[ian-reeds-games] Re: helping players to help the AI
- From: "Allan Thompson" <allan1.thompson@xxxxxxx>
- To: <ian-reeds-games@xxxxxxxxxxxxx>
- Date: Mon, 1 Apr 2013 17:58:08 -0400
Hi Ian and Carlos,
If this is what I think it is, then this is very exciting!
I just wanted to get some clarity on some thing first.
The ai_enemy_within will work just fine, thanks for considering that.
The random selection of skills, where each skill has an equal chance of being
selected is not part of this, right? I mean a priority list would kind of ruin
that, so I am assumeing that is a diffrent and separate flag.
So for Carlos's idea, there is a list of skills, and they recive priority based
on order, which overrides, so to speak, the hard coded groupings of skills Ian
has implemented int he game, did I get that right?
so if that is the case, then units would be getting these priority lists, and
they would perform them based on whatever the map maker wants.
So in other words, a warrior woudl be given a priority of attacking skills
first, while a wizard would be given a set of spells first, and a range might
have ranged attacks first, and a cleric would have healing first, I mean, in a
general sense.
Or am I totally getting the wrong idea from this? I am having a hard time
wrapping my brain around it. I was thinking range was the issue, but I was way
off, so Sorry Carlos.
al
al
"The truth will set you free"
Jesus Christ of Nazareth 33A.D.
----- Original Message -----
From: Ian Reed
To: ian-reeds-games@xxxxxxxxxxxxx
Sent: Monday, April 01, 2013 4:10 PM
Subject: [ian-reeds-games] Re: helping players to help the AI
Carlos, I think this thread is a good idea.
Let me give some first thoughts.
I really like your implementation of the first idea.
I might rename the flag to ai_skill_order or something else.
The way AI currently works is somewhat like this except that I have
predefined the groups based on the type of skill.
I'm doing this from memory so the order might be slightly off.
Basically the AI groups skills into these groups.
move,
summon,
heal,
uncurse
buff,
curse,
tile curse,
uncurse,
attack
defend.
Then it uses the groups for 2 things:
1 To determine how to use the skill and whether it is worth using, such
as recognizing that a buff should not be cast on a unit who already has
the effect that buff grants or recognizing that nobody is missing enough
health to warrant healing.
2 Determining the order in which to use it's skills, for instance
favoring healing and buffing over cursing and normal attacks.
The list above is the order the AI currently uses.
So the new flag will let you override the order and will add randomness
for any skills within the same group and I will need to keep the old
logic for whenever you don't specify the new flag.
One interesting thing to point out here, and you brought it up as a
concern, is range and how moving works.
Currently move skills are always placed at the top of the order.
But moving is only considered necessary when the AI is not in a position
where it can use a skill worth using and then it moves to the nearest
point where it can use any of it's skills worth using.
So it has the range issue you described.
In Tomb Robber Freddy would move as close as he can to perform any of
his actions which would always be pistol before attack.
And leaving move in the very last group actually works great as long as
the AI loops so that it again tries the first group after a successful move.
Perhaps the additional fix is to change move so that instead of moving
to the nearest point that it can use a worthwhile skill it instead goes
through the list and tries to move somewhere that it can use skills in
the first group containing attack and if it can't move that far in this
turn it would then try the second group containing pistol.
It will take me a while but I'm sure I can make this work, though it may
cause some extra thinking on the AI's turn.
I think there are trade offs here. The choice is really between
favoring the best skill (meaning the skill in the first group) or
favoring performing some skill this turn.
Imagine a different scenario where a unit can heal or attack and starts
in a position 1 space away from an enemy and 10 spaces away from a unit
that suddenly needs healing.
It can move 3 spaces in a turn and has this set of flags defined:
random_skills=heal|attack|move
In this case, instead of moving one space to use the attack skill he
will move 3 spaces toward the unit that needs healing, but will not
perform any worthwhile actions this turn.
So it's really a choice between whether to favor performing some skill
this turn or to favor getting closer to performing a first level skill
at the cost of sometimes not performing any skill in a turn.
I guess this could be yet another flag like ai_favor_best_action versus
ai_favor_some_action.
Well, I couldn't find a clear way of describing all that so I hope it
made sense.
Regarding your second item:
I think this problem is because the AI determines there is no path it
can follow to use a skill.
For instance, if there is one bridge across a river and a ranged AI unit
is standing on it then there is no path across the river.
Then if the ranged AI unit can shoot from where it is then it has no
reason to move.
And 100 melee AI units behind it and far away from the river won't move
at all because when they use path finding to determine how to get
somewhere they can use their melee attack they find that as the map
stands there is no path to where they can use their attack skill.
Fixing this would be something like determining that if there is no
valid path then they should choose a path that at least gets them closer
to the enemy.
For your third item it actually sounds like there is a bug when buffing
units who share a tile.
I'd have to look deeper to know for certain though.
Allan, I just read your email.
I think that letting map creators define the skill groups and priorities
is a good idea.
The problem you outlined is a little specific to summoning since the AI
considers any empty space to be a good space to use a summon skill
whereas you want it to only summon shuttles when it is close to an opponent.
So I think your idea should actually have a flag on the skill itself
such as:
ai_enemy_within=3
Then the AI would never use this skill unless it was within 3 tiles of
an enemy.
I think this would blend well with the grouping and randomization of
skills you and Carlos brought up.
Ian Reed
Other related posts:
