A stupid hunch on that craptastic AI.

Ducimus · 05-21-06, 12:41 PM

This is probably going Waaaaayyy off the deepend, and i havent fully experimented to verify this behavior.. but.. its a hunch, and i wonder if anyone else has ever had similar thoughts.

The key is this, has anyone noticed a reduction of the AI's use of active sonar after reducing the active sonar cone? This is central to this stupid hunch of mine. and Admittidly this might be hard to verify because the AI is so unpredictable and wonky to begin with.

Here's my tinfoil hat, pulled it out of my ass, totally unsubstantiated, throw me in a rubber room with a vest with real long sleeves, theory.

Code:

- AI first listens passively.

- If sound bearing it recieves is [u]within its sonar cone[/u,] it starts active sonar.

- When active sonar bearing is receieved,  [u]THEN[/u] the AI changes course to intercept contact.

So heres the Picture:

Red represents the cone size most of us adjust for
blue represents the default.
Grey line indicates Passive sonar bearing.

Again, if passive bearing is within active sonar, start pinging.
If ping gets a return, change course.
If no ping, continue search.

Thats the hunch.

In everyones bid for realistic settings it could be that we've been nerfing the AI because it needs unrealistic sonar cones due a routine like this to function.

The developers of Ubisoft , im quite certain were not stupid. SH3 took years in development, and a 5 -10 minute google search will show what active sonar cones looked like in WW2, so on that thought, it would almost seem to me that the 90 degree cone, was deliberate.

I know, start throwing the shoes, im probably way off the deep end and i admit it, but its just a naggying feeling that this is whats going on.

Now if anyone expereinces the same behavior CONSISTANTLY regardless cone size, then im just blowing alot of hot air with this post
:rotfl:

CB.. · 05-21-06, 12:55 PM

sounds logical to me..."borrows tin hat"--

im guessing the AI might not even be that particular regarding the sensors..i'm guessing that it just uses the one with the longest range to find you---then switches to sonar (if the game is that well made) to get depth reading before making a DC run...
i would not be at all surpirsed if it doesn't actually use any sensors to get your depth..and that it in effect simply "cheats" getting depth info directly from the game engine in some manner---
i've been successfully blown out of the water by DD's with no active sonar on them at all..so how did it get the depth right...co-incidence??? how did it know what sort of depths to try? co-incidence?

it may be an assumption to even think the game follows any sort of realistic real world pattern...

but the end of the day the main problem with the AI is that it just doesn't co-operate terribly well (tho i still maintain the AI was great untill summat went wrong after a few hours of play---never to return --those creeping attacks were IMO real not imagined - i spent hours just watching them--as much as i tried to convince my self other wise..either way for me they simply stopped happening after maybe 10-20 hours gameplay never to return)

that and the woe full lack of sensor blindspot after a DC drop- makes the whole gig a bit of a lash up--

mind you i dunno---what you say sounds ok to me..

Ducimus · 05-21-06, 12:59 PM

im almost 100% positive the game "cheats" on its depth settings. When the AI does acutaly attack, it rarely misses with its default settings. Infact, when depth charges are dropped, i acutaly go deeper to get away from that depth. Assuming that the depth charges are "set" at whatever depth i was when they were first launched.

Of course, those depth charges still seem to manage to explode close, regardless of depth change

CB.. · 05-21-06, 01:09 PM

Quote:

Originally Posted by Ducimus

im almost 100% positive the game "cheats" on its depth settings. When the AI does acutaly attack, it rarely misses with its default settings. Infact, when depth charges are dropped, i acutaly go deeper to get away from that depth. Assuming that the depth charges are "set" at whatever depth i was when they were first launched.

Of course, those depth charges still seem to manage to explode close, regardless of depth change

and i agree they are rarely very far out depth wise-- in fact if they just learned to drop the DC's in front of you instead of directly over head- i'd be dead every time..
any sensible AI routine would allow for the time the DC take to reach the set depth--that would in it self make life far more interesting and tactical--and satisfying--

mind you again as default (again once the AI had inexplicably "broken"--i'd stake a lot on that being true by the way) in my game the DD's suddenly just couldn't find you at all---only surfacing would alert them to your presense and that with the game as stock no mods at all--

so you have one group with uber DD's one with utterly dumb DD's with no obvious explanation for the difference in behaviour--

maddening

there are no hidden registry entrys for AI behaviour any one know --i cant see any---older games allowed for the use of registry entrys to customise elements of the gameplay--there's nowt like that with SH3 am i right?

there's summat kooky going on some where--

Ducimus · 05-21-06, 01:23 PM

Well heres what got me thinking.

