WW2 BAD-T IV BDAc AI Dogfight Tournament

[Pds314]

On 11/1/2018 at 5:13 PM, SuicidalInsanity said:

Round 3, between @Van Disaster's D.A.C. Deimos X and dundun92 @dundun93's P-009: 

 

Two four-tonne dry planes vs each other.

 

Well, these were really evenly matched. I have to say the P-009 seems a bit tougher and a bit better in instantaneous turn rate.

Edited November 4, 2018 by Pds314

[Pds314]

So far, power/weight at dry mass seems to be the determining factor whether there is any difference.

 

Although wing/weight could also work as a theory. 

Edited November 4, 2018 by Pds314

[aleksey444]

1 hour ago, Pds314 said:

So far, power/weight at dry mass seems to be the determining factor whether there is any difference.

 

Although wing/weight could also work as a theory. 

It's all of these and more.

The weight is the wet weight.  If your plane ever reaches dry weight, it means you've run out of fuel and ammo.  If that happens, you lose.  You do shed some weight throughout the match, by burning fuel and shooting bullets, but I would guess that the weight at the end of the match is closer to the wet weight you started with than to the dry weight.  In any case, you start out with the wet weight, and that's the weight you have to push as you are slowly shedding some of it.

It is also the power of your engine.  The more power you have, and the less (wet) weight, the higher your Thrust-to-Weight Ratio (TWR).  All other things being equal, the plane with a higher TWR will be able to catch up to the plane with a lower TWR if behind, and run away quicker if in front.

It is also the wing size. Larger wings allow you to do two things.  First, they let you turn better.  You can put huge elevators in the back, but without adequately-sized wings, your plane will pitch hard but then snap back.  Large wings allow the plane to better keep its orientation after a pitch.  The second thing large wings provide is a smaller Area of Attack (AoA) needed to maintain level flight.  If you want your plane to fly perfectly horizontal, you need to point it a few degrees up.  Just how many degrees depends on wing size.  The larger the wings, the smaller AoA that's needed.  This tends to make the plane more accurate.

It is also the rate at with your plane pitches, rolls, and (to a lesser extent) yaws.  Generally, faster is better, provided it doesn't make the plane too shaky.

However, the most important thing, by far, is overall accuracy.  That is to say, how well the plane is able to keep its nose on the target.  Low AoA for level flight is one, but not the only thing that influences it.  For example, if the plane, when rolling, also pitches a lot, that decreases the accuracy.  Or, to a lesser degree, if the plane pitches/rolls when yawing - although yawing is not as important. I can't really explain how to make it more accurate, other than by trying it out and tuning -- basically by trial and error. If somebody has a more systematic approach, I would honestly love to know.

Guns are also important.  It seems that most people picked the 15mm guns.  Most likely they picked one of the MG 151's. They look great on paper, because they have the highest muzzle velocity (960 m/s) and decent rate of fire (700 or 710 rounds/min).  But in practice, they do not seem to do much damage.  A part can survive many 15mm hits.  In comparison, it takes 1-2 hits with a 30mm to destroy just about any part (at least the ones with 500 hit points, and probably 1000hp parts as well), and maybe 2-3 20mm hits.  Most hits happen at close range, say about 500 meters, so muzzle velocity is not as important. The rate of fire for 20mm is either the same (700/710 for Hispanos and 700 for 151) or even higher (800 for ShVAKS).  The rate of fire of the 30mm is comparable at 650 rounds/min.

Another factor is redundancy.  You want your airplane to be able to lose a few parts and still be able to fly.  You actually did a decent job there - I think your plane can fly with half of its wing blown off.

 

To give you some examples:

Both your airplane (Shadowbird Mk IV) and mine (Benign Lurker) have about the same wet weight.  If fact, yours is a little lighter at 3085kg than mine at 3131kg.  However, I have a more powerful engine.  Also, I suspect that because of how you constructed your cockpit, out of many small parts, your plane experiences more drag than mine.  The result is that I can easily catch up to you when behind, and once I overtake you, I can rapidly get away.

Another example is @dundun93's Tallyhawk Mk XIVb. It has huge wings.  I tested it, and it turns (pitches) faster than my plane. Much faster.  It is also faster in flight.  I can't explain why, but given two airplanes with the same TWR, one plane with one engine, and the other with two, the plane with two engines will accelerate faster.  Tallyhawk's only problem is that is not very accurate.  I tested it against my plane, and it wins about 50% of the time, although it's not a very reliable number because I tested only a few times.  But if it was more accurate, I suspect it would kill me closer to 100% of the time.

The plane I'm really looking forward to seeing in action, and the one I'm genuinely afraid of, is @Noir's Snowpiercer.  It's dry weight is exactly 5000kgs (as low as can be), and the wet weight is only 600kgs more (for comparison, my wet weight is 631kg more than the minimum 2500 kgs of dry weight).  It turns faster than me.  It accelerates faster than me.  It has powerful guns (20mm).  And it's pretty accurate.  I expect it to win 6:0 against the bot. Maybe I'll get lucky if I face it, but I think it's got a really good chance of winning the competition.

