My model airplane output for the last few months

[Tim Wescott]

I, uh, got obsessed.  This video represents about three months of flogging the simulation to get the pattern sorta OK, followed by half a day learning how to get rendering working on Blender.

Obviously, I need to work on rendering...

I'd like to say the pattern is perfect, but if you look at the square and triangular maneuvers, you'll see that it needs some work to be as pretty as possible -- and even then folks will probably argue over "perfect".

Interestingly, the camera placement is right at the judges position, and boy does it not make for a nice video -- I think this is the biggest problem with trying to present the pattern in video format: video is static, or moves the camera around without the viewer being connected to it.  Actually watching control line involves moving your head around while you watch: that motion feedback informs your brain of what you're watching.

Eventual goal is to see if I can get this rendered for 3D goggles, so you can stand at the judge's position and see the plane while moving your head in a realistic fashion.

[EricV]

Interesting project. I have questions...

What is the FOV (Field of View) chosen in your video display at the mid point? (at the pilot)
Is this based on the natural view of the human eye?

Orientation without a background and horizon is a little difficult, I had to fast-forward to the V8 to find your 90Deg then re-watch to get my brain to settle down and stop complaining. LOL!

Either way, nice work. It will be interesting to see how far you go with this and if you meet your end goal.
EricV

[Tim Wescott]

The ultimate goal is judge training.  I want to get a realistic view of a flight from the judges position that someone could watch in their living room.

Interesting project. I have questions...

What is the FOV (Field of View) chosen in your video display at the mid point? (at the pilot)

FOV at the pilot?  That doesn't scan.

The camera is placed at the judge's position, with a field of view wide enough that the airplane fits into the scene at all times (well, mostly).  That means the FOV is 120 or 125 degrees (I can't remember which).

This is a stupid-wide field of view, but it's necessary to capture the whole flight in landscape mode.

For a prettier staging, and for fairly good pilot training, I could narrow the field of view and back the camera off -- but then the maneuvers are distorted differently than what a judge would see.

Is this based on the natural view of the human eye?

No camera can fully replicate the natural view of the human eye -- hence my ambition to eventually render this stuff for VR goggles.  The natural tendency is to keep the plane centered in the fovea, while really only consciously seeing the plane and its path through the air.

A normal judge moves their head while watching, while unconsciously using proprioception and the background in their peripheral vision track the plane's motion through the sky.  As an experiment, I've tried holding my head still and staring at the center point of a maneuver.  This may be a valid judging technique for someone, but it's not going to work for me -- and that center point will vary from maneuver to maneuver anyway.

Orientation without a background and horizon is a little difficult, I had to fast-forward to the V8 to find your 90Deg then re-watch to get my brain to settle down and stop complaining. LOL!

Sorry about that, but only a little bit.  I wanted to throw this video out there early, and then add refinements like better rendering, a background, etc. as I learn how to use Blender.

Either way, nice work. It will be interesting to see how far you go with this and if you meet your end goal.
EricV

My rough set of steps are

• Get a flight synthesized (done!)
• Get the thing satisfactorily rendering in 2D in Blender (started but obviously not done)
• Figure out how to render in a format for 3D goggles.
• Get some 3D goggles and try it out, iterate as necessary.
• Release it to the wild and see how folks like it.
• Perfect the flight (ugh, those squares...)
• Introduce realistic imperfections into the flight (this is -- not trivial)
• Go back to "release it to the wild" and iterate until I give up or it's a useful tool.

[Ty Marcucci]

Even from the judges point of view, those maneuvers seem a tad too small. A great training aid other wise.

[Tim Wescott]

Even from the judges point of view, those maneuvers seem a tad too small. A great training aid other wise.

Trying to render it from the spot the judges are standing does not, I believe, work if you're just using a camera.

I am probably going to do a render of this with the camera standing further back. Basically from the point of view of the tabulator's tent for our local contests. That way the airplane won't be radically changing size as it goes backward and forth, and the maneuvers will look more sensible.

But then, everything will be distorted compared to what the judges perceive.