You'd agree that ELITE's are the most consistant in behavior, most aggressive, and usualy the ones that would cream you no?

Ive acutally had this hunch for awhile. Orginally i was modding my game leaving the sonar cone as default, only fiddeling with a min range, and then tried to add for more inaccuracies within the depth charges and hedge hogs themselves.

That was awile ago, and i forgot why i was doing it that way.

Over the last couple weeks i went back to fiddling with sonar. Made the cone 70 degrees, increased the min distance.. all the semi realistic stuff that mos of us do. Well, in my game ive added quite a few elite DE's.

You know.. those dreaded buckleys?

I have 3 ship convoys escorted wtih two elite DE's in the carribean after 42. Out of 4 convoy attacks, i had only ONE close call. All the other 3 attacks resulted in me silpping away without so much as one depth charge being dumped anywhere near me.

Thats when i remembered this hunch of mine that i forgot about.

caspofungin · 05-21-06, 01:40 PM

i' not so sure...

i've set up my ai_sensors file so that at silent running, i'm impossible to hear unless the escort is w/in a couple of hundred yards. nevertheless, i can start getting pinged when the escort is 1400-1500m away. of course, that doesn't mean the ai isn't cheating...

w/ respect to the ai getting the depth automatically -- maybe. but also consider the possibility that the info it gets from hydrophones isn't limited to bearing only (as irl) but gets bearing, elevation, and range -- allowing it to pinpoint your position in 3d space.

caspofungin · 05-21-06, 01:44 PM

also, re what cb said -- i agree completely -- it's probably got more to do w/ the coded ai routines than the sensors themselves.

CB.. · 05-21-06, 01:53 PM

it's a problem which ever way you approach it---on the one hand we definitely want some sort of dangerous response from the escorts-- on the other we don't want to killed every time----the area in between being ignored and being killed every time by the DD's is fairly small with not that much room for maneuver-- definitely using the crew rating entrys to see how any set up changes under different crew ratings has to be a good solid way to go---a overly uber set up might actually be spot with the crew rating reduced to 2 say--or vice versa---

problem tho with getting constistent results is well long term anyway -consistency can be just a tadge dull? swings and round abouts all the way

it may be that the design philosophy was wrong--(LOL definitely borrows tin hat here!!) in that the AI is trying to sink u-boats--- instead of trying to prevent the sinking of merchants which involves slightly different tactics i reckon..
i'd much rather gameplay wise be beaten off a convoy attack by the escorts-- having failed to sink any shipping -- perhaps heavily damaged only able to limp back to port---then get four or five ships then get blown out of the water after a ten hour DC attack---

neither is perfect but gameplay wise the reduction in tonnage achieved by the first out weighs the potentail realism of the second---

so get the sub submerged --keep the sub down till the convoy is safely out of range then return to post--based on the idea that whilst the escort is concentrating on you another sub could be approaching from a different course--this isnt true of course as we have no wolfpacks-but no harm in having the DD's behave as if it were true---after all with the less than realistic numbers of escorts in the game for convoys having two thrids of your defensive force attacking one possible target for many hours whislt the convoy travels over the horizon virtually unprotected isn't perhaps the best way to get the job done-----that was my ad hoc feelings on possible tweaks to the escort behaviour--

if we can get the escorts to reduce the tonnage achieved by the player -- while keeping the player alive but damaged perhaps so he can try again (at least most of the time) --then the reduced tonnage results should make playing the game seem more realistic--

PS cheers Caspofugin---wonder if it is the hydrophone like you say

HEMISENT · 05-21-06, 05:25 PM

Here's a thought. If someone could figure which settings or combination of settings make the AI absolutely, positively dead nuts accurate then perhaps use a set of settings which give so-so accuracy we now have parimeters. Plug these perimeters into SH3 Commander Randomized events.CFG and set it up for whatever odds you like. The result will be that the dead nuts accurate AI will appear at a random time unkown to the player.

Anyhow, just my two cents worth.

Salvadoreno · 05-21-06, 05:56 PM

CB.. Your destroyer aggression mod. How new is that?? I wanted to try it out but i enabled NYGM TW. How will this interact with nygm??

CB.. · 05-21-06, 06:54 PM

Quote:

Originally Posted by Salvadoreno

CB.. Your destroyer aggression mod. How new is that?? I wanted to try it out but i enabled NYGM TW. How will this interact with nygm??

it's very old Salva' and not with-out it's own problems..was never really even intended to be a mod at all--i have zero idea about it's compatibility with any large scale set of mods---i'm fairly sure it won't cause a crash but beyond that zip...make absolutely sure you make back up copys of the NYGM files before you over write them with mine---so you can revert again if you don't like it--the backs ups that i supply with the mod are from the stock game--important to note that before you install--allso i don't do realism mods i do gameplay mods so no tales of woe regarding realism please--it is what it is full stop--

you might hate it -you might think it's useless --you might find it..well fill in as appropiate--

Hemisent... i reckon that sort of thing is a great way to go for those folks who use SH3 commander at least --has to a good addition..getting the parameters right is still a big issue tho...

