WW2 BAD-T 2 - BDA AI Dogfight Tournament [SEMIFINALS]

[Doke]

I've got a few screenies of various designs I'll share when I have a moment. Most of them are much more conventional than what others have been showing, but oh well.

Till then, I have a serious question: I understand (more or less) the purpose of an elliptical wing, but what's the point of the 'reverse elliptical' wings you see on some prop aircraft? I'm thinking especially of Soviet fighters, particularly the Yak-3, where the wing taper actually decreases toward the tip. I remember reading once, long ago, that there was a specific aerodynamic reason for this configuration, but I can't for the life of me recall what it was.

[Wjolcz]

I still haven't figured out how to balance my fighter. If I were a ministry of defence I would cancel that project a long time ago. Will try those BD improvements ferram is talking about. Maybe that will help.

Hope I don't run out of time.

Edited February 29, 2016 by Veeltch

[tetryds]

@Veeltch No one is going to run out of time here, that should be the last thing to worry about.

[Doke]

Some pictures:

 

[Pds314]

So, um. I apparently made a plane with a 4-meter wing?-span that has superior lift characteristics to most of my fighter designs and can easily kill the dummy.

This seems like it's worth experimenting with.

P.S. with the second image, the clock should read 1:10. As in, that's the time between switching teams and a kill, starting both at the same place facing opposite directions.

Edited March 1, 2016 by Pds314

[tetryds]

I didn't want to explicit piston monoplanes but you should not abuse multi winged designs too much.

(In fact after a point every airplane was a monoplane so I could technically enforce it)

[Pds314]

3 minutes ago, tetryds said:

I didn't want to explicit piston monoplanes but you should not abuse multi winged designs too much.

(In fact after a point every airplane was a monoplane so I could technically enforce it)

Ah. So I guess making a flying wafflemaker is probably not in the spirit of the competition..

[tetryds]

It's funny that it's even viable, wing interaction should kill its perfornance.

Where is @ferram4 with the wing overhaul?

[Pds314]

24 minutes ago, tetryds said:

It's funny that it's even viable, wing interaction should kill its perfornance.

Where is @ferram4 with the wing overhaul?

Well, I did intentionally space out the wings to give it at least SOME hope. Moving them too close definitely kills performance.

The main point of the design was to reduce induced drag while maintaining low span.

Though I suppose this could be...

Abused:
 

Supermaneuverability:

 

 

Although that particular aircraft actually has terribad performance besides that.

Edited March 1, 2016 by Pds314

[Pds314]

So, for ex, this abomination does actually fail at any sort of performance.

 

 

[gridghost]

I like the designs tho  Both innovative and creative, Even if terribad performance...

[Pds314]

15 minutes ago, gridghost said:

I like the designs tho  Both innovative and creative, Even if terribad performance...

The orange one is a viable fighter. The white boxed multiplanes are extremely awful.

Though they're not totally original. The orange is based on an idea that periodically gets thrown around called annular wings. In theory, you can get extremely high span efficiency with them. The front lower Beechcraft-staggered wing actually has to be within plus or minus a few cm, and actually improves performance in every way. Move it 6" forward and the maximum L/D drops from 18 or 20 down to 12. Unsweep it and it drops to 10.


The white ones are inspired by this, uh, "creative" design: 

I'd imagine that that flying box though would be FORMIDABLE as a fighter. I'm having trouble imagining a bullet causing any wing damage whatsoever. Unfortunately, this doesn't apply in KSP because parts are always attached in a tree structure.

Edited March 1, 2016 by Pds314

[Pds314]

Hmm. I should try making evolution design me a fighter. It'd be massively time-consuming, preferably automated and distributed on multiple PCs, but I don't really know how to mod well enough to do that. Basically, make a super-generic fighter and use a genetic code to store part lengths, settings and offsets, as well as AI behavior. Survival is determined by whether the craft wins a 2v2.

[gridghost]

That seems a bit excessive 

[Pds314]

4 minutes ago, gridghost said:

That seems a bit excessive 

The orange plane, the real massively multiplaned box, or the Darwin fighters?

Edited March 1, 2016 by Pds314

[gridghost]

9 minutes ago, Pds314 said:

The orange plane, the real massively multiplaned box, or the Darwin fighters?

The real massively multiplaned box... 

The idea of a darwin fighter seems intresting and the orange plane, creative and cool.

I'm actually doing something like the darwin fighter, making a lot of potential designs w basic control and AI setup, fights them in 1 vs 1 and keeps the ones that shows promise and then tweak the hell out of them

I think i have 6 different designs at the moment... (more if you count heavy and light versions)

[Pds314]

4 minutes ago, gridghost said:

The real massively multiplaned box... 

The idea of a darwin fighter seems intresting and the orange plane, creative and cool.

