Author Topic: Potential Spoilers: Aurora C# Enemy AI Ship Design and Tactics Thread.  (Read 8639 times)

0 Members and 1 Guest are viewing this topic.

Offline iceball3 (OP)

  • Captain
  • **********
  • Posts: 454
  • Thanked: 47 times
In the spirit of the new mentioned AI potential smartness due to Aurora C# heralding in better performance, I figure it's a good idea to start this thread up. The specific purpose of this thread being where we suggest ideas for ships and tactics for the AI to employ, for Steve to check at his leisure.
Mainly since we're lacking any sort of centralized "ship-meta", due to multiplayer being difficult, and our current iteration of AI being somewhat legged by this.
For the sake of simplicity and meaningfulness of ideas, let's try to stick to already existing weapon systems, or ones confirmed for upcoming versions of Aurora.
I'll start off with a few ideas:

-NPRs feeling a value of "threatened" by certain types of weapons systems and strategies, based on damage taken, perceived amount fielded by their enemies (such as missiles in flight and tonnage of ships mounting it (mitigated/amplified based on AI's salvage findings)). This can allow NPRs to build fleets against things they think might threaten them at some point, and making a singular potent strategy difficult to maintain if engaged in an extended campaign against an AI.

-NPR Ship: Pinnace.
A small fighter or FAC sized craft that may mount some combination of active sensors, passive sensors, gravity sensors. Always mounts a jump drive, and the most efficient engine it can.
This ship's designed to explore jump points and scout ahead of their gravity survey ships, alerting the controlling NPR of any threats and allowing them the barest intel on said threats, and have enough fuel spaces in range to justify going so far.
These designs in general are meant to be rather cheap, and not to run with any other ships in unison.


-NPR Ship: Breacher.
Designed to be uncharacteristically tough military design (for it's tech level), due to it being large and packing substantial shields and armor for it's size. It could mount light weaponry (for it's size), designed to strategically counter whatever technology the NPR feels more threatened by when it encounters you, as well as a few sensors. Anti missile designs onboard if the NPR considers itself threatened by such.
The main purpose of this ship is to draw fire, to test the waters against enemy ships of unknown capability, to waste enemy ordnance, and to be part of the leading assault squadron pushing through a jump point.
In general movement, could either sit with the main task group, or move at a position ahead of the task group at threatened bearings, to provide long range active sensor contacts for them and to take fire from enemies who do not know that that is what the ship is built for yet.


-NPR Designs: Troop transport.
Encourage NPRs to mount some amount of size of troop transport on nearly all ships, as well as to run boarding craft on occasion.
The main thing this overarching strategy would encourage is to:
--Make boarding parties a lot less potent, going both ways. Knowledge about the game that NPRs run both boarding craft and troop transport to counter such will encourage players to include such in their design to prevent hostile boarding.
--Make troop training and combat rating tech a lot more important across the board, as players will look for ways to save tons and increase the efficacy of their ship-based boarding combat defense. More potency per battalion/company means more security against boarding.

This will also make ground combat more relevant all around, as all warships will have at least some minor amount of ground troop carrying capacity, ground combat becomes an option just by incidental availability.

NPR Designs: Missiles.
Basically, allow AI to use varying combinations of ECM, armor, and sensors in their missiles. Perhaps allow them to specialize against tactics/designs they feel threatened by. (area defense fire, saturation of gauss or similar beams, fleets of fighters, respectively, but other situations can call for different things as well.)
 
The following users thanked this post: lordcirth

Offline Detros

  • Commander
  • *********
  • Posts: 389
  • Thanked: 26 times
NPR tracking station tactics:
During NPR generation:
  • Set your style with "wanted amount of DSTS at colony". It should be number somewhere between 5 and 20.

Each month check:
  • If level of Planetary Sensor Strength tech is lower than level of Engine tech
    • add research of Planetary Sensor Strength to the next batch of "to research next"

Each month check for each system with colony with factory power (industry/factories/construction brigades):
for SYS = given system, COL = colony with most  factory power in SYS
  • If amount of Deep Space Tracking Station at COL is less than "wanted amount of DSTS at colony"
  • AND there are no  Deep Space Tracking Stations in the construction queue of COL
    • add "wanted amount of DSTS at colony"/3 Deep Space Tracking Stations to the construction queue of COL
    • OTHERWISE pick random body in SYS, ideally some currently not in coverage of DSTS network of SYS (some weighted chance), transport random amount (from 3 to "wanted amount of DSTS at colony" / 2) of DSTS from COL to this new body 
« Last Edit: June 08, 2017, 07:18:31 AM by Detros »
 

Offline alex_brunius

  • Vice Admiral
  • **********
  • Posts: 1240
  • Thanked: 153 times
Regarding NPR Tactics what I feel is important is to give the AI a basic sense of tonnage/threat and not mindlessly throw ships away.

