Home » MODDING HQ 1.13 » v1.13 Idea Incubation Lab  » HAM 3.6 Full - RELEASED
Re: HAM 3.6 Beta - RELEASED[message #244714] Fri, 19 February 2010 19:32 Go to previous messageGo to next message
CptMoore

 
Messages:224
Registered:March 2009
Ok replying to myself :naughty:, is it already possible to make NV goggles lower the fov? Right now they give insane range bonuses at night but I don't think they restrict your FOV, which they should no?
Re: HAM 3.6 Beta - RELEASED[message #244715] Fri, 19 February 2010 19:38 Go to previous messageGo to next message
Sot

 
Messages:86
Registered:April 2009
Apparently edit settings using XML_editor or Notepad.
Re: HAM 3.6 Beta - RELEASED[message #244721] Fri, 19 February 2010 20:35 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Yeah, increase the tunnelvisionpercentage.


Re: HAM 3.6 Beta - RELEASED[message #244723] Fri, 19 February 2010 21:30 Go to previous messageGo to next message
Wil473

 
Messages:2842
Registered:September 2004
Location: Canada
Taking a break from real life, and thought I should pop in to offer a word of warning about a easy way to swap weapon optics. Oh and voice a suggestion for the AI situation.

Weapon Optics
Unlike goggles, weapons optics typically have an AP cost associated with attaching them to the gun. I hope this hot key or button will only be allowed to function in realtime, otherwise it may be open to exploitation.

Agressive AI
Before devoting work to granting the AI better situational awareness (something I whole heartly support anyways), can a test exe be created where different CTH conditions can be set to test AI responses. Ie. the real CTH is 30 but the AI uses 30+(ini defined bump) to make its decision on shoot or run. The real CTH would of course apply for the actually shot. This way savegames of "stupid AI" behaviour can be produced, and then tested against the theory that the CTH calculation can make it decide between running or shooting smarter.

EDIT:
Oh and is a button even worth it, as there is more to optics selection than Day/Night vision range bonus.

[Updated on: Fri, 19 February 2010 21:32] by Moderator



Re: HAM 3.6 Beta - RELEASED[message #244724] Fri, 19 February 2010 21:40 Go to previous messageGo to next message
Sot

 
Messages:86
Registered:April 2009
wil473
I hope this hot key or button will only be allowed to function in realtime, otherwise it may be open to exploitation.
True. At the same time it would be a good idea to ban the device to attach certain weapons (scopes, collimators, etc.) during the battle! Maybe something else ...

to Headrock
Could you explain how the game differentiates between night and day optics and other device?

[Updated on: Fri, 19 February 2010 21:50] by Moderator

Re: HAM 3.6 Beta - RELEASED[message #244731] Fri, 19 February 2010 23:41 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Quote:
Unlike goggles, weapons optics typically have an AP cost associated with attaching them to the gun. I hope this hot key or button will only be allowed to function in realtime, otherwise it may be open to exploitation.


Or, more logically, to have it cost the proper amount of APs in turn-based mode (and not work if not enough APs available).

However, the more I think about it, the less I like the idea of having a hotkey for this at all. Sounds like more trouble than it's worth.

Quote:
Before devoting work to granting the AI better situational awareness (something I whole heartly support anyways), can a test exe be created where different CTH conditions can be set to test AI responses.


Yes, it can. I'll try to make one with HAM 3.6 as a base (it's easier for me). If you'd rather have one with 1.13 as a base, let me know.

Quote:
Could you explain how the game differentiates between night and day optics and other device?


BrightLightVisionBonus versus NightVisionBonus. If the item has a positive amount of BrightLightVisionBonus, it's considered "day goggles". If it has a positive amount of NightVisionBonus, it's considered "night goggles". If it has both, things get messy.

If you want to know more about swaps, you can read this.

[EDIT: Thanks Dieter/MM]


Re: HAM 3.6 Beta - RELEASED[message #244734] Fri, 19 February 2010 23:56 Go to previous messageGo to next message
Sot

 
Messages:86
Registered:April 2009
Headrock


Or, more logically, to have it cost the proper amount of APs in turn-based mode (and not work if not enough APs available).

