flies that approach so well but can't fly the Canarsie approach correctly.
I will forget Canarsie approach... I tried to move the start locations to the runway, the models are the same as for other sceneries like Kai Tak and every aircraft lands in the middle of the airport... It is almost impossible to use this file. I thought that changing the little runways on the afcad would have some effect but no. I am doing spotting at runway 13 and I am seeing all the landings in the buildings...
harpsi
-
Umberto
I found your response to my post very informative. We are very much in agreement with what you said and I for one appreciate your concern to address many different Users that want FSDT to work based on their needs.
Jim Vile rightly says that the default textured runway placement has an impact over the AI/ATC behaviour. Ok, we agree. This means we need to place a default runway, which is something that many users hates in a payware scenery.
I did not explain that very well and want you do know that what you are doing is correct. You can use any runway texture you want (default or your own) and that has no effect on AI Plane behavior. If you open up the FS9 or FSX JFK AFCAD the Apron link lines (green) as taxiway paths is a good idea which FS does not have to use resources to draw a taxiway texture since you use your own. The runway is different. Notice you left the default runway texture in place. Even if you use your own runway texture your AFCAD is correct because FS always works with the default XML runway or the one copied into the AFCAD. Embedded in the default XML runway (that is always present) if you decide to use your own runway are codes for the target touchdown point for AI Planes, ATC release commands from one state to another state, wheel contact code for the main gear so ATC knows where the AI/User plane is after touchdown, etc.
If a 3rd party runway is placed to give the airport a more realistic look it is very important that the AFCAD default XML runway be of the same width and length underlying the new runway texture. If a AFCAD designer shortens the XML runway as seen with AFCAD then they must also be prepared to answer certain AI/User Plane behavior issues that will show.
If a User or AI Plane does not land on the XML runway texture regardless of what runway is used (new 3rd party scenery or default) ATC looses the where abouts of that plane. This is easy to see when a plane misses the runway on landing because the ATC command state changes and ATC Tower releases the next plane for takeoff that is waiting in line at the entrance hold short node. You now will see 2 AI planes on the same runway where one is starting its takeoff roll while the other one landing has not exited the runway yet.
Contrary to popular belief the exit hold short node has nothing to do with a command to contact the ground controller for parking instructions. That ATC command comes from the main gear contact points as soon as the main gear exits the side of the XML runway texture.
I just wanted to be sure you and I are on the same page if I failed to explain that any enhanced runway texture works in FS9/FSX as long as the underlying XML default texture stays in place. If you decide to add a more realistic looking runway rather then the default runway textures you can also remove all runway markings and lights from the default XML runway and only leave the texture and runway black link line.
Jim
-
Harpsi
In my 1st post I said
An example is a AI Plane may sometimes go missed especially in a curved approach because the Start Location is off the runway texture and located on a taxiway.
I did not say by moving the Start Location would help a AI Plane to find the runway after the turn to short final.
-
Hi Jim
After some posts and as a conclusion, what can we do to have Canarsie approach flown in a good behaviour? ;D
harpsi
-
Harpsi
I will take a deeper look into the approach code if thats OK with FSDT. Presently the AI Planes are attempting for a circle to land rather then a right curve approach. This is no fault of FSDT but unique to the AI behavior and some times a compromise has to be added to the approach code.
When AI are thinking in code for a circle to land (starting the approach) and then at the last minute see the extended runway target point to the right of the plane (RWY 13L/R because they are on a right base) it can cause some planes to have difficulty in the last turn. The Canarsie Approach for the User Plane is not a problem but based on certain geometry within the AI Plane code when it is flying the Canarsie Approach it is thinking circle to land and not curve to land. No worries, that is correctable with heading angles in respect to runway angles.
The reason some will see a difference on any one type AI Plane from one computer vs another computer is simply a FPS issue. Each User looking at the AI Plane landing is using a different time, different amount of AI are parked or in the visual zone and FPS based on speeds of computers play a part in the smoothness of when a AI starts the turn to final.
I may have the same B747 that you have and it will land ok for me put not for you. Any straight in type approach for a AI is always seeking the target point of the XML runway texture so slight hesitations do not have the same affect on a AI plane vs a curved approach. A curved approach has no runway texture target point to seek from 7-10 miles out and only has a heading to work with. Only when the AI nears the runway does it then try to correct itself and makes a turn toward the runway target point in the XML code for the touchdown point.
If a AI Plane could fly Navaids it would have targets like the User plane but AI does not fly Navaids. All the default FS9/FSX weather related curved approaches that are in use in the approach code database are nothing more then tricks to make a AI plane think it is on a long final for a straight in approach. Only at the last few seconds does the AI plane fiqure out it is not in line with the runway and makes a turn to compensate for the correct final. If my computer can cause AI to fly a smoother curve then your computer I have a better chance of seeing the AI find the texture of the runway.
Of course the approach that I write compensates for both the slowest and fastest computers out there so each user should see about the same thing even if FPS are different between each computer.
-
Hi Jim
Very good explanation. I wait for your solution, otherwise I need to change to ILS 13 definetelly because I didn´t see more than 3 good approaches until now. I am sure that like all the other work you did, you will find a good way to solve this one as always. :D
harpsi
-
Jim,
Just out of curiosity, how diffictul would it be to make AI Aircraft fly SIDS? At JFK, all aircraft departing 31L do a sharp left turn and do the Canarsie climb. I thought I remembered reading somewhere (Maybe on the FlyTampa forums) that it wasn't possible to modify what an AI Aircraft does after departure but I was just wondering.
-
MHTripple7
Up to now I have not found anyway to control AI on departure. The most I can see in the AI code is they always turn using the shortest radius in a circumference to join up with their entended Flight Path. The AI Flight Plan is from the parking spot to the closest Start Location of a runway at the arrival airport based on a world setting of winds.
Departing 31L/R to KPHL = left turn
Departing 31L/R to KBOS = right turn