I'm actually doing something like the darwin fighter, making a lot of potential designs w basic control and AI setup, fights them in 1 vs 1 and keeps the ones that shows promise and then tweak the hell out of them

I think i have 6 different designs at the moment... (more if you count heavy and light versions)

Oh, yeah. Well I'm more-or-less doing that. I've got... let's see.. at least 7 designs from the ground up so far, although a few of them are either obsolete or scrapped, and most of them have several variants, especially because, the way I do the cockpits, converting to an open/closed cockpit is really, really easy (though I like the closed ones for their aerodynamics, appearance, and potentially protection).

Edited March 1, 2016 by Pds314

[Doke]

The problem with a 'Darwin fighter' is that there's no way of controlling for AI and control deflection. I find that every plane needs different AI and control settings, and small changes to either can make a big difference to performance.

[Pds314]

3 minutes ago, Doke said:

The problem with a 'Darwin fighter' is that there's no way of controlling for AI and control deflection. I find that every plane needs different AI and control settings, and small changes to either can make a big difference to performance.

Just make the AI settings subject to evolution? I don't see why they shouldn't be. Bad AI = dead bird.

And make AI settings adjust slowly, like everything else. Make radical change unlikely in one generation.

Edited March 1, 2016 by Pds314

[tetryds]

@Pds314 Unfortunately even if you programmed such a plugin you would face a sad truth.

Apart from local minimums another big problem you would face would be the way you select and evolve these airplanes.

Evolution works super well against environments and/or clear rules, but evolving an airplane that fights the best against itself or another specific group or type of airplane would not mean you will behave well against every other possible airplane.

Another weird fact is that the "best" is not always going to win, sometimes you will get sniped by a barelly flying brick with some heavy weaponry over and over again.

This could both screw up your evolution or frustrate you in a battle.

But I cannot recommend tweaking and retesting more, engineer your own desin, tweak it against something like the dummy, then manually evolve its AI, that should make sure it's at least piloted as well as it can.

[Pds314]

18 minutes ago, tetryds said:

@Pds314 Unfortunately even if you programmed such a plugin you would face a sad truth.

Apart from local minimums another big problem you would face would be the way you select and evolve these airplanes.

Evolution works super well against environments and/or clear rules, but evolving an airplane that fights the best against itself or another specific group or type of airplane would not mean you will behave well against every other possible airplane.

Another weird fact is that the "best" is not always going to win, sometimes you will get sniped by a barelly flying brick with some heavy weaponry over and over again.

This could both screw up your evolution or frustrate you in a battle.

But I cannot recommend tweaking and retesting more, engineer your own desin, tweak it against something like the dummy, then manually evolve its AI, that should make sure it's at least piloted as well as it can.

If there's a sufficient population in each generation, then it's acceptable for several lucky kills to occur. There's also no guarantee that only one design would evolve. It's quite possible that with a sufficiently large population, you could end up with divergent evolution producing multiple archetypes.

As for local minima, with something as complex as an airplane with lots of avenues for mutation (I'm currently thinking of using procedural wing segments in a branching tree structure as the body, so there will be literally at least a hundred values which can be modified, including the addition of totally new parts to nearly anywhere on the plane or the total removal of parts), I don't really see it as being particularly likely that every single design will get trapped in a local minima early on. There are probably evolutionary routes out of most near-minima.

Edited March 1, 2016 by Pds314

[KamikazeF0X]

Is it possible to use bombs or rockets as a ballast ?

[Pds314]

9 minutes ago, KamikazeF0X said:

Is it possible to use bombs or rockets as a ballast ?

The dummy uses bombs, though what exactly they're for besides bombing I'm not sure.

[tetryds]

@Pds314 that is exactly the problem, there are so many possibilities that if you were to do that it would end up being computationally unfeasible.

This was never ever done in real life for a good reason.

You can get away with evolving things towards one or a few goals, especially when they are simple, but this is much more complex than that.

Also, changing that little does not improve it enough so that you can really compare, your best design will eliminate all others right away and keep fighting against itself, and on this case the output of the battle is pretty much random, you would need to change way too much to notice differences.

So it would be neat to evolve the AI, or a specific part of the airplane, but you would have to judge by yourself or develop something to analyze the behavior, win/loss can only be used for two totally different designs.

That is what makes BAD-T so nice, the best doesn't always win.

Another strategy would be develop two designs yourself and improve one to defeat the other, when one gets stuck and always loses, experiment until something defeats the previous, and keep doing that.

The hard part is making a new one to defeat the previous but you learn with time and will learn how to attack its weaknesses.

 

@KamikazeF0X bombs yes, but not missiles, who knows when the AI will start using them against other airplanes?

[gridghost]

2 minutes ago, tetryds said:

Another strategy would be develop two designs yourself and improve one to defeat the other, when one gets stuck and always loses, experiment until something defeats the previous, and keep doing that.

The hard part is making a new one to defeat the previous but you learn with time and will learn how to attack the its weaknesses.

Ya, that's pretty much what I'm doing at the moment... but I have two or three contestants right now.