However, the more I think about it, the less I like the idea of having a hotkey for this at all. Sounds like more trouble than it's worth.


to Headrock
Still, I beg you not to depart from my idea Smile As an option - enter this option on / off in the INI.

Re: HAM 3.6 Beta - RELEASED[message #244736] Sat, 20 February 2010 00:06 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Here's the EXE for testing:

Download.
NOTE: This is a HAM 3.6 exe. You need all the HAM files installed for this to work.

You need to add a new INI setting in the "JA2 HAM Settings" section of your INI file:
CTH_REQUIRED_TO_CONSIDER_COVER_MODIFIER

The value of this setting will increase or decrease the CTH requirement for considering taking cover in BLACK decision mode. In fact, it will also show a message telling you the current CTH (without modifier) and the range to the target.

Increase the modifier to make it EASIER to seek cover.

Also note: This only applies to seeking cover in BLACK mode. That means when the AI character can see you. If the character is seeking cover but the message doesn't display, it means one of two possible things:

A) The enemy can't see you. He is in RED decision mode. If this occurs a lot, it means the problem is located elsewhere.
B) The enemy couldn't find a good attack at all, and resorted to taking cover because it's the only thing he could do.


Re: HAM 3.6 Beta - RELEASED[message #244737] Sat, 20 February 2010 00:07 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Quote:
Still, I beg you not to depart from my idea \:\) As an option - enter this option on / off in the INI.


The effort involved may not be worth the end result. Frankly, I don't think there are many people out there who would even have a use for this hotkey.


Re: HAM 3.6 Beta - RELEASED[message #244739] Sat, 20 February 2010 00:18 Go to previous messageGo to next message
Sot

 
Messages:86
Registered:April 2009
Is it convenient to change the scopes for each rifle (considering that the total detachment may be 32 mercenary!) In hand?
Re: HAM 3.6 Beta - RELEASED[message #244740] Sat, 20 February 2010 00:19 Go to previous messageGo to next message
Faithless

 
Messages:442
Registered:October 2009
Location: The safe end of the barre...
You'd probably just want to switch the squad members guns at most.
Re: HAM 3.6 Beta - RELEASED[message #244744] Sat, 20 February 2010 00:50 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
I'm finding more and more issues with the AI, I think I'm starting to understand why it resorts to taking cover so much. Thing is - the AI checks several times to see whether it has enough APs to add extra aiming levels to its shots. However, from what I'm seeing, it is NOT properly calibrating to the current AP system. In addition, it may or may not take Increased Aiming Costs into account (assuming this feature is activated at all, of course). This is getting more complicated as I go along. It may be required to build a debug EXE with error reporting to see exactly what's happening. Once I'm done analyzing the routine, I'll try to create such an EXE.


Re: HAM 3.6 Beta - RELEASED[message #244745] Sat, 20 February 2010 01:18 Go to previous messageGo to next message
ChonkE

 
Messages:17
Registered:January 2008
Location: Utah
Brilliant work yet again HR. I am almost done with my 3.5/WF/STOMP/Alpha campaign and school is slowing up for a bit so I can fully spend time with your revision. Is there a way to switch the Bloodcat ambushes to be something different like changing them to Kingpin's assassin group? Also what are the limitations of the facility XMLs? Keep up the good work!
Re: HAM 3.6 Beta - RELEASED[message #244751] Sat, 20 February 2010 02:09 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Thanks, good to see you back here.

Quote:
Is there a way to switch the Bloodcat ambushes to be something different like changing them to Kingpin's assassin group?


You mean, spawn assassins instead of bloodcats? I think that would be very, very, very complicated.

Quote:
Also what are the limitations of the facility XMLs?


That is a very broad question, I'm not sure exactly what you're asking. Seeing as the Facility System isn't a living god with omnipotent powers of farting thunder and pissing brimstone, there are plenty of limitations. Please be more specific.

