For those of you who've been here long enough, you've seen this topic covered in great detail in past posts, and can probably dig up some material that deals with the topic at hand. If so, please let me know - I can't find anything, which is why I'm posting in the first place. Ok. So. Right now one of my game dev groups is working on what is turning out thus far to be a low-tech, somewhat vanilla Turn Based Strategy/Tactics game. I've been mining some of the older ideas brought up here, and I think I've come up with an element I'd like to try to introduce into our design (I'm not really the designer, so "try" is as much as I can do). The heart of the idea lies in what I'm calling "fuzzy" command structure - this is where the player acts at only one level of a hierarchy (in this case, the top level) and influences a set of officers who translate and put into action the player's decisions. These officers are required to be capable of autonomous action when necessary, and should ideally have some form of "personality" governing the character of their actions. Nothing new even in my estimation here - the RTS-heads have discussed this one to death over time. I would, however, like to extend this system using morale as a driving factor. Specifically, it would be great to be able to create an engine which is capable of simulating the mental state of officers and delivering their decisions with that information in mind. Many a conflict has been won or lost by breaking the men in charge instead of the men on the ground. If the decisions being made are of low quality, then the outcome, obviously, is going to suffer. The player's job then has to include not only making good decisions in terms of how to play the war, but also how to play the peace, or at least the sanity, of the officers in question. Do you think this extension has merit? What other variations on the theme are possible? Is there a stronger way to involve the player with the consequences of their decisions?

A variation I am looking at is command and supply lines.
For example, an army sent to the frozen north will have dispatch riders who will take messages to the nearest town to be relayed back (with delays?). The longer the distance or more dangerous the area the less chance the rider has to get the message back.
With no command line you cannot command or get information from thr army. Maybe the general sends a message to say 'Sickness. 20% casualties. Retreating to town X from town Y. Need reinforcements'. If the message doesn't get through your map shows them as in Town Y and with a full compliment of men.
If you receive no message in a certain amount of time there is an alarm and a '?' placed with the army on the map.
Each army will also 'eat' either from the surrounding farms (reduces support in that area) or from supplies. Supplies act the same as a messenger. Think Peninsular War where the French lived off the land but incurred the wrath of the Spanish while the Brits paid for their food and carried it with them (more expensive and slower marching).
If you have no command line then the General will use 'fuzzy' command. If the general is a hard-nose then he will force the soldiers to carry on despite the epidemic taking huge losses.
A 'softer' general will rest until the epidemic is under control.
The problem I foresee is player involvement. Maybe you can only control the army when they are in command so the delay would have to be dropped and a more abstract command line used.
I was thinking of 'storying around' the problem by having a limited number of telepaths who keep an army in command line, relaying your orders on battlefields (i.e. you control the battle) but end if the telepath dies.
Obviously, not too good for realism but that's the advantage of doing your own games (my world... my rules ;-)
Anyway, still very much in the pre-design stages

Hmmm. Interesting. In terms of reward cycles, I think there would need to be something the player could do to have an immediate effect. Maybe with this type of system there needs to be more of an operational scope, so that you're politicking at home to make sure that you have cash for supplies & weapons while you're waiting for opportune moments to send & receive messages. It doesn't sound like you're doing a modern game, but if you are then of course sat/radiophones and all those lovely items solve your communication lag 90% of the time. Your game definitely sounds like a stronger instance of fuzzy control than mine - but then I'm not concentrating so much on that aspect, as it doesn't appeal to me as a player.

I intend that the player have direct control of a few men, who are in turn controllers of the mass of forces. I think I failed to get to the heart of this in my first post. In addition to standard concerns re: morale - keeping your troops supplied, only fighting if you can win, and scheduling your officers some downtime - my thought is that there should be a set of attack styles or sequences specifically dedicated to breaking the morale of these officers. The player might have a measure of how will the officer is handling the stress, and if they continued to ignore it the troops under that officer would start losing effectiveness. Once it got to a certain point, orders would be mangled or even violated directly. It seems like this type of system would at least change the focus of play from strictly winning to a higher level of concern.

