Better AI Idea - modders please read - Page 2

Jourin · #11 November 28th, 2001, 09:18 PM

This past weekend I ran a series of 5 test games with 10 AIs (I sometimes took control). I changed the minimum number of mine sweepers to 6 in all the AI scripts. Prior to an attack, the AI built 6 minesweepers and sent them in as a group. The six minesweepers cleared the 100 mines in usually in the first pass (depending on mine level - in one case the did not, but cleared the majority that the fleet, although depleted still made it through the minefield) and then sent in the 40 ship fleet. I could no longer depend on minefields for the AI to kill itself on. I recommend that all the modpack AI mod makers seriously consider upgrading the minimum number of minesweepers to 6. Conduct your own tests but please consider.

Other tests:
I added a separate queue for attack, but left the prepare to attack queue the same. Did not achieve the expected results. I could not find a clear distinction between attack and prepare to attack by the AI. Recommend keeping the current setup on one queue for both.

I added a separate queue for defend (long term), and in all but one case achieved the expected results. The AI consistently went into this state when it was loosing. I used this queue to have the AI build lots of mines, satellites, fighters, weapon platforms, bases, and then ships. The AI put up such a strong defense that it actually stopped me. Usually once I get the AI on the ropes, it folds, but it actually stopped me. The one case that it didn't work occurred when an Ork ship was transported by an event into Norak space and moved into the Norak home system. The Norak's went into the defend long term state for over 50 turns. Maybe since the Norak's could not attack because of no direct contact, the AI put them into the defend long term state, or maybe since the home system was "attacked", the AI went into the defend long term state. So I added a colonizer back into this queue but with a minimum of 1. The AI needs to build at least 1 colonizer so it doesn't kill itself like the Norak's by not expanding, but does not need to constantly waste resources by building colonizers that immediately get destroyed when actually in a real defensive situation. I replayed the Norak game and with the addition of the colonizer, the Norak's changed states and moved out of defend long term much sooner then 50 turns. So it played differently the second time around.

Recommend that the modpack AI makers seriously consider and test a separate queue for defend long term.

I also made a separate research queue for defend long term, so it would research defensive techs. Maybe could expand to include separate techs for infrastructure, etc. Might also help the AI.

Going to take a break from SEIV since it is the only game I have played for a year and try some Civ III, but since I still love SEIV I will be back.

Jourin · #12 November 28th, 2001, 09:21 PM

This past weekend I ran a series of 5 test games with 10 AIs (I sometimes took control). I changed the minimum number of mine sweepers to 6 in all the AI scripts. Prior to an attack, the AI built 6 minesweepers and sent them in as a group. The six minesweepers cleared the 100 mines in usually in the first pass (depending on mine level - in one case the did not, but cleared the majority that the fleet, although depleted still made it through the minefield) and then sent in the 40 ship fleet. I could no longer depend on minefields for the AI to kill itself on. I recommend that all the modpack AI mod makers seriously consider upgrading the minimum number of minesweepers to 6. Conduct your own tests but please consider.

Other tests:
I added a separate queue for attack, but left the prepare to attack queue the same. Did not achieve the expected results. I could not find a clear distinction between attack and prepare to attack by the AI. Recommend keeping the current setup on one queue for both.

I added a separate queue for defend (long term), and in all but one case achieved the expected results. The AI consistently went into this state when it was loosing. I used this queue to have the AI build lots of mines, satellites, fighters, weapon platforms, bases, and then ships. The AI put up such a strong defense that it actually stopped me. Usually once I get the AI on the ropes, it folds, but it actually stopped me. The one case that it didn't work occurred when an Ork ship was transported by an event into Norak space and moved into the Norak home system. The Norak's went into the defend long term state for over 50 turns. Maybe since the Norak's could not attack because of no direct contact, the AI put them into the defend long term state, or maybe since the home system was "attacked", the AI went into the defend long term state. So I added a colonizer back into this queue but with a minimum of 1. The AI needs to build at least 1 colonizer so it doesn't kill itself like the Norak's by not expanding, but does not need to constantly waste resources by building colonizers that immediately get destroyed when actually in a real defensive situation. I replayed the Norak game and with the addition of the colonizer, the Norak's changed states and moved out of defend long term much sooner then 50 turns. So it played differently the second time around.

Recommend that the modpack AI makers seriously consider and test a separate queue for defend long term.

I also made a separate research queue for defend long term, so it would research defensive techs. Maybe could expand to include separate techs for infrastructure, etc. Might also help the AI.

Going to take a break from SEIV since it is the only game I have played for a year and try some Civ III, but since I still love SEIV I will be back.

Puke · #13 November 28th, 2001, 10:21 PM

Jourin, thats is an awesome development! depending on the design of minesweeper, you can probably get away with less than 6. BCs with MS4s should probably do fine with about 3, so that might be hard to balance between the early and midgame.

I really like the long term defense idea. can you post an example of the edited files for us to check out?

Thanks!

DirectorTsaarx · #14 November 28th, 2001, 10:28 PM

I agree with Puke: nice work Jourin.

If we can make the AI do human-like things in the different states (like researching and building defensive items when defending), we'll be closer to making a challenging AI.

I think some of the other modders have determined that the AI always needs at least one colonizer in the build queue for each state, otherwise the AI will stagnate and eventually lose.

Tampa_Gamer · #15 November 28th, 2001, 11:09 PM

Nice job Jourin! I tried separating some of the queues a few Versions ago with my TDM races. I was having mixed results with some queues stopping altogether but I beleive this was as a result of a bug that has since been fixed sinced. I will try your suggestion this weekend. As for Civ3, I have played two games and am now waiting for the patch before my next game due to several bug frustrations (patch rumor mill has it coming out next week).