----------------

Back on the AI topic:

I've just completed an analysis of how the AI finds the best target to shoot at. It's obvious that some things have to be altered here, especially now that CTH can drop to 0 - it has the potential of eliminating consideration of targets that might otherwise be selected for blind autofire. Also, as I mentioned above, not scaling the AimingTime APs properly can easily lead to problems where the AI forgoes a reasonably good attack because he doesn't have sufficient APs to actually perform it. Hopefully solving these two issues may increase AI shooting probability, although I'm reasonably sure that the problem is still located elsewhere.

I'll put out a new testing EXE shortly, with fixes to these issues as well as on-screen messages that trigger at the appropriate times. Stay tuned.


Re: HAM 3.6 Beta - RELEASED[message #244753] Sat, 20 February 2010 02:27 Go to previous messageGo to next message
Wil473

 
Messages:2842
Registered:September 2004
Location: Canada
This almost sounds like decision making in real life: the AI needs to make decisions based on what it hopes to be possible (a hit when CTH is approaching 0) instead of what is actually real (don't waste the ammo). Now based on what your finding, an additional question is, has anyone seen how the AI behaves under a 25AP game recently?

I hope to try out messing around with the CTH (precieved by the AI), but won't have time till March. I need to try HAM 3.6Beta with the last UC-1.13 Hybrid release anyways, so I'll see if the AI gets more shooty with the .ini tweaked then. Thanks for hearing out the idea.

[Updated on: Sat, 20 February 2010 02:45] by Moderator



Re: HAM 3.6 Beta - RELEASED[message #244757] Sat, 20 February 2010 03:04 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Well, the AI shouldn't conserve ammo, its only chance is by throwing everything it has on the player as soon as possible. The thing is that, especially with HAM suppression, a stray shot isn't always a bad idea. At least it's a shot.

In any case, I've repaired the AI's target acquisition a bit. It will no longer consider 0 CTH as completely hopeless (though yes, it does consider it a bad idea, like it always did). That should prevent it forgoing a target completely.

In addition, I've repaired the Aiming Costs calculation. The AI will now be able to tell how many APs it will ACTUALLY need to add aiming levels. This should avoid situations in which it has picked a target to fire at but doesn't have enough APs to actually perform that attack (causing it to give up and go hide somewhere).

I'm going to try to add a few "debug messages" at key points, so we can get more reports on what exactly the AI is doing in combat. This will be available soon.


Re: HAM 3.6 Beta - RELEASED[message #244760] Sat, 20 February 2010 03:36 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Spotted another AI inconsistency in the Search-for-cover formula. Again, it is assuming that each tile of movement costs about 1 AP. In 100AP mode, this would cause the AI to look for cover far from its current position, without an actual ability to reach that cover in time.

This is getting very frustrating. I'm going to join wil here in asking you guys to report whether the AI is as stupid in 25AP mode as it is in 100AP mode.


Re: HAM 3.6 Beta - RELEASED[message #244761] Sat, 20 February 2010 03:37 Go to previous messageGo to next message
Kaerar

 
Messages:2045
Registered:January 2003
Location: Australia :D
Why not have a counter test in 250AP mode too to test if the AI gets even more stupid Wink


Re: HAM 3.6 Beta - RELEASED[message #244762] Sat, 20 February 2010 03:43 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
LOL that might actually work, although from what I'm seeing, it is already reaching its own stupidity cap in 100AP.

Goddammit I got tangled in the AI routines again. This is extremely unrewarding work done with a goddamn toothpick. It's making my head hurt. And I've still got HAM 3.6 to complete, WIKI articles to write, and now WarmSteel is going to have a go at NGAP...

Uggggghhhhhhhh


Re: HAM 3.6 Beta - RELEASED[message #244765] Sat, 20 February 2010 04:04 Go to previous messageGo to next message
Wil473

 
Messages:2842
Registered:September 2004
Location: Canada
Wait a minute, 1AP per tile is still 25 tiles in the old system. I don't remember being able to run 25 tiles in one turn(or am I just being forgetful). Or are you saying that the AI should only consider cover that is within a turn or two of running?