[Steve Helmick]

The tricks seem too small, like Ty said. I liked it better when I cranked the replay speed up to 2x. I'll try to watch the whole pattern soon.  Steve

PS: Adding a level "ground" would be my first suggestion. Speeding up the level laps to 2x or 3x would be awesome, while leaving the actual maneuvers at 1x. No idea if/how to do that, of course. 

[Tim Wescott]

The tricks seem too small, like Ty said..

A lot of that is the rendering. Some of that is that they're by the book: 45 degrees, period.

[Keith Renecle]

Hi Tim,

Did you ever see my old CL-Sim from 2004? In case you missed it, you can download it from my Google Drive here: https://drive.google.com/drive/folders/1Onf9bLUjRvuoW6b7-z531ld55Y0KzB8i?usp=sharing

I used the old Blender Game engine back then, so that instead of a video from a fixed position, the user could move around the basic scenery. The squares are not correct to the newer F2B rules and the sharp corners are 5 foot radius. It took me many months to do this back then. It had about 8500 frames or something like that. It was a fun project but started many silly debates, some of which are still ongoing. The latest versions of Blender are incredible, but then, so is the learning curve!

Regards,
Keith

[Tim Wescott]

Keith, good to hear from you!

I do remember it! I would have tried to talk you out of source code, but I thought the plane simulation part would be easy-peasy.

You showed that it's possible (and that, yes, it sparks debate - but debate means folks are thinking, so that's good).

[Ken Culbertson]

 I am baiting my breath for the version that will show the hemisphere for reference.  I too was a little shocked at how small things looked until I went through my videos and found one shot from the judges position.  I normally shoot from about 60' behind to get the full frame but this one was shot 20' from the side to measure elevations and what antics the pilot's hand was up to.  I measured the length of the plane as it crossed in front of me on the screen (120mm) then when it crossed the tree on the opposite side of the circle (12mm).  That is almost exactly what your simulation shows.  If your math is right, which I suspect it is, then the maneuvers really are that small when viewed from the judges position and if all of that is true then I am really not flying too small.     Now all I have to do is convince the three national judges, two national champions and a recent world team member that I fly with that they are all wrong.    Why do I feel that the term "Dead Right" is taking on a more literal meaning.

Great Work! - Ken

Ken

[Keith Renecle]

Keith, good to hear from you!

I do remember it! I would have tried to talk you out of source code, but I thought the plane simulation part would be easy-peasy.

You showed that it's possible (and that, yes, it sparks debate - but debate means folks are thinking, so that's good).

Hi Tim & Ken, It's been 20 years since I created the CL-Sim, but if you email me (krenecle<at sign>netactive.co.za) I can give some shortcuts to creating the background scenery etc.  For Ken, I found that for judges training, that the maneuver shapes on the sphere without he model flying works better for understanding the shapes. The old sim also shows the spherical grid if you press F11. All of the functions are in the user manual.

A few years ago, I did another useful tool called JTS or Judges Training System which shows a higher resolution graphics and all of the maneuver shapes as per the latest F2B rules. Some folks found the hemispherical grid confusing so I reduced it to just the downwind side of the grid. Personally I find the full grid a lot better. Here is the download link:

JTS User Manual:
https://drive.google.com/file/d/1LL9aQcznMRvXU-CnP63_-UR26SjXDyzv/view?usp=sharing
 
JTS Program files:
https://drive.google.com/file/d/1dDr6A-GLvPazOqaC1nkEiyU_mNt-qBqy/view?usp=sharing

Keith R

[Tim Wescott]

Here's an update, with a horizon and ground and stuff.  It shows some imperfections in my programmed flight that didn't show as badly as before.

[Ken Culbertson]

Here's an update, with a horizon and ground and stuff.  It shows some imperfections in my programmed flight that didn't show as badly as before.

I am loving it!  If you show it with 8' corners I would watch it every day.  Fantastic judge training tool.

Ken

[Howard Rush]

  If you show it with 8' corners I would watch it every day.

As is, it looks like Orestes flying.

[Keith Renecle]

Good job Tim! I think though that your model has way too much flap area and/or throw!  You can see how it lifts without pitching the nose up.