I guess to some extent this type of system is best abstracted to an operational game where the units represent officer+troops. I think, however, that the presentation of the effect could be more effective for casual players if they could see the troops themselves.

Just do a search on my nick and you should find some ideas relating to what you are looking for.

http://www.gamedev.net/community/forums/topic.asp?topic_id=85599
http://www.gamedev.net/community/forums/topic.asp?topic_id=92774
http://www.gamedev.net/community/forums/topic.asp?topic_id=84724
http://www.gamedev.net/community/forums/topic.asp?topic_id=119571

Actually, in most of these of these posts, you replied to them so they may seem familiar to you.


I think the real trick in having AI Commanders is figurint out how to implement their autonomy and learning capability. Simple FSM's aren't going to do a very good job. Moreover, even though your Commanders are automous agents, they should also be able to collaborate and share information with one another.

The more I thought about dealing with a command structure, the more I realized how important data compartmentalization was. What do I mean by that? Well, information is only privy to those agents which have access to it. Access to information can then be disseminated to other agents (including the player) only if communication lines are open. This sharing of information is vital in order to have all of your agents (and the units under their command) coordinate effectively. This compartmentalization is very important to recreate the fog of war correctly, while also validating the use of reconnaisance and intelligence gathering.

Unfortunately, the orders that the player gives may be ambiguous, which is something that computers can't handle well. Here, a tight definition of tactics and strategy is important. Tactics are essentially the implementation of a plan, whereas strategy is what you want to do in the first place. AI shuld be able to handle strategy fairly well, but the implementation of tactics can be quite difficult. The AI would have to understand what certain objects do (for example, that rocks provide cover from fire, or smoke makes enemy units unable to see them), or how certain moves can affect outcomes (for example, firing from an enfilade position, or flanking your opponent).

I think this necessitates a couple of things. Firstly, the agents need a "knowledge domain". This is a set of knowledge about basic tactics (maneuvers, formations, etc) and world knowledge (rocks provide cover, and big huge tanks are dangerous enemies to light infantry). Secondly, the agents need some kind of personality qualifier which can affect the morale of his troops. Morale acts as a self-preservation constraint to limit certain actions. This can actually be a good thing. How many times have you groaned when a unit in a RTS game was too stupid to find cover when an enemy unit that had a longer range pummelled it to death? Of course, morale can always come into conflict with an order, which is half the fun.

okay, why does opera not store the cookie information right?

That was me in the anonymous post above.

Quote: Original post by Anonymous Poster
Just do a search on my nick and you should find some ideas relating to what you are looking for.

Threads now bookmarked (such a useful feature that i never ever knew anything about...).

Quote: I think the real trick in having AI Commanders is figurint out how to implement their autonomy and learning capability.

We agree on autonomy, although in the scenario under discussion, I have no real ambition to implement a learning characteristic - the officers should have a well-realized personality, fixed throughout the game. The axes of change will be rather in terms of simple values, which will, at threshold values, trigger state transitions.

Quote: This compartmentalization is very important to recreate the fog of war correctly, while also validating the use of reconnaisance and intelligence gathering.

In our case, the fog of war rules are already determined satisfactorily, at least for the moment. The latter point, I think, is something I haven't touched on. In some ways, our group isn't concentrating on these concerns, due to the idiom of the backstory. However, it does seem reasonable to question how knowledge or lack thereof affects morale - in particular, its effect on the commanders who are on the frontline.

For an officer who is doing "well", defined by pushing back or causing casualties to the enemy without losing many troops of their own, I think a confidence factor would be appropriate. A confident commander could become overconfident, or alternatively if they're of a specific persuasion they might be susceptible to personal destruction by way of a sudden, massive loss in battle.

An officer who is doing "poorly", conversely, could be completely undermined by their inabilty to make progress. Or they could be subject to a rapid upswing witnessed upon a significant victory, by hook or by crook.