Ducimus · 05-21-06, 07:44 PM

I think we can toss the tinfoil hat. I renabled all sonar cones back to 90, and i really havent noticed much of a difference.

AI is just stupid. What is odd how its universally dumb, and yet, theres an occasional elite that can pound the crap out of you.

caspofungin · 05-21-06, 11:11 PM

Quote:

Originally Posted by HEMISENT

Here's a thought. If someone could figure which settings or combination of settings make the AI absolutely, positively dead nuts accurate then perhaps use a set of settings which give so-so accuracy we now have parimeters. Plug these perimeters into SH3 Commander Randomized events.CFG and set it up for whatever odds you like. The result will be that the dead nuts accurate AI will appear at a random time unkown to the player.

Anyhow, just my two cents worth.

easy -- commander already changes the values for the lost contact time. set it up so the sim.cfg has the hydrophone and sonar settings rem'd out (w/ the ; ) -- the effect is ubersensor.

gouldjg · 05-22-06, 04:24 AM

Guys, I have exams in 3 weeks so am very busy and will be out of the loop for a while. Its good to see some ai chat back on the scene CB. This time I think we all know the hardcore issues unless a miracle arrives.

I really wish I was not so tied up but thats life. Anyway, I just wanted to say,

Should any of you come up with a new set of ranges to plug into sh3 commander randomisation, I have already written most hex placements for all current sensors. http://hollywoodchaos.blogspot.com/2...-settings.html

It does not have to be a drawn out process as I did it. I wanted probability based settings so went this way however there is a way to get it shorter and simpler for any randomisation. Just PM me and I will try and show how.

I will be starting the whole file in three weeks time but that is not to say nobody can use it to produce models now (they may be better and I am no expert). I only created the file so we all can abuse it to suit our requirments.

So if you have kept tract of the randomisation notes made by Jscones, you may want to set up a decent test file specifically for the DD. Even if not random, it allows faster changing of the settings.

Anyway See you all in three weeks,

Ducimus · 05-22-06, 11:05 AM

Another dumb thoguht.

Lets say you make an HK group in the editor, and to save time, you select "GENERIC Destoryer Escort". I once heard that "generic" entries use the AI_hydrophones and AI_sonar in the AI_sensor.dat instead of all the specifically named hydrophone or sonar sets that have much better stastics, i never put much thought into that, but now i have to wonder if thats true.

05-21-06, 12:41 PM	#1
Ducimus Rear Admiral Join Date: May 2005 Posts: 12,987 Downloads: 67 Uploads: 2	A stupid hunch on that craptastic AI. This is probably going Waaaaayyy off the deepend, and i havent fully experimented to verify this behavior.. but.. its a hunch, and i wonder if anyone else has ever had similar thoughts. The key is this, has anyone noticed a reduction of the AI's use of active sonar after reducing the active sonar cone? This is central to this stupid hunch of mine. and Admittidly this might be hard to verify because the AI is so unpredictable and wonky to begin with. Here's my tinfoil hat, pulled it out of my ass, totally unsubstantiated, throw me in a rubber room with a vest with real long sleeves, theory. Code: - AI first listens passively. - If sound bearing it recieves is [u]within its sonar cone[/u,] it starts active sonar. - When active sonar bearing is receieved, [u]THEN[/u] the AI changes course to intercept contact. So heres the Picture: Red represents the cone size most of us adjust for blue represents the default. Grey line indicates Passive sonar bearing. Again, if passive bearing is within active sonar, start pinging. If ping gets a return, change course. If no ping, continue search. Thats the hunch. In everyones bid for realistic settings it could be that we've been nerfing the AI because it needs unrealistic sonar cones due a routine like this to function. The developers of Ubisoft , im quite certain were not stupid. SH3 took years in development, and a 5 -10 minute google search will show what active sonar cones looked like in WW2, so on that thought, it would almost seem to me that the 90 degree cone, was deliberate. I know, start throwing the shoes, im probably way off the deep end and i admit it, but its just a naggying feeling that this is whats going on. Now if anyone expereinces the same behavior CONSISTANTLY regardless cone size, then im just blowing alot of hot air with this post :rotfl:

