WWII BAD-T V: The AI Strikes Back - BD AI Dogfight Tournament

[OmegaForce]

49 minutes ago, Pds314 said:

7:07... what even just happened?

That was a first... It's as if both pilots cooperated together for a suicidal ram.

[Pds314]

2 hours ago, OmegaForce said:

That was a first... It's as if both pilots cooperated together for a suicidal ram.

And then collided into the third...

[Pds314]

I'm tempted for the next BAD-T-related tournament to make a tank. I don't think I can make it as tough as when part density abuse was easier, but maybe something like my DJA tank. It was pretty tough. Albeit very overweight at 9 tonnes.

 

Back in BADT-IV, there were some ways to get part health up to much higher levels. You could strafe them on the ground with minimal damage.

 

Or shoot the relatively soft wing panels to bits and they'd still fly on spars and whatever was left of the wings.

[sturmhauke]

7 hours ago, Pds314 said:

7:07... what even just happened?

The Krakenhound pilots were both annoyed at their planes' peashooters and decided to engage ramming mode at the same time.

[SuicidalInsanity]

9 hours ago, Pds314 said:

7:07... what even just happened?

Double Crash Cockpit Crush K.O!
As for how it happened, your guess is as good as mine, one of those 1-in-a-million chances, but the sequence of events is Krakenhound 1 rear ends the AeroImperial - this does minor damage to the AeroImp, but crushes the nose of the Krakenhound. Pilot is fine, plane is controllable, but all weapons were smashed off. The AeroImp then bounces off, into the cockpit of the second Krakenhound, crushing the pilot. To add insult to injury, the AeroImp survived with minimal damage. My only regret is that I cut the recording there - the AeroImp managed to circle around and rammed one of the Krakenhounds a second time before the 'hounds crashed.

[Jeb-head-mug kerman]

OMG 7;07, that's just insane....

[SuicidalInsanity]

And so it ends.
Match 16, the final match of BAD-T V, between @aleksey444's White Collar Criminal and @Dwerto's PD-2 Honeybee; two capable and agile fighters that have proven themselves thus far, both built around similar design philosophy, that now come head to head in a not-quite mirror match to determine the champion:

Congratulations, winner (and second place too); and to everyone else, regardless of how good (or bad) your craft did, thank you for participating and I hope enjoyed the tourney, both building your craft as well as watching them compete.

Spoiler

Final Standings:
First Tier:

Spoiler

Champion: Dwerto's PD-2 Honeybee
Second Place: aleksey444's AF-735 White Collar Criminal
Third Place: Bot Julius' AeroImperial

Honorable Mention - Pds314's La-9 DPRK - Tier 1 Battle Royale winner

Second Tier:

Spoiler

Champion: Alioth81's AL-25 Vulture
Second Place: OmegaForce's AFC-88 Thalo
Third Place: dundun92's Tallyhawk Mk XXVIa

Honorable mention - Box of Stardust's FB-97B-5 Sea Fairy - Tier 2 Battle Royale winner

 

Edited December 19, 2019 by SuicidalInsanity

[Jeb-head-mug kerman]

17 hours ago, SuicidalInsanity said:

Champion: Dwerto's PD-2 Honeybee

Congratulations! That Honeybee is just incredibly deadly! I really hope there will be a sequel, @SuicidalInsanity, as this was a very fun challenge.

[aleksey444]

I’m going to sound like a sore loser, but I’ll say it anyway.

This result is nonsense.

I ran this battle 10 times.  My plane won 7 out of 10.  The Honeybee is a decent plane, but it is not better than the White Collar Criminal.  It is not even its equal.  WCC outperforms it in just about every way.  WCC is faster (much better TWR), has much better pitch rate, and is about as or even more accurate.  About the only metric in which the Honeybee is better is roll rate.

I uploaded the White Collar Criminal to kerbalx.  The Honeybee is also still available.  If you think I sound arrogant, or if you don’t believe the previous paragraph is true, or you’re just curious – download both craft files and see for yourself.

There is some weirdness in the match itself, too.  Is the first battle supposed to have the camera follow the WCC?  Then why the switch to the Honeybee?  I noticed in my own runs, every time the camera switches to another airplane, it causes the AI to stutter for a moment.  It makes a chasing plane more likely to miss, a chased plane more likely to get shot, and a diving plane more likely to not recover in time.  It’s probably why a WCC crashed into the ground at 1:04.

Second, what happened to a WCC off-camera in the third battle?  At 11:11, you can see it shooting down a Honeybee.  The WCC is undamaged, it’s flying at high altitude, and it’s turning away from the debris field.  It’s a perfectly healthy airplane.  And then at 11:52 is just crashes?  Why?  How?  This is not a 2nd tier plane – it doesn’t just crash.

Also, by the time the final comes, the two remaining contenders are usually going to be somewhat closely matched – they are, by definition, the two best planes of the competition.  The Honeybee is indeed a good plane.  The game is not deterministic, there is some randomness in the process.  Which is probably why earlier BAD-T iterations had 5-match final battles – to offset some of that randomness – to find the better plane instead of the luckier one.