To add the sky and clouds, just add a "skydome" which is a hemisphere that covers your scene and that's a lot bigger than your scene area. You then wrap a cloud picture (texture) around the sphere. If you need some help with this then let me know.

Ken, why do you want to see 8 ft. corners? Most of our good stunter's are flying about double that radius. I seem to remember that Brett calculated that 13 ft. would be about the sharpest possible corner. Just btw, did anyone look at my JTS program. I would appreciate some feedback. Thanks.

Keith R

[Ken Culbertson]

Good job Tim! I think though that your model has way too much flap area and/or throw!  You can see how it lifts without pitching the nose up.

To add the sky and clouds, just add a "skydome" which is a hemisphere that covers your scene and that's a lot bigger than your scene area. You then wrap a cloud picture (texture) around the sphere. If you need some help with this then let me know.

Ken, why do you want to see 8 ft. corners? Most of our good stunter's are flying about double that radius. I seem to remember that Brett calculated that 13 ft. would be about the sharpest possible corner. Just btw, did anyone look at my JTS program. I would appreciate some feedback. Thanks.

Keith R

I totally agree that most top cornering fliers are in the 10'-13' range depending on the corner.  However, an 8' corner is possible, I turned them that tight with my canard. They looked ridiculous.  I was curious to see what they would look like in this simulation.  For a training tool and as a goal, I think 12 would probably be better.
My personal view on corners is that they should be measured against an attainable size and not against what the top fliers are doing.

As a side note, I measure using plane length and distance traveled from the fuselage mid point on the last frame where the plane is horizontal to the first frame where the plane is vertical.  This would not count slide which probably should be included but the eye really sees the angle.  It is not fair to measure the vertical to horizontal corners as a benchmark since they are flown much slower and have a positive boost from gravity.

Ken

[EricV]

As is, it looks like Orestes flying.

Own Richards (no longer with us) would have said I was flying "butt cheeks" if he saw me flying the square 8's in this rendering and tell me to fly them wider.
Always made me wonder who's butt he was looking at?
EricV

[Tim Wescott]

I am loving it!  If you show it with 8' corners I would watch it every day.  Fantastic judge training tool.

Those are 7' 4" corners; I'm not sure that 8' would look substantially different.

I need to open them up a lot to look good.  If you watch the square corners frame by frame you'll see the nose of the airplane getting below the down-line in the turns to level -- that's just a consequence of geometry.

Eric: yup, all the square maneuvers are at least a bit wrong -- the triangle and hourglass are too fat, and the squares and square eights are pinched on top.  I'm going to live with that until I have things rendered the way I want.

Keith:  About the flaps: anytime the thing is descending it's using a separate bit of code that keeps the plane level while magically descending.  That's another thing to fix, later.  Interestingly, it wasn't nearly as apparent until the thing was five feet away from a ground surface -- then it jumps out.

And yes -- I know about painting a picture of a sky onto the background.  That's on my list of rendering improvements.

Render, then make the "perfect" pattern more perfect, then start introducing controlled errors of the sort that people typically do.

[Keith Renecle]

I totally agree that most top cornering fliers are in the 10'-13' range depending on the corner.  However, an 8' corner is possible, I turned them that tight with my canard. They looked ridiculous.  I was curious to see what they would look like in this simulation.  For a training tool and as a goal, I think 12 would probably be better.
My personal view on corners is that they should be measured against an attainable size and not against what the top fliers are doing.

As a side note, I measure using plane length and distance traveled from the fuselage mid point on the last frame where the plane is horizontal to the first frame where the plane is vertical.  This would not count slide which probably should be included but the eye really sees the angle.  It is not fair to measure the vertical to horizontal corners as a benchmark since they are flown much slower and have a positive boost from gravity.

Ken

I have never seen any of the top fliers flying a 10 ft. corner, or even a 13 ft. corner, and I've traced just about all of the world's best fliers. Many appear to have really sharp corners but when you trace them, frame by frame and measure them, they are closer to 16 ft. I watched Yuriy Yatsenko flying his Classic and his Gee Bee and the corners really look sharp. I also have the full set of DVD's from Richard Oliver of the 2004 world champs and US Nats. At around that time, I asked my young friend and good stunt flier to fly me his sharpest corner with his lightweight Trivial Pursuit. I then copied each frame of the corner and measured the radius, using as you did, the length of the fuz which is close to 4 ft. As you can see from the attached pics, the model skids through the corner and it ends up as a 16 ft. radius. In real time the corner really looks sharp.