05-21-06, 12:55 PM	#2
CB.. Ace of the Deep Join Date: Apr 2002 Location: UK Posts: 1,278 Downloads: 0 Uploads: 0	sounds logical to me..."borrows tin hat"-- im guessing the AI might not even be that particular regarding the sensors..i'm guessing that it just uses the one with the longest range to find you---then switches to sonar (if the game is that well made) to get depth reading before making a DC run... i would not be at all surpirsed if it doesn't actually use any sensors to get your depth..and that it in effect simply "cheats" getting depth info directly from the game engine in some manner--- i've been successfully blown out of the water by DD's with no active sonar on them at all..so how did it get the depth right...co-incidence??? how did it know what sort of depths to try? co-incidence? it may be an assumption to even think the game follows any sort of realistic real world pattern... but the end of the day the main problem with the AI is that it just doesn't co-operate terribly well (tho i still maintain the AI was great untill summat went wrong after a few hours of play---never to return --those creeping attacks were IMO real not imagined - i spent hours just watching them--as much as i tried to convince my self other wise..either way for me they simply stopped happening after maybe 10-20 hours gameplay never to return) that and the woe full lack of sensor blindspot after a DC drop- makes the whole gig a bit of a lash up-- mind you i dunno---what you say sounds ok to me.. __________________ the world's tinyiest sh3 supermod- and other SH3/SH2 stuff http://www.ebort2.co.uk/ The best lack all conviction, while the worst Are full of passionate intensity. W.B.Yeats

05-21-06, 01:53 PM	#8
CB.. Ace of the Deep Join Date: Apr 2002 Location: UK Posts: 1,278 Downloads: 0 Uploads: 0	it's a problem which ever way you approach it---on the one hand we definitely want some sort of dangerous response from the escorts-- on the other we don't want to killed every time----the area in between being ignored and being killed every time by the DD's is fairly small with not that much room for maneuver-- definitely using the crew rating entrys to see how any set up changes under different crew ratings has to be a good solid way to go---a overly uber set up might actually be spot with the crew rating reduced to 2 say--or vice versa--- problem tho with getting constistent results is well long term anyway -consistency can be just a tadge dull? swings and round abouts all the way it may be that the design philosophy was wrong--(LOL definitely borrows tin hat here!!) in that the AI is trying to sink u-boats--- instead of trying to prevent the sinking of merchants which involves slightly different tactics i reckon.. i'd much rather gameplay wise be beaten off a convoy attack by the escorts-- having failed to sink any shipping -- perhaps heavily damaged only able to limp back to port---then get four or five ships then get blown out of the water after a ten hour DC attack--- neither is perfect but gameplay wise the reduction in tonnage achieved by the first out weighs the potentail realism of the second--- so get the sub submerged --keep the sub down till the convoy is safely out of range then return to post--based on the idea that whilst the escort is concentrating on you another sub could be approaching from a different course--this isnt true of course as we have no wolfpacks-but no harm in having the DD's behave as if it were true---after all with the less than realistic numbers of escorts in the game for convoys having two thrids of your defensive force attacking one possible target for many hours whislt the convoy travels over the horizon virtually unprotected isn't perhaps the best way to get the job done-----that was my ad hoc feelings on possible tweaks to the escort behaviour-- if we can get the escorts to reduce the tonnage achieved by the player -- while keeping the player alive but damaged perhaps so he can try again (at least most of the time) --then the reduced tonnage results should make playing the game seem more realistic-- PS cheers Caspofugin---wonder if it is the hydrophone like you say __________________ the world's tinyiest sh3 supermod- and other SH3/SH2 stuff http://www.ebort2.co.uk/ The best lack all conviction, while the worst Are full of passionate intensity. W.B.Yeats

05-21-06, 05:25 PM	#9
HEMISENT Ace of the Deep Join Date: Feb 2003 Location: Northern Illinois Posts: 1,052 Downloads: 135 Uploads: 0	Here's a thought. If someone could figure which settings or combination of settings make the AI absolutely, positively dead nuts accurate then perhaps use a set of settings which give so-so accuracy we now have parimeters. Plug these perimeters into SH3 Commander Randomized events.CFG and set it up for whatever odds you like. The result will be that the dead nuts accurate AI will appear at a random time unkown to the player. Anyhow, just my two cents worth. __________________ Nuke 'em till they glow!