The battle is supposed to be a representation of the competing airplanes' capabilities.  And this one just ain't.  It is, in my sincerest opinion, garbage.

When BAD-T VI rolls around, and people are looking for a sparring partner to test their new designs, the smart ones will use the WCC.  And if I bother to participate again, I’ll probably just pull a sturmhauke and resubmit the same plane again.  Why spend a few weeks building and testing, putting real-life stuff on hold, when I can just reuse some of my old crap.  I might even do better the next time!

 

Edited December 19, 2019 by aleksey444

[sturmhauke]

BDA doesn't always behave the same way on different computers. I've observed this in other competitions as well. One plane I submitted for a Jolly Roger PFC tournament didn't fly properly on his computer, but it was fine on mine. He offered to edit the control surface settings, and then it flew ok on his, but was not great on mine. Maybe camera switching has an effect, but I don't think that's the whole story. I understand that it's frustrating, but unless you want to debug the BDA code I don't think there's much that can be done about it.

[tetryds]

@aleksey444 when I created BAD-T, it became clear that the matches are definitely nowhere near being deterministic, neither does war.

Sometimes it will be quite frustrating indeed, but I would not get attached to that.

We run best of three to filter out luck sniper headshots from 6km away being the dominant factor, but finals were run best of 5 because they tend to be way nicer and more exciting, not to filter randomness.

If your craft lost despite being better, and the match execution affected it, that does not matter the slightliest.

I believe that everyone here trusts @SuicidalInsanity just like I do, and whichever effect he has on the match has the only purpose to get better angles and a nice to watch experience for everyone, making it fully random and unbiased.

The best doesn't always win.

[SuicidalInsanity]

It seems there is some contention over the results of the final match.
 

4 hours ago, aleksey444 said:

There is some weirdness in the match itself, too.  Is the first battle supposed to have the camera follow the WCC?  Then why the switch to the Honeybee?

This is a goof on my part, due to filming the matches over two different days. In light of the deviation from my usual pattern of team 1 focus, team 2 focus, what ever has the best wide angle coverage focus, I recorded a  second take of the first bout. This had the same end score and three times the length, so I went with the first take.

4 hours ago, aleksey444 said:

Second, what happened to a WCC off-camera in the third battle?  At 11:11, you can see it shooting down a Honeybee.  The WCC is undamaged, it’s flying at high altitude, and it’s turning away from the debris field.  It’s a perfectly healthy airplane.  And then at 11:52 is just crashes?  Why?  How?  This is not a 2nd tier plane – it doesn’t just crash.

Since it was offscreen I cannot give a definitive answer. Best guess, based on what I know of the BDAc AI, both in code and observed behavior, is either A) the AI remained target fixated on the wreckage of the Honeybee and followed it into the ground, or B), the AI went into a dive at the end of the climb after shooting down the PD-2, and dove into the ground due to the AI not really having any controls to pull out of a dive unless the craft crosses the min alt threshold, at which point the plane GLOC'd the pilot and crashed (or had too much speed and couldn't pull up in time). While the WCC isn't a second tier plane and doesn't ordinarily GLOC the pilot, it does ride the edge of the pilot's G-tolerance at times, so a GLOC after a long dive isn't inconceivable.

4 hours ago, aleksey444 said:

I noticed in my own runs, every time the camera switches to another airplane, it causes the AI to stutter for a moment.  It makes a chasing plane more likely to miss, a chased plane more likely to get shot, and a diving plane more likely to not recover in time.  It’s probably why a WCC crashed into the ground at 1:04.
The battle is supposed to be a representation of the competing airplanes' capabilities.  And this one just ain't.  It is, in my sincerest opinion, garbage.

I went and re-ran the final match 9 times under near-identical* conditions to the original 3 matches (partially to investigate, partially to satisfy my curiosity on overall outcome, partially for the purposes of adjudication, since this is the first time a match has been contested in this manner). Not how I was planning on spending my evening, but c'est la vie. 3 bouts focused on a WCC with no vessel switching, 3 bouts focused on a PD-2 with no vessel switching, and 3 bouts with vessel switching every 30 seconds. Final score the PD-2 won 6-3; PD-2 scored 13 kills to the WCC 9. The bouts without vessel switching had a similar WCC crash - incoming fire from the PD-2s is causing it to dive and evade. Usually it pulled up in time, occasionally not.
*These weren't official matches and the UI was left on, mainly to investigate GLOC. They were recorded, though, to ensure that RAM and CPU overhead were consistent with the official bouts. If people want to see an extended Finals reel I can put them up in some form.

I do not interfere/intervene/modify/or otherwise influence their outcomes. Like Sturmhauke said, sometimes results differ between different hardware. I simply run the matches to provide a uniform, consistent playing field for all entrants to ensure all craft compete under the same, as near-identical-as-I-can-make conditions.