 

In short, building a good fighter is hard. There isn't any one or two things that makes it good. It's more like a dozen or two dozen things that need to be optimized to make it deadly.  Some things are easier to achieve than others.  Make your dry weight as close to the minimum as possible -- ideally it should be exactly 2500 or 5000 kgs.  Don't overload with fuel and ammo and keep the wet weight as low as you can.  Higher TWR is always good.  Larger wings are generally good (up to a point, I guess).  Two engines, for reasons I can't explain, seem to be better than one.  High turn rate is extremely important.  And accuracy is the most important, but I can't explain how to make it accurate by reducing it to some rule-of-thumb.  I did mine through trial and error, and some luck.

Edited November 4, 2018 by aleksey444

[sturmhauke]

I tested my plane against the Tallyhawk and Snowpiercer, and lost. Based on that, I'd rate my Krakenhound like so:

++Maneuverability

++Toughness (in some cases it could still fight with only one engine)

+Firing accuracy

-Speed

-Climb rate

--Firepower (1 15mm + 4 Brownings means it scores a lot of superficial hits)

[sturmhauke]

7 hours ago, aleksey444 said:

I can't explain why, but given two airplanes with the same TWR, one plane with one engine, and the other with two, the plane with two engines will accelerate faster.

I suspect it's because the two-engine plane has a better power/drag ratio.

[aleksey444]

17 minutes ago, sturmhauke said:

I suspect it's because the two-engine plane has a better power/drag ratio.

Yeah, that makes sense.  Its thrust would be 2x (assuming same engine), but it's surface area, volume, and number of parts (whatever it is that actually produces drag), while larger, would not twice as large.

[Guest]

1 hour ago, aleksey444 said:

Yeah, that makes sense.  Its thrust would be 2x (assuming same engine), but it's surface area, volume, and number of parts (whatever it is that actually produces drag), while larger, would not twice as large.

Yes I noticed that my single engine designs were noticably inferior to the tallyhawk, even though they had equivalent engines. 

8 hours ago, aleksey444 said:

Another example is @dundun93's

8 hours ago, aleksey444 said:

Tallyhawk Mk XIVb. It has huge wings.  I tested it, and it turns (pitches) faster than my plane. Much faster.  It is also faster in flight.  I can't explain why, but given two airplanes with the same TWR, one plane with one engine, and the other with two, the plane with two engines will accelerate faster.  Tallyhawk's only problem is that is not very accurate.

The accuracy issue may be because I angled one of the pairs of 15mms up by about 1.5°. I did that because it was consistently aiming below the target. Tweaking the AI solved that, but I decided to keep it. From my testing, the "spread" in the gun pattern helped to ensure more hits.

[Guest]

8 hours ago, aleksey444 said:

It seems that most people picked the 15mm guns

One reason I picked it was because of the fact that it takes a while to overheat. In my testing at least, overheating became a slight. Additionally, I set up a "gun stand", with several guns on a stand(in the different legal configs, e.g, 4x 15mm, 3x 23mm), aiming at a target of about 4 wing layers. In the testing, the 23 only did marginally better than the 15mm, and I figured that the higher ROF would equal it out in combat. The 30mm did a lot of damage, but the low ROF made it less effective than the 15mm in my combat testing. Plus, the high muzzle velocity can help in long range firing (which seemed to happen a lot in my testing), and helped in utilising "snapshot" opportunities

[Guest]

10 hours ago, aleksey444 said:

Another example is @dundun93's Tallyhawk Mk XIVb.

Everyone seems to mistake @dundun93 from @dundun92..... Its Dundun92 who made the tallyhawk.

4 hours ago, sturmhauke said:

I tested my plane against the Tallyhawk and Snowpiercer, and lost. Based on that, I'd rate my Krakenhound like so:

++Maneuverability

++Toughness (in some cases it could still fight with only one engine)

+Firing accuracy

-Speed

-Climb rate

--Firepower (1 15mm + 4 Brownings means it scores a lot of superficial hits)

That is hard to believe....

18 hours ago, Pds314 said:

Two four-tonne dry planes vs each other.

 

Well, these were really evenly matched. I have to say the P-009 seems a bit tougher and a bit better in instantaneous turn rate.

I gave it thick wings. The wings can withstand a head-on collision. I tried to make it maneuverable, but I'm a terrible FAR plane designer....

[Guest]

4 minutes ago, dundun93 said:

That is hard to believe....

Eh not really. It has the biggest wings, and two powerful engines.

[53miner53]

Random question: what happens if a plane runs out of one kind of ammo and is fast and maneuverable enough to keep the other plane within the minimum engage range of the guns that use the other type? Does it just come down to fuel at that point? 

[Guest]

13 minutes ago, 53miner53 said:

Random question: what happens if a plane runs out of one kind of ammo and is fast and maneuverable enough to keep the other plane within the minimum engage range of the guns that use the other type? Does it just come down to fuel at that point? 