Re: HAM 3.6 Beta - RELEASED[message #244768] Sat, 20 February 2010 04:35 Go to previous messageGo to next message
Khor1255

 
Messages:1826
Registered:August 2003
Location: Pleasantville, NJ
Going from standing still to movement takes 2 aps, you then move 1 tile per ap in running mode. But if you ended your turn while in mid stride running you can indeed have a character run 25 tiles in one turn.
Re: HAM 3.6 Beta - RELEASED[message #244769] Sat, 20 February 2010 04:45 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
The range at which the AI looks for cover is dependent on several factors:

A) Wisdom. For every 8 points of wisdom, he/she can look 1 tile away.
B) Roaming Range. As defined by the character's internal data, only applies if can't see enemies.
C) How far we can go with the APs we've got.

With the 100AP system, option C gives values that are ridiculously large. Of course, these values cap one another, so it doesn't matter in most cases.

However, consider this case:
1) The character has 80 Wisdom. He has just finished running to cover, and has 20 APs left.
2) The character can't find anything to shoot at with reasonable certainty. The AI routine kicks in and looks for better cover.
3) With the bad calculation, APs cap the search range at 16 tiles: 4 are reserved for the APBP Constant for starting to run (that at least is handled correctly), and then 1 tile per AP (wrong for 100AP, where it costs 4AP, and that's on FLAT GROUND!).
4) Wisdom caps the range at 10 tiles.
5) The character finds better cover 10 tiles away.
6) The character starts running to that cover, but...
7) Runs out of APs after 5 tiles, because that's how much it costs to run that far. (20 APs = 4 to start + (4AP*4 tiles=16AP) to move).
Cool The character is now standing in open ground, halfway to his destination.

I've fixed this now. I'm actually close to solving all the overt problems in the BLACK AI routine, at least the ones I could spot after analysis. Hopefully, the AI is going to behave a little bit smarter now... Hopefully.


Re: HAM 3.6 Beta - RELEASED[message #244770] Sat, 20 February 2010 04:52 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Fixed another problem. The original code was supposed to practically FORCE enemies to choose burst-fire at targets less than 10 tiles away if the difficulty is not NOVICE. Unfortunately, this was completely borked by bad coding. Fixed that. Expect death at close range.


Re: HAM 3.6 Beta - RELEASED[message #244772] Sat, 20 February 2010 04:58 Go to previous messageGo to next message
usrbid

 
Messages:1536
Registered:December 2008
Hi Headrock, I still feel bad for messing up your post, please let me know if I can do anything for you! Thanks for being so patient, Dieter


Re: HAM 3.6 Beta - RELEASED[message #244773] Sat, 20 February 2010 05:04 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
LOL don't worry about it. Water under the bridge.

Fixed another problem. Looks like if the enemy decides not to use Autofire, it removes all extra aiming from the shot. This causes shots fired with 0 aiming, even though the "Best Attack" calculation dictated extra aiming should be used.

God it's 5 a.m.

I still need to fix aiming cost calculations. These are a serious problem - the AI tries to figure out whether it has enough APs to fire at you, but doesn't take 100AP nor Increased Aiming Costs into account. The end result is probably trying to fire but failing miserably due to lack of APs.

Man... ok no, this waits until tomorrow.


Re: HAM 3.6 Beta - RELEASED[message #244775] Sat, 20 February 2010 05:23 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
BTW I figured out why enemies seek cover all the time.

I'll tell you all about it... TOMORROW.

MUAHAHAHAHAAHAHAHA :vader:


Re: HAM 3.6 Beta - RELEASED[message #244782] Sat, 20 February 2010 09:25 Go to previous messageGo to next message
Faithless

 
Messages:442
Registered:October 2009
Location: The safe end of the barre...
Pfsh, that's like saying;
Hey I've got a cookie, want it? Well... TOO BAD :devilaugh:

Awesome though, seems like you're fixing alot of problems
Re: HAM 3.6 Beta - RELEASED[message #244785] Sat, 20 February 2010 10:21 Go to previous messageGo to next message
Deane

 
Messages:62
Registered:January 2010
Headrock's dedication is scary..