05-21-06, 05:56 PM	#10
Salvadoreno Grey Wolf Join Date: May 2005 Posts: 887 Downloads: 119 Uploads: 0	CB.. Your destroyer aggression mod. How new is that?? I wanted to try it out but i enabled NYGM TW. How will this interact with nygm?? __________________

05-21-06, 12:59 PM	#3
Ducimus Rear Admiral Join Date: May 2005 Posts: 12,987 Downloads: 67 Uploads: 2	im almost 100% positive the game "cheats" on its depth settings. When the AI does acutaly attack, it rarely misses with its default settings. Infact, when depth charges are dropped, i acutaly go deeper to get away from that depth. Assuming that the depth charges are "set" at whatever depth i was when they were first launched. Of course, those depth charges still seem to manage to explode close, regardless of depth change

05-21-06, 01:23 PM	#5
Ducimus Rear Admiral Join Date: May 2005 Posts: 12,987 Downloads: 67 Uploads: 2	Well heres what got me thinking. You'd agree that ELITE's are the most consistant in behavior, most aggressive, and usualy the ones that would cream you no? Ive acutally had this hunch for awhile. Orginally i was modding my game leaving the sonar cone as default, only fiddeling with a min range, and then tried to add for more inaccuracies within the depth charges and hedge hogs themselves. That was awile ago, and i forgot why i was doing it that way. Over the last couple weeks i went back to fiddling with sonar. Made the cone 70 degrees, increased the min distance.. all the semi realistic stuff that mos of us do. Well, in my game ive added quite a few elite DE's. You know.. those dreaded buckleys? I have 3 ship convoys escorted wtih two elite DE's in the carribean after 42. Out of 4 convoy attacks, i had only ONE close call. All the other 3 attacks resulted in me silpping away without so much as one depth charge being dumped anywhere near me. Thats when i remembered this hunch of mine that i forgot about.

05-21-06, 01:40 PM	#6
caspofungin Commander Join Date: Apr 2005 Posts: 459 Downloads: 41 Uploads: 0	i' not so sure... i've set up my ai_sensors file so that at silent running, i'm impossible to hear unless the escort is w/in a couple of hundred yards. nevertheless, i can start getting pinged when the escort is 1400-1500m away. of course, that doesn't mean the ai isn't cheating... w/ respect to the ai getting the depth automatically -- maybe. but also consider the possibility that the info it gets from hydrophones isn't limited to bearing only (as irl) but gets bearing, elevation, and range -- allowing it to pinpoint your position in 3d space.

05-21-06, 01:44 PM	#7
caspofungin Commander Join Date: Apr 2005 Posts: 459 Downloads: 41 Uploads: 0	also, re what cb said -- i agree completely -- it's probably got more to do w/ the coded ai routines than the sensors themselves.

05-21-06, 07:44 PM	#12
Ducimus Rear Admiral Join Date: May 2005 Posts: 12,987 Downloads: 67 Uploads: 2	I think we can toss the tinfoil hat. I renabled all sonar cones back to 90, and i really havent noticed much of a difference. AI is just stupid. What is odd how its universally dumb, and yet, theres an occasional elite that can pound the crap out of you.

05-22-06, 04:24 AM	#14
gouldjg Grey Wolf Join Date: Mar 2005 Location: Manchester UK Posts: 881 Downloads: 4 Uploads: 0	Guys, I have exams in 3 weeks so am very busy and will be out of the loop for a while. Its good to see some ai chat back on the scene CB. This time I think we all know the hardcore issues unless a miracle arrives. I really wish I was not so tied up but thats life. Anyway, I just wanted to say, Should any of you come up with a new set of ranges to plug into sh3 commander randomisation, I have already written most hex placements for all current sensors. http://hollywoodchaos.blogspot.com/2...-settings.html It does not have to be a drawn out process as I did it. I wanted probability based settings so went this way however there is a way to get it shorter and simpler for any randomisation. Just PM me and I will try and show how. I will be starting the whole file in three weeks time but that is not to say nobody can use it to produce models now (they may be better and I am no expert). I only created the file so we all can abuse it to suit our requirments. So if you have kept tract of the randomisation notes made by Jscones, you may want to set up a decent test file specifically for the DD. Even if not random, it allows faster changing of the settings. Anyway See you all in three weeks, __________________ My Mods Gouldjg's Crew Ability Balancing Mod for SH5 1.2 http://www.subsim.com/radioroom/showthread.php?t=169630

05-22-06, 11:05 AM	#15
Ducimus Rear Admiral Join Date: May 2005 Posts: 12,987 Downloads: 67 Uploads: 2	Another dumb thoguht. Lets say you make an HK group in the editor, and to save time, you select "GENERIC Destoryer Escort". I once heard that "generic" entries use the AI_hydrophones and AI_sonar in the AI_sensor.dat instead of all the specifically named hydrophone or sonar sets that have much better stastics, i never put much thought into that, but now i have to wonder if thats true.