This is done by estimating a few things:

- Technology of an enemy can be estimated by comparing the speed of enemy ships & missiles observed to those of your own.

- Estimating how offensive your ships are based on previous encounters weight of missile/beam fire or salvage results ( scoring their threat from for example 0.2 to 3.0 based on how much tonnage they have dedicated to weapon systems and how deadly the salvo fire is ).

- Total tonnage of an enemy fleet can be estimated by knowledge of their known ships and amount ( If the NPR have observed you having a total of five 10k ton ships and fifteen 2k ton ships previously it will estimate your fleets to be worth 16k ton "threat" for each 10k ton ship it observe even before it's sensors can detect a 2k ton signature ).


A practical example:

The AI is observing an enemy fleet made up of eight 20k ton warships approaching in one of it's systems and is tracking it on resolution 160 actives from a long range scout.

Based on a speed advantage of 15% the fleet is estimated to be 1 TL above the AI and as such is given a x1.15 multiplier in perceived "threat". Based on previous encounters the fleet has been observed to be fairly balanced with a bit of offensive focus, so it's given a x1.3 multiplier in perceived "threat". Based on a total of known enemy ships  (108x 4k ton screens, 16x 20k ton warships and 12x 12k ton warships), combined with the fact that the actives can detect any warships larger then 8k ton in the fleet, the fleet approaching is estimated to consist of 8x 20k ton warships and 37x 4k ton screens (160k ton observed / 464k known tonnage that should be observable = fleet is estimated to be composed of 34.5% of the enemy empires warships ). This gives the enemy fleet an estimated calculated tonnage of 308k tons, which is multiplied by x1.15 and x1.3 for technology and offensive firepower. The AI NPR thus judges that it will need to amass at least 460k tons of own warships in a single fleet to be able to meet the threat. This could further be multiplied by the inverse of the racial aggressiveness scoring, such that an aggressive/reckless race can attack with less and a defensive race will want more just to be safe.

If the AI can meet or exceed the fleet size needed (without leaving colonies without defense), it tries to gather the fleet and engage. If not it tries to get as close as it can and defend either the next jump point (if possible to beat the enemy fleet to it), or the closest / likely targeted colony the enemy fleet is approaching.
 
The following users thanked this post: Happerry, Scandinavian, lordcirth

Offline sloanjh

  • Global Moderator
  • Admiral of the Fleet
  • *****
  • Posts: 2805
  • Thanked: 112 times
  • 2020 Supporter 2020 Supporter : Donate for 2020
    2021 Supporter 2021 Supporter : Donate for 2021
I stickied it.

John
 
The following users thanked this post: iceball3

Offline TCD

  • Lt. Commander
  • ********
  • T
  • Posts: 229
  • Thanked: 16 times
I guess one important question to discuss if how much we care about the AI cheating if it make it a better threat? alex_brunius excellent thoughts are prefaced on things the AI could reasonably know if it was a human player. But it does add a lot of extra calculation for Steve to make those estimates. Alternatively, the AI could just "know" your relative tech level and your total fleet size etc. That would reduce your ability to be sneaky and deliberately try to tempt the AI out to a battle it can't win. But it might make for a more satisfying challenge.

I think I'm okay with the AI cheating a bit if it makes it easier/possible to code in more intelligence.
 

Offline lennson

  • Warrant Officer, Class 1
  • *****
  • l
  • Posts: 76
  • Thanked: 10 times
Given the direction of making larger modules more efficient the NPC empires should be able to 'scale up' ship designs as their economy grows. Currently the NPC empires appear quite restricted in terms to the size of ships that they field. I think that it would be useful if the NPC empires had a ship template that can be scaled up as large as their economy permits. Enabling the NPC empires to build 'super capital' ships (i.e. one ship making up a significant fraction of an entire fleet's tonnage) probably is needed for them to make effective use of the newly strengthened shields.
 

Offline iceball3 (OP)

  • Captain
  • **********
  • Posts: 454
  • Thanked: 47 times
I guess one important question to discuss if how much we care about the AI cheating if it make it a better threat? alex_brunius excellent thoughts are prefaced on things the AI could reasonably know if it was a human player. But it does add a lot of extra calculation for Steve to make those estimates. Alternatively, the AI could just "know" your relative tech level and your total fleet size etc. That would reduce your ability to be sneaky and deliberately try to tempt the AI out to a battle it can't win. But it might make for a more satisfying challenge.