I do think that it may be possible to hit that 8 ft. corner with a very light, slow flying model, but then of course, the rest of the pattern will be difficult, especially in any wind or turbulence. I would love to have seen Scott Bair's Stuntfire flying at 6.5 secs/lap, but once again, I reckon that in lousy weather (normal competition weather!) it would be difficult to fly a good pattern. That flying wing with the 1/2A engine pointing out at around 45° was supposed to be able to hit a 5 ft. corner. Just btw, my old CL-Sim's square corners are 5 ft. The last pull-out after the clover is 10 ft. which I threw in for fun!

[Keith Renecle]

Those are 7' 4" corners; I'm not sure that 8' would look substantially different.

I need to open them up a lot to look good.  If you watch the square corners frame by frame you'll see the nose of the airplane getting below the down-line in the turns to level -- that's just a consequence of geometry.

Eric: yup, all the square maneuvers are at least a bit wrong -- the triangle and hourglass are too fat, and the squares and square eights are pinched on top.  I'm going to live with that until I have things rendered the way I want.

Keith:  About the flaps: anytime the thing is descending it's using a separate bit of code that keeps the plane level while magically descending.  That's another thing to fix, later.  Interestingly, it wasn't nearly as apparent until the thing was five feet away from a ground surface -- then it jumps out.

And yes -- I know about painting a picture of a sky onto the background.  That's on my list of rendering improvements.

Render, then make the "perfect" pattern more perfect, then start introducing controlled errors of the sort that people typically do.

I used the Blender Game Engine to do the CL-Sim, and I don't have good coding skills, especially back in 2003 when I started on the project. I drew a wire-frame sphere scaled to have a 70' radius, and then I drew the maneuver shapes in the correct place "downwind" on the sphere. Then I simply rotated the model that's center was fixed to the center of the sphere, and did each keyframe like that, manually rotating each axis  around each maneuver shape. I can remember that Igor noted at that time that there was no skidding into the corners. I threatened to add that but never got around to it. It did however serve to illustrate the errors in the rules and I used it to create my presentation that I gave in 2007 at the Euro Champs in Serbia to good effect. It still took until 2010 to actually change the F2B rules to correct the many errors that were there since the early 60's!

There was one "funny" at the top of the vertical 8 in my sim that had something to do with Blender's animation system when the axes change around. I could not fix that, and put it down to "turbulence"! Blender dropped their game engine, but recently a group of enthusiasts revived it again calling it UPBGE and it runs in Blender version 3.6. I wanted to use the sim as a game rather than a video, so that the user can move around and get the idea of viewing the maneuver shapes from any perspective. I had a lot of fun back then and I'm sure that you're having some fun now too....and loads of frustration!

Just btw, I see that you "fly" the model in your squares with the tops following the 45° parallel line. I looked at it frame by frame. My CL Sim does the same, and we put that in the F2B rules at that time. I understand that many feel/believe that this is how the rules should specify this line....BUT, no pilot actually flies it that way. They all fly it as a spherical straight line or great circle path. If you look at the traced flight path's in Alberto Solera, and Andrey's tracker program, you will see this clearly. I've had a hard time trying to convince the majority of pilots and judges about this in the F2B rules, but I guess that just as before, it will take some time!

[Ken Culbertson]

I have never seen any of the top fliers flying a 10 ft. corner, or even a 13 ft. corner, and I've traced just about all of the world's best fliers. Many appear to have really sharp corners but when you trace them, frame by frame and measure them, they are closer to 16 ft.