Yep. I think so

[MightyDarkStar]

Shame to have missed it this year  I'll be there for the next one.

Is the infinite turning battle thing still present with the BDA AI? I'm in the mood for a big, heavy, powerful energy fighter but alas it's never seemed to work.

[Guest]

41 minutes ago, 53miner53 said:

Random question: what happens if a plane runs out of one kind of ammo and is fast and maneuverable enough to keep the other plane within the minimum engage range of the guns that use the other type? Does it just come down to fuel at that point? 

If you run out of ammo you loose IIRC.

16 minutes ago, MightyDarkStar said:

Shame to have missed it this year  I'll be there for the next one.

Is the infinite turning battle thing still present with the BDA AI? I'm in the mood for a big, heavy, powerful energy fighter but alas it's never seemed to work.

The AI will engage in extended turn-fights. But energy fighting isn't a really good strategy now.

[SuicidalInsanity]

3 hours ago, 53miner53 said:

Random question: what happens if a plane runs out of one kind of ammo and is fast and maneuverable enough to keep the other plane within the minimum engage range of the guns that use the other type? Does it just come down to fuel at that point? 

From my experience of BDA AI behavior, both planes then proceed to fly straight until one runs out of fuel/manages to outrun the other by 2.5km
The better question is is this the case for both planes on a team, and/or have both sides taken a loss and it's down to 1v1?
On one hand, I'm loathe to introduce rules changes mid-tourney, on the other hand, who wants to watch two planes practice parade formation for 10+ minutes?
Three ways around this I can think of: A): establish if the lead plane can reasonably be expected to not shake its pursuer - ie both planes have a similar top speed and cannot extend or overtake, and rule in favor of the pursuer on the grounds that if it wasn't for AI derpiness, the pursuer could reasonably be expected to shoot down the lead plane in this scenario, or B) rerun the match, which would likely end with the same final score, unless the craft are really close, ability-wise, but could easily end with running out of close-range ammo again.
Option C is I customize the AI engagement code to provide for this exception and allow it to use guns inside min range if no other option is available anymore.
Barring the adoption of one of the above, yeah, it would come down to fuel and a fastforwarded section.

[sturmhauke]

23 hours ago, dundun93 said:

That is hard to believe....

 

23 hours ago, dundun92 said:

Eh not really. It has the biggest wings, and two powerful engines.

I posted my .craft file after the tournament started. You're all welcome to try it out. https://kerbalx.com/sturmhauke/SAI-K9-Krakenhound

[SuicidalInsanity]

Round 4, between my IA-23 Caelus II and @sturmhauke's SAI K9 Krakenhound, or, 'Any Landing you Walk Away From', an introspective in three acts:

Edited November 6, 2018 by SuicidalInsanity

[sturmhauke]

Well that was embarrassing. I don't understand why one of my planes did a bellyflop though, it didn't appear damaged going into that dive.

[SuicidalInsanity]

It was doing a very steep dive at the time; the unplanned landing is likely either from the AI deciding evasion takes priority over pulling up to prevent a crash, or the K9's minimum altitude of 300m wasn't enough. At least the pilot survived?

[sturmhauke]

I tested it manually in a full-throttle, 90° dive, and was able to pull out at 300 meters. Must have been the evasion priority, or it got confused because of that hill.

Whatever, it obviously needs work for next year.

[SuicidalInsanity]

Since we've now seen it (and, I suppose, in the interests of adjudicator transparency), here's the IA-23 Caelus. And since I never got around to it earlier, here's the IA-22 Stribog from last tournament as well.

[SuicidalInsanity]

Did more tinkering with my personal copy of BDAc, and ended up with something that might be of use for in-flight tuning of competition craft:

Those of you who are interested my fork of the code's on GitHub.

Also got the last two battles of Round 1 recorded, with some surprising results, so expect those shortly.

[Noir]

1 hour ago, SuicidalInsanity said:

Also got the last two battles of Round 1 recorded, with some surprising results, so expect those shortly.

Calling it now - both bots won.

[Pds314]

On 11/6/2018 at 12:41 PM, sturmhauke said:

I tested it manually in a full-throttle, 90° dive, and was able to pull out at 300 meters. Must have been the evasion priority, or it got confused because of that hill.

Whatever, it obviously needs work for next year.

The thing is, the AI often gets itself into very bad positions and then doesn't apply full pitch-up immediately. All of my planes can pull out of a 90-degree dive from 500 meters, even the overweight underwinged Russian Bias, yet I needed to turn them up to 1000 meters minimum and they still crash without taking fatal damage maybe 1-2% of the time, and my heavy crashes even more often because it doesn't have that great a roll rate.

Edited November 8, 2018 by Pds314

[aleksey444]

17 hours ago, SuicidalInsanity said:

Also got the last two battles of Round 1 recorded, with some surprising results, so expect those shortly.

Guys, if you don't hear from me again, it's because the suspense has killed me.