Author Topic: Response to Improving AI Strategy  (Read 3923 times)

Offline zeusalmighty

  • Full Member Mark II
  • ***
  • Posts: 151
Response to Improving AI Strategy
« on: December 12, 2018, 02:36:40 am »
TLDR: This is a response to the post regarding ways of improving AI strategy. My response was too long to submit in that thread so I posted it here. 

Link to the original post: https://forums.arcengames.com/ai-war-ii/improving-ai-in-combat-strategy/

General Overview:
The roles of the hunter and the warden need to be more clearly delineated (e.g., classic’s special-forces riot ships). To this end I propose to make the sentinels to be the Jack-of-all-trades, while Hunters and Wardens specialize in offense and defense respectively. By focusing on their distinct roles, special tactics can be allocated to each based on their respective strengths (waves tactics doesn’t apply to wardens, for instance). I divide my comments based on the main AI opponents, some attention to general functions as well as giving specific scenarios when further context is needed.
***

Sentinels:
1. Can engage in “General Tactics” (see below)
2. Raiding Waves: Sends some waves (indistinguishable from normal waves) that purely consists of raiders and raid guardians that should split into “raiding parties” upon arrival to do a multi-planet attack—raiding parties should engage in “general tactics.”
3. Coordinate with Hunters (see below)

Hunters:
1. No defensive units—gravity guardian, spider guardian, temperamental guardian, fragmenting guardian, and vanguards are banned
2. Diversion: When a CPA or exo-galactic attack is imminent, Hunters will send small squadrons to medium defended planets (those without fleet usually) and hit-and-run resource collectors. (The idea is that CPA’s are big threats which will occupy players’ attention, but the Hunters will confuse the player by attacking several places at once, possibly obscuring where the CPA is coming from, and hurting the players’ economy)
3. Raiding parties: Basically, the same as the Sentinels version, but emphasis entirely on CPA’s (since it doesn’t use direct waves). Raiding parties are synchronized with the next available wave, randomly choosing to attack before, during, or after the wave launches—*this interval needs to adjust for the approximate time the Hunters can get into position.
4. Flank: Consolidate significant amount of available of Hunter forces to launch a wave at a highly valued planet, randomly choosing to attack before, during, or after the next non-direct wave. Goal should be to destroy the command center and then either retreat or engage in “general tactics”

Wardens:
1. No offensive units—tractor guardian, raid guardian, and raiders are banned
2. Usurper Exception: The wardens are permitted to assist a planet with a usurper present (since the usurper is supposed to reclaim the planet, it seems appropriate that the wardens can do special convey missions (**actually, that would be really cool to see a CPA of usurpers escorted by wardens). Might be good if the wardens suffered attrition damage if the usurpers die.
 
General Tactics:
1. Economy Harass: Divide and conquer resource collectors (metal and energy harvesters), retreat or regroup to continue harassment on adjacent planet
2. Tachyon Hit-and-run: Destroy detectors, retreat, repeat (this behavior is to facilitate infiltration of cloaked ships into player’s territory; see also special tactics)
3. Kill Irreplaceable: Focus fire an irreplaceable structure (this may be moot currently)
4. Deep Strike: Target HW or nearby planet, ignoring player defense on in-between planets (Goal should be to rush target by overwhelming force—*must meet high strength threshold to initiate)

Special Tactics:
1. Cloaking Behavior: Cloaked units should gather and attempt to hide on a planet near HW (using “scout behavior”). They will become active when the have sufficient strength and synchronized with next available wave. Cloaked units of any AI faction should act in tandem (so Hunter and Sentinels should synchronize cloaked forces) and can engage in “general tactics” (with some self-evident exceptions)
2. Anti-bubble shield units (Plasma guardians, stingrays, vicious raid starships…) should focus target forcefields, prioritizing nearest to command station (if applicable).

Miscellaneous Suggestions for Increasing Difficulty:
1. AI Strength Proportional to Minor Faction Difficulty: This probably needs a discussion unto itself, but basically, I think we need to base the AI’s response to minor factions by making it proportional to the difficulty of the minor faction. In other words, if the Marauders are at 10, the AI’s response is 10, regardless if the player sets the AIs’ difficulty each at 4. Otherwise, the player basically handicaps the AI out of being relevant. I suspect that this would be the simplest way of balancing the minor factions at the most general level.

2.Mines for the AI:
  a.Change Stealth Guard Post into Mine Field Guard Post, which has a variety of mines orbiting it (~avg. fleet ship range), 85mm armor, and “cluster rocket” (i.e., long-range grenade launcher). (Mines should counter fleet ships, guns should focus things beyond the mines range, especially concussion corvettes.) ***Stealth Guard Posts currently have no purpose, so something has to happen to them.

  b. AI should be able to place “patches” of mines, spaced in zones in the area between wormholes. Patches would ideally be randomly picked from a set of geometrical patterns—and at the very least—not just randomly scattered.