Now if I wasnt to obsessed with STOMP mod, I might actually give HAM3.6 a try O_o
Re: HAM 3.6 Beta - RELEASED[message #244791] Sat, 20 February 2010 13:14 Go to previous messageGo to next message
Telpscorei

 
Messages:13
Registered:September 2007
Location: United Kingdom
Headrock, you are the God-King of coding. I am always amazed at how much work you put in. I'm a little busy prepping for job interviews at the moment, but if you need a hand, I can look over code and offer analyses, improvements, etc. Though not at your crazy rate, but I figure I can help take some of the pressure off of you. Just let me know what you want me to look at, and I'll get back to you with info asap.


EDIT: I may have forgotten why you should trust me with this sort of thing. I've just finished a masters in computer science with modules in software engineering and artificial intelligence. In fact my dissertation focused on both of those. So if you need an extra pair of eyes with that kind of training, I'm here.

[Updated on: Sat, 20 February 2010 13:16] by Moderator

Re: HAM 3.6 Alpha v7 - RELEASED[message #244794] Sat, 20 February 2010 13:33 Go to previous messageGo to next message
datakurs

 
Messages:166
Registered:June 2009
Location: Budapest, Hungary
Some positive feedback on 3.6b after several hours of playing (I hope helps a bit to finally remove the greek alphabet and encourages Sandro to make his mod compatible with the newest HAM):
- no more interrupted movements
- no more immobile Hans
- no more immobile opponents in boxing ring
Re: HAM 3.6 Alpha v7 - RELEASED[message #244800] Sat, 20 February 2010 15:12 Go to previous messageGo to next message
dzeller

 
Messages:91
Registered:February 2003
Location: USA
Seems like there is some good progress being made both on the bugfix and AI front! Thanks for all your hard work Headrock!

I'm looking forward to being able to play the newest HAM with the AI fixed (or even improved!). HAM adds a lot not only because of the suppression system but also beacause of all those little nifty features (off-road Hummer, militia upkeep costs, ...). So once again, kudos!
Re: HAM 3.6 Beta - RELEASED[message #244808] Sat, 20 February 2010 17:33 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
@ Telpscorei: Thanks kindly for the offer, I'll keep that firmly in mind. At the moment I'm still trying to figure out what I need to do at all, I've got several things pressing, and can't decide which one to hit. My problem is that it's hard for me to sometimes keep track of what I'm doing and how I did it, so "out-sourcing" stuff can easily confuse me, but if I run into something that looks way too big on me I'll look you up!

Quote:
- no more interrupted movements
- no more immobile Hans
- no more immobile opponents in boxing ring


Excellent news, thanks Datakurs.

Quote:
I hope helps a bit to finally remove the greek alphabet and encourages Sandro to make his mod compatible with the newest HAM


I don't think he needs encouragement, he's expressed interest in doing that in the past, I'm just not yet done with it. However, RoWa is already well on the way to releasing 1.13+HAM3.6 so I'm guessing it won't be long before STOMP is updated to that.

Quote:
I'm looking forward to being able to play the newest HAM with the AI fixed (or even improved!).


If I can iron out some more AI errors I'll probably release Beta v3 later today. I'm just hoping that AI behavior will be better rather than worse. If you remember, in the past some attempts to fix the AI have resulted in it being borked the other way...


Re: HAM 3.6 Alpha v7 - RELEASED[message #244810] Sat, 20 February 2010 17:55 Go to previous messageGo to next message
datakurs

 
Messages:166
Registered:June 2009
Location: Budapest, Hungary
One more thing if you have time to patch:
When I want to attack the rednecks many party members just don't want to do it, probably they think they need to kill innocent civilians. No problem, finally the Hicks family will be wasted. When I go back to Keith the unnamed shopkeepers attack me in that sector, and even the militia kills them with the sad side-effect of decreasing loyalty. This must be something wrong with the factions.
Re: HAM 3.6 Beta - RELEASED[message #244811] Sat, 20 February 2010 18:05 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Ok, I promised to tell you all about why the AI seeks cover so often. It's actually not as complicated as it seems, although it took me a couple of read-throughs of the code to figure out what should've been pretty obvious to begin with. I was going to publish an article with a detailed analysis of how the AI works in actual combat (I.E. when it can see you) but then I started finding all the above problems and started fixing them. So instead of an article I'll just give you a run-down.

