Naval AI - Part 2

[Edwin P.]

In SC1 the Naval AI never knew when to be aggressive and when to play defensively.

Example: Italian Navy

While playing with FOW off, naval and air superiority in the Med the Italian Navy would always assume a defensive posture and never seek to hunt down and attack allied naval units in the Med.

During the same game when the Allied forces achieved superiority the Italian Navy would venture forth to attack Allied Naval units.

To test the SC1 Naval AI in a custom 1939 Scenario I game the Axis 15 Subs in the Baltic Sea and they never ventured forth from the Baltic to attack Allied naval forces.

Additionally, the AI would not use its naval forces to bombard shore targets, even when their was no chance of taking losses or being attacked, in order to gain experience and season its sailors for battle.

On a "tactical" level in SC2 I am looking forward to seeing the AI utilize its naval forces in a more intelligent manner.

Ideally, the Naval AI would be able to select from 3 or more AI instruction sets - an AI for Naval Superiority in a region, an AI for Naval Inferiority in a region, and an AI for uncertain superiority in a region.

Naval Superiority = Aggressive AI

Inferiority = Defensive AI

Uncertain = Cautious Probing

[ December 15, 2004, 09:05 AM: Message edited by: Edwin P. ]

[Marky]

i agree with that

tho a ?

with the subs, did u give the axis control of the port of constantinople?

if u didnt, then of course, all those boats would be bottled up in the black sea

the AI definitel needs to be more agressive, i agree

ive left britain with NO GROUND TROOPSS, and the axis didnt even TRY to invade, and i know they could see cuz i they had fighters, and they were raiding near manchester, so they had pretty good L.O.S.

i like that instruction set too

makes perfect sense

how about -

Bold = Will try sumthing akin to the taranto raid

Kamikaze = will try anything to win any battle, and will continuously launch spec ops units - more on this in my next post

any thouts?

[Edwin P.]

The Baltic Sea is the sea bordering Germany, Sweden and Finland. This is where I placed the German subs in my mod.

The Black Sea is the sea bordering Russia, Turkey, Romania and Bulgaria.

As for the Axis never invading England if it was undefended you are right, in SC1 the Axis would never attempt a Sealion as long as Russia was in the game, perhaps because it could trigger an early entry of the US into the war. I would like to see this oversight fixed in Sc2. The question arises is how to do it in a manner that the AI can handle or how to penalize the Allies for leaving the UK totally undefended.

Perhaps, in a game with FOW on, for for the Human Allied Player each unit in the UK gives a 20% that UK units are protected by FOW (Fog of War). If you pull all of your units out of the UK there is a 100% that the Axis AI can see the location of all of your UK Units. If you leave only 3 units in the UK then the Axis AI has a 40% to pierce the Fog of War (FOW) covering all of your units. With 5 or more units in the UK your UK units are 100% covered by the FOW.

[pzgndr]

I am as curious as everyone else about what Hubert is going to come up with as far as AI and AI scripting goes. If the new AI flexibility proves to be as robust as the other new features in the Editor, we should all have a lot to look forward to.

FOW spotting range advantages for the AI based on difficulty level would help. Even a total FOW off option for the AI would be good. More information should result in better decisions, but not always.

AI scripts creating some sort of UK garrison requirement and/or other conditions might be useful for triggering a Sealion if the garrison drops too low. The current AI knows how to do Overlord (D-Day) fairly well once conditions are met so we should be able to get the AI to do other historical invasions. Norway, North Africa, Sicily, Italy, southern France, etc. all need to be implemented somehow in SC2. Plus other hypothetical invasions.

The basic AI with whatever weighted decision-making it has, even with FOW advantages and scripted AI and/or event triggers, may still calculate a Sealion to be a greater risk than it's worth. Cold hard number crunching is often unforgiving. Should we then somehow "force" the AI to make a "bad" decision, just so we get an occassional exciting game? We need to be very careful with this. Nobody wants to play against a reckless AI.

[Edwin P.]

Many thanks for the comments Pzgnder.

In SC2 it appears amphibious invasions will be much easier to an AI to launch with the new marine units.

Here are a few simple AI invasion options for the Axis AI to consider when the allied player leaves the city and surrounding hexes undefended.

1. Edinburgh Raid - Axis Marine Unit moves Adjacent to Edinburgh and takes it in one turn.

2. Bucharest Raid - Russian Marine Units invades from the Black Sea and takes Bucharest. ( I have done this to overly confident Axis human opponents many times, of course it only works once with each opponent)

3. Scarpa Flow Raid

4. Cairo Raid

5. Icelandic Invasion where Iceland is undefended by Allied units.

[Edwin P.]

The basic AI with whatever weighted decision-making it has, even with FOW advantages and scripted AI and/or event triggers, may still calculate a Sealion to be a greater risk than it's worth. Cold hard number crunching is often unforgiving. Should we then somehow "force" the AI to make a "bad" decision, just so we get an occassional exciting game? We need to be very careful with this. Nobody wants to play against a reckless AI.

Excellent point, the key here I think is to define what are the conditions for a Sea Lion.

I have played in games where the Allied player left the UK undefended after invading France and other games where the AI player sent all of his units to the Mediterrean leaving the UK undefended.

You also have games where the Axis strategy calls for an early Sea Lion and most games where the Axis focuses his efforts on defeating Russia first. So, I think that the AI has to begin by selecting a strategy and then altering that strategy based on events.

Example: 20% SeaLion Strategy > Position Forces > Sea Lion doable? Yes then Launch, No then switch to Russia first strategy.

Now what would encourage the AI to select a sea lion strategy - perhaps +10% if Most of UK navy is in the Med, perhaps +10% if UK is lightly defended, perhaps +20% if UK is totally undefended.

[Edwin P.]

One AI improvement that I would like to see is allowing the AI to know, not the location of enemy units but how many there are in a region.

For example:

I would like the AI to know how many Allied Naval units are in the Med. without knowing their location. Based on this it could select a strategy.

Example:

Italian AI has 6 ships in the Med. Allies have 1 ship = Italian AI selects Aggressive Strategy 90% of the time.

Italian AI has 6 ships in the Med. Allies have 12 ships in the Med = Italian AI selects defensive strategy 90% of the time.

With this system you can still surpise the AI as it does not know where your ships are but it will adjust its strategy according to the perceived balance of forces.

From a naval point of view, I would like to see the AI know the number of ships deployed in the Med, N. Atlantic, S. Atlantic, Baltic and Black Seas.

This knowledge would also be useful in writing AI scripts that select from a list of AI postures in a region - Highly Aggressive vs Aggressive vs Defensive vs Highly Defensive.

[ December 30, 2004, 07:59 AM: Message edited by: Edwin P. ]