« Last Edit: December 12, 2018, 11:55:05 am by zeusalmighty »

Offline AnnoyingOrange

  • Jr. Member Mark II
  • **
  • Posts: 71
Re: Response to Improving AI Strategy
« Reply #1 on: December 12, 2018, 06:36:46 am »
In other words, if the Marauders are at 10, the AI’s response is 10, regardless if the player sets the AIs’ difficulty each at 4. Otherwise, the player basically handicaps the AI out of being relevant.

Isn't the entire point of minor faction intensity to give the player more fine grained control over the difficulty level they want to play at?
I suppose you could have factions with two intensity settings (one for benefits and one for maluses), or factions that compliment each other such as Alt Champion Progress and Nemesis in classic.

Offline Draco18s

  • Resident Velociraptor
  • Core Member Mark V
  • *****
  • Posts: 4,251
Re: Response to Improving AI Strategy
« Reply #2 on: December 12, 2018, 10:21:27 am »
General like of the ideas here. No specific comments.

Offline zeusalmighty

  • Full Member Mark II
  • ***
  • Posts: 151
Re: Response to Improving AI Strategy
« Reply #3 on: December 12, 2018, 11:41:06 am »
Quote from: zeusalmighty on Today at 02:36:40 AM
In other words, if the Marauders are at 10, the AI’s response is 10, regardless if the player sets the AIs’ difficulty each at 4. Otherwise, the player basically handicaps the AI out of being relevant.

Isn't the entire point of minor faction intensity to give the player more fine grained control over the difficulty level they want to play at?
I suppose you could have factions with two intensity settings (one for benefits and one for maluses), or factions that compliment each other such as Alt Champion Progress and Nemesis in classic.

This point needs more clarifying. I think raising the minor faction difficulty should raise the "floor" of the Ai difficulty proportionally, whereas the AI difficulty adjusts the ceiling. There should be two variables at play. But the basic reasoning behind this is that AI seems to be struggling to deal with multiple factions at once, and often sends insignificant waves to that do nothing or, in the nanocaust case, just feed the factions

Offline AnnoyingOrange

  • Jr. Member Mark II
  • **
  • Posts: 71
Re: Response to Improving AI Strategy
« Reply #4 on: December 12, 2018, 02:59:49 pm »
But the basic reasoning behind this is that AI seems to be struggling to deal with multiple factions at once, and often sends insignificant waves to that do nothing or, in the nanocaust case, just feed the factions

Then rebalance the AI response, redoing the entire difficulty system (as of now difficulties are discrete, with no idea of floor or ceiling) seems excessive at best.

Offline RocketAssistedPuffin

  • Arcen Volunteer
  • Sr. Member
  • *****
  • Posts: 260
Re: Response to Improving AI Strategy
« Reply #5 on: December 12, 2018, 04:25:13 pm »
This point needs more clarifying. I think raising the minor faction difficulty should raise the "floor" of the Ai difficulty proportionally, whereas the AI difficulty adjusts the ceiling. There should be two variables at play. But the basic reasoning behind this is that AI seems to be struggling to deal with multiple factions at once, and often sends insignificant waves to that do nothing or, in the nanocaust case, just feed the factions

Reconquest Waves seem to be working again somewhat, and I'm going to test them properly in a small bit. In theory, they should let the AI fight factions a bit more, especially the Nanocaust. There is also the fact that because these were never used, all of the budget that went into them was wasted - in the end meaning the AI was fighting handicapped all this time.

As for the ideas as a whole, the point of having Guardians generally be delegated to specific things I like in theory, though getting this done in a reasonable manner might prove tricky. Barring Raiders and Vanguards from things would be the worst part to figure out. I believe this would likely extend to Astro Trains that spawn units for the Hunter and Warden as well. In fact, I believe they already do, but can include the relevant Guardians pretty easily.

There is the problem noted in Discord I think that any Sentinel unit can become threat, then join the Hunter, including units you would have barred from there. This is likely unavoidable.

Mines...I had thought of them, but I wouldn't try it for now. That's something new entirely to figure out.

The general abilities of the AI factions I can't do. I appear to ah...not be so good at the whole program thing!
« Last Edit: December 12, 2018, 04:27:27 pm by RocketAssistedPuffin »
Autistic, so apologies for any communication difficulties!

Offline zeusalmighty

  • Full Member Mark II
  • ***
  • Posts: 151
Re: Response to Improving AI Strategy
« Reply #6 on: December 12, 2018, 04:35:43 pm »
Mines...I had thought of them, but I wouldn't try it for now. That's something new entirely to figure out.

In the long run I think mines are desirable in the AI's toolkit. It will provide more "puzzle" elements to AI planets (need to avoid that area) and makes sentry starships have more utility to assist your offensive fleets. Mines were definitely great and fun in classic, imo, and I feel that they could have more impact here (being able to upgrade mines sounds cool to me)