When I redid my calculations including the skid I found your numbers to be more realistic.  If they are then that means that out of a 54' arc (45 degrees on 70' lines) there are 32' of corners and 22' of flat.  To me this explains why what we see is so different from what the plane is actually doing and as a result why video judging may be a bad idea.  I think the human eye pretty much disregards the skid in viewing a maneuver.  If it didn't a square flown to rulebook size would look like a round with flat spots.  We see the vertical shape of the fuselage and our brain filters out the rest.  I think that the experiment with the canard may have changes the way that the plane behaved in the skid that it appeared to be vertical sooner that it would be without the canard but overall about the same size if you include the skid if that makes any sense.  It also explains why Gieseke convinced me back in the late 70's that my plane should have a long white stripe down the fuselage center line.  "It makes the flats look longer...".  Now we have the technology to know why.

Thanks Keith - Ken

PS - another thing I noticed in your tracings which I do not have the ability to duplicate is that there is a definite skid going into the corner before the plane actually starts to rise from the horizontal plain.  With the canard directly behind the prop on my plane I was getting a definite thrust vectoring effect and the nose came up immediately and reached vertical very quickly but the skid was about the same.  Bottom line, can see that I was not doing much tighter corners, they just appeared that way.  Also speed made a huge difference.  I had very strong breaking from the timer making the last corners mush easier to turn.

[Tim Wescott]

I used the Blender Game Engine ... and did each keyframe like that, manually rotating each axis  around each maneuver shape.

It may have taken me less time to do it that way -- I've got the whole pattern encoded programmatically.  On the minus side, it took me months of evenings and weekends.  On the plus side the maneuvers will be easy to change -- things like turn radius and the size of squares are already in most of them.

Once I'm showing a "perfect" pattern, the goal is to show a pattern that's flawed in various typically-human ways.

I can remember that Igor noted at that time that there was no skidding into the corners.

Noted.  Thinking about making that happen just makes my brain hurt -- but noted.

There was one "funny" at the top of the vertical 8 in my sim that had something to do with Blender's animation system when the axes change around. I could not fix that, and put it down to "turbulence"!

If you had Blender set up to use Euler angles, that turbulence is baked into the underlying math: trying to represent three-dimensional angles with three variables inevitably leads to glitches somewhere on the sphere.  With Euler angles, it's at the poles -- so if you'd tried to do underhead eights you'd have had the same problem directly below the pilot's feet.

https://en.wikipedia.org/wiki/Hairy_ball_theorem

Just btw, I see that you "fly" the model in your squares with the tops following the 45° parallel line. I looked at it frame by frame. My CL Sim does the same, and we put that in the F2B rules at that time. I understand that many feel/believe that this is how the rules should specify this line....BUT, no pilot actually flies it that way. They all fly it as a spherical straight line or great circle path. If you look at the traced flight path's in Alberto Solera, and Andrey's tracker program, you will see this clearly. I've had a hard time trying to convince the majority of pilots and judges about this in the F2B rules, but I guess that just as before, it will take some time!

Hmm -- I can vary that.  I do that with all of the "level" 45-degree lines -- squares, eights, and the clover.

An unstated goal in all of this is that when I start feeling this pattern is close to perfection, then I'll shop it around to be judged and critiqued, and see if there's a consensus on how to make it better.

[Tim Wescott]

Ken:

Interesting point about skidding and the advisability of video judging -- I think we'd want to approach video judging very carefully.  My overall argument -- which encompasses what you said -- is that we judge what looks pretty, and computers don't do "pretty".  If we did embark on some giant AI training adventure to train video judges, the best we could come up with is some average of what judges think is pretty now, with oddball quirks that may or may not be pretty to us.

That would mean that in future, any innovation would be in the direction of exploiting those quirks, which may well lead how the pattern is flow in ugly directions.

I saw your point about Bob Giseke and graphics on the plane -- that all-white ghost plane really needs some help in that department; it's on the list when I tire of trying to make a realistic background.