I sympathize, I've had my share of BDA contests where I make something that can trounce the leaderboard, only to watch as AI quibbles and RNG decide that no, it's going to eat all the missiles(or dirt) and fail to score a single kill in the official matches. And yes, it's annoying to see my time and effort fail; I suspect that's why tetryds set up BAD-T with a double-blind entry system. You build the best craft you can, and not have to worry about someone else coming along and building something tailored to counter it. You have fun building and designing an entry (and if you don't, then why play KSP?), you submit it, then sit back and watch the show. You don't know how your craft will ultimately perform against the competition, but no one else will, either. There's a reason the rules include "...but for the most part, this tournament is aimed at having fun, watching submitted craft fight bravely and die gloriously". The WCC did both, with aplomb.

 

Edited December 20, 2019 by SuicidalInsanity

[Dwerto]

Wow, I just got around to watching the final match and I'm happy and a little surprised to see my plane won. It seems the match was very close and both planes had positive and negative attributes.

I haven't personally tested the white collar criminal, but from the stats and matches I've seen these are some things I've noticed:

• The WCC is lighter, has lower wing loading, a better sustained turn rate and much better energy retention. 
• The PD-2 has a faster roll rate, is a smaller target, and possibly has better instantaneous turn performance

In my testing I found that roll rate matters A LOT, partly because the AI often gets into rolling scissors type turning battles. A plane with poor sustained turning and high roll rate often beats a plane with a high sustained and a low roll rate. I also noticed that the White collar criminal often seems to overshoot. Maybe something to do with really good energy retention without a super high roll rate? There is definitely randomness in these fights, and crashes happen. I actually gave my AI more aggressive setting that I knew were more likely to cause crashes because it was still better than the safer settings. I think it's totally conceivable that different planes could lose to the WCC and beat the PD-2 or vice versa. for example some planes have the ability to take down a flying tank type build while others can't. I just had to hope my planes flew well during the actual fights, and that I got opponents my plane could actually defeat. I guess luck was somehow in my favor and it all worked out.

Overall this tournament was super fun. I hope everyone enjoyed it and I look forward to the next one whenever it happens.

[sturmhauke]

1 hour ago, Dwerto said:

a flying tank type build

[Alioth81]

I am currently traveling and will be until end if January but I watched the matches when I had WiFi.

Thanks @SuicidalInsanity for running the tournament. I hope I am back in time for the next one.

Out of curiosity could you run a fight with my flak fighter (75mm). I never really finished and tested it under true battle conditions and wondered whether it would have been viable at all.

 

[Pds314]

On 12/20/2019 at 10:41 PM, Dwerto said:

Wow, I just got around to watching the final match and I'm happy and a little surprised to see my plane won. It seems the match was very close and both planes had positive and negative attributes.

I haven't personally tested the white collar criminal, but from the stats and matches I've seen these are some things I've noticed:

• The WCC is lighter, has lower wing loading, a better sustained turn rate and much better energy retention. 
• The PD-2 has a faster roll rate, is a smaller target, and possibly has better instantaneous turn performance

In my testing I found that roll rate matters A LOT, partly because the AI often gets into rolling scissors type turning battles. A plane with poor sustained turning and high roll rate often beats a plane with a high sustained and a low roll rate. I also noticed that the White collar criminal often seems to overshoot. Maybe something to do with really good energy retention without a super high roll rate? There is definitely randomness in these fights, and crashes happen. I actually gave my AI more aggressive setting that I knew were more likely to cause crashes because it was still better than the safer settings. I think it's totally conceivable that different planes could lose to the WCC and beat the PD-2 or vice versa. for example some planes have the ability to take down a flying tank type build while others can't. I just had to hope my planes flew well during the actual fights, and that I got opponents my plane could actually defeat. I guess luck was somehow in my favor and it all worked out.

Overall this tournament was super fun. I hope everyone enjoyed it and I look forward to the next one whenever it happens.

Yes the instananeous turn rate and high roll rate is part of why my little tiny La-9 performed as well as it did. I think the combination of the two actually prevents the sustained turning fights that planes with big wings can do well at and instead gives the big wings a situation where they're just gonna rip if they follow it, and it doesn't matter anyway because it's rolled 140 degrees off.

Edited December 29, 2019 by Pds314

[SuicidalInsanity]

On 12/21/2019 at 2:09 AM, Alioth81 said:

Out of curiosity could you run a fight with my flak fighter (75mm). I never really finished and tested it under true battle conditions and wondered whether it would have been viable at all.

BAD-T V Bonus Video 3:

Performance of the Rhino isn't quite what it was in your test vid, mainly due to my copy of BDArmory still having a 500m minimum arming distance for the flak rounds.

[Alioth81]

Thanks for running it. Very satisfying to watch a 75mm kill even if it is friendly fire and when it misses it makes nice new year fireworks.

I agree it would be more on the fun then competition side but without the 500m arming distance the 75mm would have been a very solid choice for a different type of warplane.

With arming distance it could have been interesting on an even slower (opponent spends more time at 500m plus) plane. More agility and more optimization (probably smaller explosion distance setting to maximize damage) could have helped too I guess.

Wish you all a good end of 2019 and a great start into 2020

Edited January 8, 2020 by Alioth81
Corrected autocorrected spelling from my phone