Some suggestions for AI makers

Ideas and suggestions for how to improve the Warzone 2100 base game only. Ideas for mods go in Mapping/Modding instead. Read sticky posts first!
Post Reply
User avatar
Arreon
Trained
Trained
Posts: 324
Joined: 15 Feb 2009, 05:57
Contact:

Some suggestions for AI makers

Post by Arreon »

I have some ideas to hopefully improve the effectiveness of any AI that people make, whether it's Niker's BecomePrey, or the Dydo-AI, I feel that common sense could be added. I don't know if this is possible at the moment, but maybe someday, someone will make an AI that will incorporate these ideas into their gameplay. These ideas are:

1: When the AI gets hit by artillery, they know which direction it came from.

2: The AI can divide units into groups, based on their propulsion. Hovers can be used to travel long distances and secure important areas, half-tracks can be used as support for the main force, tracks can either be used to absorb damage or to defend important areas, cyborgs can be used to absorb anti-tank damage and be used for support, and VTOLs can be used to individually choose their targets (spread the damage from the Anti-tank weapons for tanks, for instance).

3: If the AI knows that its enemies have a lot of artillery, they can build their own artillery BEFORE they put up the CB tower.

4: The AI will use moves like flanking, ambushes, and sandwiches.

5: They will use mobile sensors and try to keep them out of harms way. For example if the AI is using a VTOL strike turret and enemy tanks approach, either have the VTOLS take care of the tanks, or move it out of sight before it's spotted.

6: The AI, in the case of VTOLs, will make AA units and move them to vulnerable areas that are under construction or some Aerial form of difficulty.

I know that some of these ideas are difficult to put in an AI, but maybe someday when a breakthrough happens.
OS: Windows 7 Home Premium 64-bit
Processor: AMD Phenom II X6 1045T (6 CPUs), ~2.7GHz
Memory: 8192MB RAM
KukY
Regular
Regular
Posts: 1859
Joined: 20 Mar 2009, 21:56

Re: Some suggestions for AI makers

Post by KukY »

My opinions :

1 - Good idea, but it is extremely hard to make.

2 - Also good idea, I hate when AI mixes Tracks and Hovers in 1 group so Hovers are slow like tracks. (Unless you press F11 to turn off Formation speed limiting)

4 - No.

3 , 5 , 6 - Yes, AI-s almost never use sensors and artillery, if they would they'll be double stronger. And they need to use more AA defenses. I made group of 10 Vegenace VTOL Pulse Lasers. 2 runs of that VTOL group and Transport with 10 Plasmite flamers, AI is dead...

There are really too much combinations that can kill AI-s in no time.
xahodo
Greenhorn
Posts: 8
Joined: 12 Sep 2009, 14:21

Re: Some suggestions for AI makers

Post by xahodo »

Perhaps I have a suggestion which could, well, turn AI's in very, very hard opponents.

The basic idea is simple: generalize to the top, and specialize to the bottom. Dividing responsibilities tends to make things clearer and easier to control.

--

At the bottom is the sergeant, this one is relatively dumb and reports its status and the information it has to its lieutanant, it also blindly follows the orders of the lieutenant.

A lieutenant has control over 10 sergeants and has some intelligence.
If the lieutenant has control over some factories it is responsible for which factory does what. If the lieutenant has control over some builders, it makes sure an appropriate amount of builders are on a certain job (depending on urgency).
If this AI has control over some combat units, it controls whether they attack or retreat. This AI requests for reinforcements with the captain and checks whether it's succeeding in its mission or not, it is also able of identifying what type of reinforcements it needs, if any.

Now, a captain has control over 10 lieutenants.
In case of a base/construction captain, it manages resources (i.e. divides control over the resources to its lieutenants, based on needs).
The army captain makes tactical decisions based on the information obtained from its lieutenants and give to it from the major above it.

The major has control over at most 10 captains, but it can be less.
An army major is bound to an area.
An army major determines how well an area should be defended based on: what's there, connections (how many other areas may be reached through it), threat value and tactical value. It requests for captains and relinquishes control of captains to other majors. It also pushes defensive buildings into the build queue, along with an urgency value. The army captains under its control determine what defenses its area needs.
The base/build major determines what should be built first. It also figures out whether it needs additional builders and can handle them.

Now, there are 3 colonels.
There are the army, construction and research colonels.
The army colonel arranges the majors in importance in several criteria: army presence needed, air defense needed, ground defense needed. It maintains a list of needs along with urgency as well as success statistics for specific units in certain roles.
The construction colonel tries to divide the resources it gets among base, defense and unit construction. It does so by determining which need is most urgent by communicating with the army colonel.
The research colonel tries to do as much research as possible with the resources it gets, guided by the strategy the general chooses.

The 3 colonels are controlled by one general.
The general tries to determine its strategy by assigning weights to certain tactics based on past success ratings of its units and tactics. It also attempts to balance research and construction resource consumption.

--

Now, how exactly something like this needs to be implemented... no idea. But I trust the general idea comes across and I hope it's of help to the AI designers.
stiv
Warzone 2100 Team Member
Warzone 2100 Team Member
Posts: 876
Joined: 18 Jul 2008, 04:41
Location: 45N 86W

Re: Some suggestions for AI makers

Post by stiv »

Now, how exactly something like this needs to be implemented... no idea
.

Ah, implementation. You seem to have hit on exactly why writing a good AI is so challenging.
User avatar
Zarel
Elite
Elite
Posts: 5770
Joined: 03 Jan 2008, 23:35
Location: Minnesota, USA
Contact:

Re: Some suggestions for AI makers

Post by Zarel »

So why would that be better than any existing AI? You've pretty much described basically what an AI does, except you're calling some of its functions "captain", some of its functions "colonel", etc. I hope you realize renaming functions doesn't actually cause the AI to behave any differently. O_o Except for splitting up different parts of the map for different "major"s - that can only be worse for it. I mean, it's not like a computer has a limited attention span.

And why 10 lieutenants? Your average Warzone army can generally only be split 3 ways before they get too ineffectual to do anything. You can split trucks a few more ways, but by the time you have the money to build outposts in ten different places, you've either already won, or your outposts are absolutely useless and will get steamrolled the instant they meet an enemy army, making the whole endeavor really quite pointless, anyway. And even splitting the army 3 ways is only sensible in special cases; you generally want to attack with everything you have, when you attack.

I mean, having such a command structure in real life only makes sense because each human as a limited attention span. An AI has no such limitation.
Post Reply