[Keith Renecle]

When I redid my calculations including the skid I found your numbers to be more realistic.  If they are then that means that out of a 54' arc (45 degrees on 70' lines) there are 32' of corners and 22' of flat.  To me this explains why what we see is so different from what the plane is actually doing and as a result why video judging may be a bad idea.  I think the human eye pretty much disregards the skid in viewing a maneuver.  If it didn't a square flown to rulebook size would look like a round with flat spots.  We see the vertical shape of the fuselage and our brain filters out the rest.  I think that the experiment with the canard may have changes the way that the plane behaved in the skid that it appeared to be vertical sooner that it would be without the canard but overall about the same size if you include the skid if that makes any sense.  It also explains why Gieseke convinced me back in the late 70's that my plane should have a long white stripe down the fuselage center line.  "It makes the flats look longer...".  Now we have the technology to know why.

Thanks Keith - Ken

PS - another thing I noticed in your tracings which I do not have the ability to duplicate is that there is a definite skid going into the corner before the plane actually starts to rise from the horizontal plain.  With the canard directly behind the prop on my plane I was getting a definite thrust vectoring effect and the nose came up immediately and reached vertical very quickly but the skid was about the same.  Bottom line, can see that I was not doing much tighter corners, they just appeared that way.  Also speed made a huge difference.  I had very strong breaking from the timer making the last corners mush easier to turn.

Thanks Ken, it is a fascinating subject and one of the many reasons that we keep coming back for more! It is never boring! I enjoyed your comments on the canard. I have had the pleasure of a little stick time on two Long Eze's with pilot friends, and the canards do feel different. They're wonderful planes to fly.

I would personally love a digital tracker but not with a camera. If camera's are used then you really need multiple camera's to get accurate shape recordings. It would not readily be accepted for judging, even if someone could write the code to score the pattern, but for judges training, it would be invaluable.

[Keith Renecle]

It may have taken me less time to do it that way -- I've got the whole pattern encoded programmatically.  On the minus side, it took me months of evenings and weekends.  On the plus side the maneuvers will be easy to change -- things like turn radius and the size of squares are already in most of them.

Once I'm showing a "perfect" pattern, the goal is to show a pattern that's flawed in various typically-human ways.

Noted.  Thinking about making that happen just makes my brain hurt -- but noted.

If you had Blender set up to use Euler angles, that turbulence is baked into the underlying math: trying to represent three-dimensional angles with three variables inevitably leads to glitches somewhere on the sphere.  With Euler angles, it's at the poles -- so if you'd tried to do underhead eights you'd have had the same problem directly below the pilot's feet.

https://en.wikipedia.org/wiki/Hairy_ball_theorem

Hmm -- I can vary that.  I do that with all of the "level" 45-degree lines -- squares, eights, and the clover.

An unstated goal in all of this is that when I start feeling this pattern is close to perfection, then I'll shop it around to be judged and critiqued, and see if there's a consensus on how to make it better.

Thanks for the Euler explanation and the Hairy Ball Theorem.

The 45° "level" story was an old contentious point when I started to challenge the F2B rules. What got me going was that I could not construct the clover on a sphere by following the rules step by step. I started asking loads of questions and got told that I was sniffing too much ether. I flew stunt diesels then! I eventually got Wild Bill Netzeband who put me onto Pete Soule who had done a thorough math evaluation of how a 4-Leaf clover could fit on a spherical surface. My math is fairly useless, but Pete explained each step to me and explained that the present US and F2B rules are jokes. This is where the 40.88°, 42° and 45° horizontal joining lines all come from. As he showed the only way that you can draw a 4-loop clover with equal size loops and one loop touching 2 others, is with spherical straight lines or great circle paths. I unfortunately lost Pete's article when my pc at that time was stolen. I am rather chuffed when I managed to convince the F2B guys to pull up vertical directly to loop 1 to start the clover, because that gets rid of that old argument of what the correct angle is to get going on the clover.

I did think some years ago that the tops of the square loops should be at that constant 45°, or parallel to the ground, but after seeing and checking carefully, it is just not possible to fly that minor circle curve in the short time between the corners.

[Tim Wescott]

I would personally love a digital tracker but not with a camera. If camera's are used then you really need multiple camera's to get accurate shape recordings. It would not readily be accepted for judging, even if someone could write the code to score the pattern, but for judges training, it would be invaluable.

A digital tracker for coaching or self-coaching would be great, whether with cameras or IMU.  Judging -- it would change the event too much.