Master Belisarius · #16 November 29th, 2001, 01:08 AM

Good work Jourin!
I'll test something about it, if some day I can get a free time!

God Emperor · #17 November 29th, 2001, 02:49 AM

I agree with Master Belisarius. The AI cannot/doesnt make use of specialised components effectively so its best to design their ships as "Swiss Army Knives". All of my races Attack ships include at least one minesweeper component whilst Minesweeper ships include limited weapons.
If you play strategic combat, the Swiss Army Knife ships tend to perform better too...

DirectorTsaarx · #18 November 29th, 2001, 10:03 PM

quote:
Originally posted by Tampa_Gamer:
Nice job Jourin! I tried separating some of the queues a few Versions ago with my TDM races. I was having mixed results with some queues stopping altogether but I beleive this was as a result of a bug that has since been fixed sinced. I will try your suggestion this weekend.

I'm not certain those bugs are entirely gone; one of the races in my current single-player game appears to be having trouble with construction. I need to check on that... thanks for the reminder...

edit: it's one of the stock races that's having trouble, not a TDM race.

[ 29 November 2001: Message edited by: DirectorTsaarx ]

Jourin · #19 November 30th, 2001, 07:06 PM

Thanks,

My test environment was a custom map where I paired the AIs together. The AIs, I used where Earh Alliance (me/AI) - XiChung (AI/me), Orks - Norak, Jraenar - Colonials, Krill - Sergetti, Rage - Pryrochette.

Test #1: Increased minesweepers to 6 - Did not achieve expected results. AI would build ships and not build minesweepers.
Test #2: Increased minesweepers to 6 and moved them to the top of the queue. Achieved expected results.

In test 2, the AI would discover a minefield. The AI would then send the Ready Fleet into the minefield (and be destroyed). Then it would send in the Ready minesweepers (usually just 2) and clear some mines but then be destroyed. The AI would then build the 6 minesweepers and a new attack fleet. The AI would send in the minesweepers and then the attack fleet. Not perfect but better.

Test #3: Separated the Attack and Prepare to Attack into two separate queues. Did not achieve expected results.

Test #4: Made Defend (Long Term) into separate queue. Queue was short and I will post the actual text file later. Had the AI build mines, minelayers, satillites, satillite layer, fighters, bases, weapons platforms, and ships. Cheated in the game to set up the conditions of a very weak Earth Alliance (scrapped all the ships) and then attacked with the Xichung. Earth Alliance definitely used this queue and put up a very good defense that stopped my XiChung attack. Monitored other AIs and found previously mentioned problem with Noraks.

Test #5: Added a colonizer to the queue (but only 1), and retested Noraks. After building the colinizer, they left the defend (long term) phase. My concern that I didn't test, does the AI automatically leave defend (long term) after a colinizer is built. If so, then putting instructions into the queue to build a colinizer may defeat the purpose of a separate defend (long term) queue unless colinizer is at the end of the queue and you want to trigger a transition to another state. Conversly, having no colinizer in the queue could lead to AI growth stagnation as seen with the Noraks.

After my tests I just played a normal game for about 85 turns, but discovered my strategy of just using minefields for defense while I explore and build my economy/infrastructure didn't work anymore and I (gasp) was loosing to the AI, so I did what any good human would do. I quite the game and started playing Civ3.

PS. Played at High bonus, low tech, 1 planet start and separated each AI by five systems, so they would not meet until after colinizing the second system.

Thanks for everyone's interest.

Jourin · #20 November 30th, 2001, 07:08 PM

Thanks,

My test environment was a custom map where I paired the AIs together. The AIs, I used where Earh Alliance (me/AI) - XiChung (AI/me), Orks - Norak, Jraenar - Colonials, Krill - Sergetti, Rage - Pryrochette.

Test #1: Increased minesweepers to 6 - Did not achieve expected results. AI would build ships and not build minesweepers.
Test #2: Increased minesweepers to 6 and moved them to the top of the queue. Achieved expected results.

In test 2, the AI would discover a minefield. The AI would then send the Ready Fleet into the minefield (and be destroyed). Then it would send in the Ready minesweepers (usually just 2) and clear some mines but then be destroyed. The AI would then build the 6 minesweepers and a new attack fleet. The AI would send in the minesweepers and then the attack fleet. Not perfect but better.

Test #3: Separated the Attack and Prepare to Attack into two separate queues. Did not achieve expected results.

Test #4: Made Defend (Long Term) into separate queue. Queue was short and I will post the actual text file later. Had the AI build mines, minelayers, satillites, satillite layer, fighters, bases, weapons platforms, and ships. Cheated in the game to set up the conditions of a very weak Earth Alliance (scrapped all the ships) and then attacked with the Xichung. Earth Alliance definitely used this queue and put up a very good defense that stopped my XiChung attack. Monitored other AIs and found previously mentioned problem with Noraks.

Test #5: Added a colonizer to the queue (but only 1), and retested Noraks. After building the colinizer, they left the defend (long term) phase. My concern that I didn't test, does the AI automatically leave defend (long term) after a colinizer is built. If so, then putting instructions into the queue to build a colinizer may defeat the purpose of a separate defend (long term) queue unless colinizer is at the end of the queue and you want to trigger a transition to another state. Conversly, having no colinizer in the queue could lead to AI growth stagnation as seen with the Noraks.

After my tests I just played a normal game for about 85 turns, but discovered my strategy of just using minefields for defense while I explore and build my economy/infrastructure didn't work anymore and I (gasp) was loosing to the AI, so I did what any good human would do. I quite the game and started playing Civ3.

PS. Played at High bonus, low tech, 1 planet start and separated each AI by five systems, so they would not meet until after colinizing the second system.

Thanks for everyone's interest.