I think I'm okay with the AI cheating a bit if it makes it easier/possible to code in more intelligence.
It does mean a lot of prospective PvP obfuscation tactics might be nil.
Hmm, unless we make this a function of espionage teams, perchance? Make it so espionage can yield "fleet weapon usage by total tonnage" or "platform force value" without specifics? So for instance, an AI could use espionage on you or interrogate your crew (or vice versa) to become aware that you have a lot of sensors with high tracking speed multiplier/turrets with a specific tonnage gear without actually mentioning it's respective tracking speed, or that a fleet is mounting 200HS worth of reduced-size launchers, or so forth. It'll allow AI's and the player access to the more transcendental knowledge, perhaps AI moreso when you've cranked their difficulty, (hopefully) giving them access to more military academies.
 

Offline MagusXIX

  • Lieutenant
  • *******
  • Posts: 173
  • Thanked: 10 times
I never play with NPRs on, mostly because of how much it slows everything down. If that's fixed (or just much better) with the C# version, then I'll give it a shot.

(I mostly play vs myself.)
 

Offline alex_brunius

  • Vice Admiral
  • **********
  • Posts: 1240
  • Thanked: 153 times
It does mean a lot of prospective PvP obfuscation tactics might be nil.
Hmm, unless we make this a function of espionage teams, perchance? Make it so espionage can yield "fleet weapon usage by total tonnage" or "platform force value" without specifics? So for instance, an AI could use espionage on you or interrogate your crew (or vice versa) to become aware that you have a lot of sensors with high tracking speed multiplier/turrets with a specific tonnage gear without actually mentioning it's respective tracking speed, or that a fleet is mounting 200HS worth of reduced-size launchers, or so forth. It'll allow AI's and the player access to the more transcendental knowledge, perhaps AI moreso when you've cranked their difficulty, (hopefully) giving them access to more military academies.

Actually it would make more sense IMHO if this was not tied only to espionage teams, but instead to a general type of "knowledge of enemy ships" which is gained mainly from salvaging their warships.

Salvaging many warships of an empire should give you a pretty good idea about the general makeup of their ships. Things like their performance, techlevel and weapon percentage all included.

I do agree that interrogating crew could boost it, and it could also be boosted massively if you manage to capture ships ( making this more worthwhile except for RP purposes ).


Overall I feel this part of the game is a bit lacking. If I have destroyed half their navy, captured all their surviving crew and salvaged all their ships I should have an almost complete picture of what I am up against, and I should also know that I have destroyed roughly half of their navy by now and rough estimates on their ship building or the capacity of their yards.

Historically for example during WW2 the allies were able to estimate German tank production within a few percent accuracy by such a simple thing as counting serial numbers of parts in the tanks they were knocking out and salvaging in Normandy.



This could neatly also solve the problem of giving the AI enough knowledge without cheating, by just making sure they aggressively salvage stuff and capture surviving crew.
 

Offline Detros

  • Commander
  • *********
  • Posts: 389
  • Thanked: 26 times
Historically for example during WW2 the allies were able to estimate German tank production within a few percent accuracy by such a simple thing as counting serial numbers of parts in the tanks they were knocking out and salvaging in Normandy.
"Intelligence Update: Glasgow - Salvaging teams of SV Minnesota have found a bridge in the Wreck #27. The salvaged info suggests this wreck belongs to the 8th finished ship of class Anubis which appears to have been designed as Area Defence Carrier. Intelligence report on this class was updated. Other known wrecks of the same class have been marked as of Anubis class."

Effects:
  • designation set to ADC in Intelligence screen
  • minimal number of finished ships set to 8 in Intelligence screen
  • Wreck #27 renamed to "Anubis #8 wreck" in System map
 

Offline Senji

  • Leading Rate
  • *
  • Posts: 5
Why not specialized NPR in a weapon branch ?
It's add a variety and we'll need to modify our army.
 

Offline DuraniumCowboy

  • Warrant Officer, Class 1
  • *****
  • D
  • Posts: 88
  • Thanked: 7 times
I would try to make their behavior highly driven by a large set of parameters that could be accessible on the backend.  I would think quite a few players would be interested in tweaking and optimizing.  Add that and a way to add in some scripts, and you could get the community to do a lot of work helping with AI development.