Return

Here's another thing to consider TW about GC. (ground control)

It's going to come with a very powerful, user friendly editor called GenEd.

With this Editor you will be able to create Mission Maps with A.I. scripted specifically to that map and only that map.

This is a category of Maps that has received very little attention in the past but I believe will come into its own with Ground Control.

To make a "General Purpose Skirmish A.I." that poses a real challenge to players equivalent to playing against good human players on ANY Map is something of a "Holy Grail". Especially when you cannot "Spawn Reinforcements" and rely on "Canon Fodder" and "Ant Swarm" waves.

Within those constraints such an A.I. would have to be "Dynamically Adaptive Tactically".

Among other things, I believe, you would need a sophisticated "Threat Assessment" scheme, an ability to correlate from a DB of Tactics and also be able to modify on the fly failing Tacs or counters & only press the offensive if losses where not inordinate.

Here's a past discussion thread that deals

with this very provocative aspect of RTS A.I. - Threat Assessment:

http://www.rtsempire.net/wzpbb/forumdisplay.cgi?action=displayprivate&number=1&topic=000090

Note Member Anolis & Rastas suggestions for ways to implement "Threat Analysis".

Also related though it came up in the context of Commander Turret A.I. is this post by DarKseid D:

quote:

--------------------------------------------------------------------------------

Darkseid_D

Xtreme Gamer

posted 06-18-1999 07:05 AM

--------------------------------------------------------------------------------

AI a rambling muse.

Im not all that hot at coding but here's what I think you mean for a Commanders AI, in a pseudo code sense.

Check settings:

1) What retreat level

2) What aggression level (return, hold, ffa)

3) What movement can I perform (hold,guard,pursue)

Check current objectives:

1) Am I moving somewhere

2) Am I moving to attack

3) Am I targeting/attacking

Check Current Visibility:

1) What enemies are in sight

2) What enemies are in range

3) What enemies can I put in range with a small movement. Take moveto{} as its default zone, it can 'move' say 8 units away to engage.

Check current locality:

1) Are the enemy buildings in sight

2) Are they in firing range given my orders

3) can I put them in firing range

4) do I have permission to fire

Check Status:

1) am I under fire

a) How do I respond within my settings

i) Complete current action

ii) Disengage building and respond

iiI) retreat to point A / safe zone

2) What's my armour

a) head to repair facility

b) whistle up a repair unit

c) head to a repair unit

That's kind of vague but you get my drift

Once a UNIT is assigned to the commander, you could cut its processing over head considerably. Slave its 'thoughts' to that of the commander, and share its tactical data. This means the command knows what its units can see and its less likely to allow 'stragglers' to be picked off. The slave units move to where the commander is, using its pathfinder routine (with some flex built in) and target what the commander says.

This would shift you more toward a parallel type of AI rather than multi thread. Ie one larger routine doing tasks for many children, over each child having its own routine.

Each status check for the commander has a different weighting value which determines their sequence.

Adjusting the AI's 'personality' would also reflect in that sequence.

IE a reckless 'human wave' type ai would set his ai to do or die and this would start ignoring the armour check.

Extra status checks would need to be employed.

Things to keep his tacgroup together, ie slow down all units to allow the slowest units to keep up.

Telling AA units to spread, or keeping heavier indirect fire units to the rear of a formation.

tricky, but quite possible....

Ds

-----------------------------------------------------------------------------------

With the continued advent of more powerful CPU's becoming an inexpensive commonplace quick calculations that wouldn't bog down the flow of Real Time Play are a reality.

The real issue though, IMO, is the advent of Cheap, Fast & Wide Net Bandwith and its becoming more the rule than the exception.

In the end it just may NOT be "Cost Effective" to develop "Dynamic, Adaptive, Tactical Skirmish A.I." when you can "Drop-In" in the middle of a Live Multiplay RTS Game like you will be able to with Ground Control for a quick Skirmish against the best Human Opponents.

That's really heart of this issue, I think - the omnipresence of the Net will re-define the Single Player Game Feature Set.

-RJ