First a basic explanation. The AI works in "STATES", going from GREEN state to YELLOW state to RED state to BLACK state. In each state, it has a completely different method of decision. The two most interesting states are RED and BLACK. In RED state, the enemy character already positively knows that there are mercs in the sector, but just can't see any at the moment. In BLACK state, which is what I've been fixing, the enemy has a merc/militia in sight. He needs to decide whether to shoot, duck for cover, or perform one of the other possible actions.

Basically, the A.I. goes through a simple routine of deciding what action to choose. Here's the outline of how this proceeds:

A) The AI looks for emergency conditions and reacts to them. This includes climbing out of deep water, moving out of a gas cloud, running for an alarm switch, and other life-saving actions along this vein.
B) If it didn't perform any of the above, the A.I. now calculates his best possible GUNSHOT at any target. This takes into account the best CTH it can get at a target, how many extra aiming APs must be spent to hit that target, how dangerous the target is, etcetera.
C) It does the same calculation to figure out the best target for a STAB/KNIFE THROW, and another best target for a GRENADE/LAUNCH.
D) If any of the calculations yields 0 CTH (which is possible since HAM 2.8 was integrated, given that you don't alter the default settings), that attack is considered IMPOSSIBLE and is completely discarded. This is very important for cover checks later on. In the original game, the A.I. was expecting at least 1 CTH because of an enforced minimum, even on shots that are extremely unlikely to hit (which originally led to sharpshooting pistols).
E) So by now we've got up to three separate targets for three different kinds of attacks. If no target was found, we can still consider running up and punching someone, but that's really a last resort (unlikely that we'll actually reach that).
F) The AI compares the three attacks and chooses the one that's most likely to succeed based on CTH. Note that this CTH value takes into account the target's cover - I.E. how much damage the bullet/attack will cause to the target.
G) At this point there is a check to see whether the A.I. wants to completely forgo an attack and duck into cover instead. SpaceViking set this up to allow ducking into cover only for very high-morale, brave characters with lots of APs and good phsyical condition. I originally thought this was what was causing the AI to duck into cover a lot, but it isn't: Most characters are actually extremely unlikely to pass this check. If they do pass, of course, they will almost definitely duck into cover.
H) Now comes a check to find better cover nearby. Most characters, especially those that haven't been able to find any attack with more than 0 CTH, will be able to run this calculation. The result is a percentage that tells us how much better the new cover will be than the one we are in now.
I) This is the nasty part. The AI compares the improvement in cover (as calculated in point H) to the CTH of the best attack we could find (point F). The vast majority of AI characters will receive a MASSIVE bonus to the cover improvement percentage, pretty much guaranteeing that Ducking for Cover will be chosen over Attacking unless the CTH is ridiculously high (upwards of 50 points) or the current cover is really better than anything around it!! If this happens, the A.I. deletes all possible attack parameters.
J) If an attack is still deemed best, the AI will decide whether it wants to burst or autofire, but if the original attack required any Extra Aiming, it will probably forgo both burst and auto, sticking with a single-shot attack. This was further borked because failing to autofire was likely to erase extra aiming points, causing the character to fire a single, unaimed shot instead!
K) If no attack was made, the character defaults to seeking better cover, assuming one was found at all.
L) If neither attack nor cover were selected, the AI performs a generic action like radioing in, changing stance, changing direction, or even running away.

The end result of this grueling formula is that the AI fails to make anything close to a reasonable decision. Even if it does make the correct one, failure to take into account the 100AP system can easily lead him to perform it badly, or not have enough APs to perform it at all, leading to what looks like Artificial Stupidity (for instance, running to cover and stopping halfway there).

This AI routine runs once every time the previous AI decision ends. That means it's possible to run it several times every turn. In addition, bad AP calculations can lead to deadlock - by repeatedly choosing an action without the necessary number of APs required to perform it.

To make matters worse, the game's TunnelVision mechanic may easily cause the AI to switch back into RED mode (decision process when enemies aren't visible) even when in close proximity to mercs. That would cause choosing actions that are completely inappropriate for close-combat situations, and is far more likely to trigger suicidal movement.