Quote: Secondly, the agents need some kind of personality qualifier which can affect the morale of his troops. Morale acts as a self-preservation constraint to limit certain actions. This can actually be a good thing. How many times have you groaned when a unit in a RTS game was too stupid to find cover when an enemy unit that had a longer range pummelled it to death? Of course, morale can always come into conflict with an order, which is half the fun.

The extension I proposed would attempt to introduce a direct attack capability on exactly this quantity, among other things. The men would be subject to the effectiveness of their commander

Obviously, Dauntless, you've been one of the primary proponents of this type of system so I'm not going to try to match your enthusiasm or depth of thought for this model of play, but I think that this type of "limited implementation" serves as a good bridge between the current total-control idiom and a more vertical stack of responsibility with the player at the top.

read up on MOO3. Clever command structures, AI generals, etc. just pissed people off. If there is a level of gameplay beyond the player's control, then to the player its just an annoying and confusing and fustrating weakness.

The issue of belief of control is very important. Notice however that I said belief of control and not just control in general. The difference is that if the player thinks that he can influence events which are only partially under his control, then the player is more willing to forego some level of absolute control.

There's an old axiom that says the level of control we have over our lives directly influences how happy we are with our lives. Ergo, we want that in our games. We want to know that our actions have predictable and deterministic results. The less that the results depend upon casual effects that were not due to our own actions, the more happy we will be.

First off, this is to be frank, an immature attitude. In real life, we can only control our own lives, and if we try to manipulate others to get what we want out of things, then we've failed to discover that happiness comes from within. For example, if I'm playing on a team and I did the best I could possibly do and yet my team lost, I should still be happy.

Secondly, things in life are affected holistically by the sums of its parts. All I can do is the best of my (the protagonists) abilities. The parts of the whole which aren't under our control we can only hope do the best they can too. But games want to make players happy by eliminating all the other parts that make up the whole. Under the theory that contrl=happiness, it also introduces the need to micromanage and create a model which no longer accurately reflects what we are trying to represent.

So how do we mitigate the desire of players to control everything? You give them either the illusion that they are controlling things, or you allow them to influence, either in-game, pre-game, or post-game, how the other autonomous agents go about their business. For example, if the player has the ability to select leaders with different personalties, and these personalties in turn help determine behavior, then the player has a hand in the control of these agents. In-game influence can be something like having the player be able to modify or influence behavior, such as providing morale bonuses or override subordinate decisions with direct orders (standing orders). Post-game influence can be handled if the game uses learning algorithms, and you can influence the fitness of the learning process by remarking which actions were good, and which weren't so good.

As I've long tried to champion here, games are not just about fun, they are also things that the player can experience which makes him feel better about himself in some way for playing it. If we always create games trying to pander to the lowest common denominator, then we will never break new ground, and games will eventually stagnate.

Hi,

Interesting thread - just a comment on the implementation side of things. Maybe this could be implemented by strategy being chosen through use of a Bayesian utility function. The utility of each action could then be a parametric model of morale i.e. a positive action (advance, taking terrain, whatever) would be an increasing (sigmoid?) function of morale, so when morale is high they are more likely to take this action. Then morale is a scale that depends upon other things going on at the time i.e. win battle : Morale += 10; running out of food : Morale -= 1 per day without food (so you could have morale lowering due to lack of food and the player has to do something to keep morale high - and as positive actions are linked to high morale, this loop could become self-perpetuating. This might also lead to the opposite effect - a confident commander might overcommit due to having high utility for aggressive actions (which would outweigh the negative aspects of the superiority of the enemy).

I'm thinking of something similar where different AI's have different utility functions for different actions (i.e. some are cautious, some are aggressive etc.) By building all this into a Utility based algorithm you shouldn't have to worry about lots of different FSMs, you just specify the utility functions for each commander and then let them sort it out.

Hmm, we'll see if this works in practice though.

Jim.