Solving all these problems is more difficult than just fixing the bugs - it may be necessary to change the order of the above decision process completely. The usefulness of bursts and autofire is completely negated by the process, and forgoing a low-CTH attack may not be the AI's best choice at all.

I'm not sure how I can go about this, or if I can at all. Still, I'm going to try to fix as many of the underlying problems as I possibly can, hoping that better data will at least encourage the AI to make better decisions.

Once I'm done with the simple stuff, I'll release a new EXE. I'm hoping that you guys will give enough feedback that we can figure out whether my fixes have helped or made things worse.


Re: HAM 3.6 Beta - RELEASED[message #244812] Sat, 20 February 2010 18:06 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Quote:
When I want to attack the rednecks many party members just don't want to do it, probably they think they need to kill innocent civilians. No problem, finally the Hicks family will be wasted. When I go back to Keith the unnamed shopkeepers attack me in that sector, and even the militia kills them with the sad side-effect of decreasing loyalty. This must be something wrong with the factions.


Yes, I'm hoping to figure out why this happens. Question: Are you using WildFire?


Re: HAM 3.6 Alpha v7 - RELEASED[message #244813] Sat, 20 February 2010 18:07 Go to previous messageGo to next message
datakurs

 
Messages:166
Registered:June 2009
Location: Budapest, Hungary
Wildfire maps only + map update.
Re: HAM 3.6 Beta - RELEASED[message #244816] Sat, 20 February 2010 18:31 Go to previous messageGo to next message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Yeah, we know that WildFire is a bit quirky in that regard, though I couldn't say why. If anyone encounters this problem in normal JA2, however, then it may be more pressing to find out why it happens.

BTW, the shopkeepers attack you? Are you using HAM's "no-civilian-hostility" setting?


Re: HAM 3.6 Alpha v7 - RELEASED[message #244817] Sat, 20 February 2010 18:51 Go to previous messageGo to next message
Telpscorei

 
Messages:13
Registered:September 2007
Location: United Kingdom
From reading what you've discovered, it seems that every soldier goes through this decision making process on his own. We therefore have a couple of questions that need answering before proceeding further into big changes to the A.I.; do we want to fight 20 soldiers each working to kill the player, or do we want to fight 20 soldiers under the control of a A.I. director?

It sounds like we currently fight 20 independent soldiers, and while this is fine (and getting better with your fixes Headrock), I'm wondering about getting the soldiers to work together to kill us.

At any rate, there sounds like there's a whole lot of changes and improvements that can be made before we get there. I'm going to take a look at the current source code and see if I can figure anything out.
Re: HAM 3.6 Alpha v7 - RELEASED[message #244819] Sat, 20 February 2010 19:04 Go to previous messageGo to next message
datakurs

 
Messages:166
Registered:June 2009
Location: Budapest, Hungary
CAN_TRUE_CIVILIANS_BECOME_HOSTILE = FALSE
Only shopkeepers and the bartender, so the soldier-type civilians.
Re: HAM 3.6 Beta - RELEASED[message #244823] Sat, 20 February 2010 19:36 Go to previous messageGo to previous message
Headrock

 
Messages:1774
Registered:March 2006
Location: Jerusalem
Quote:
From reading what you've discovered, it seems that every soldier goes through this decision making process on his own. We therefore have a couple of questions that need answering before proceeding further into big changes to the A.I.; do we want to fight 20 soldiers each working to kill the player, or do we want to fight 20 soldiers under the control of a A.I. director?


LOL that's the age-old question though, isn't it. Yes, we are fighting 20 independent soldiers, there's no over-arching AI to control them all. They do interact with each other in a very limited fashion (reporting contacts to each other, requesting spotting aid for snipers, etc) but it's far from cohesive planning. In any case, I actually did start working on a squad AI, at least a basic one, but quickly realized that I had bitten more than I could chew. That was like a year ago. In the meanwhile, aechua wrote some actual research on implementing a LEARNING AI... but that would require a scripting system which we don't have.


Previous Topic: Space Viking's releases (was: Many Mercenaries...)
Next Topic: How about adding an UAV?
Goto Forum:
  


Current Time: Thu Jun 04 01:02:34 EEST 2020

Total time taken to generate the page: 0.01750 seconds