Denizen Herding Behavior

This article is about getting your denizens to herd together to a greater or lesser degree and is by Russell Ackerman. You can download the full code by downloading the game Ascii Wilderness: http://asciiwilderness.blogspot.com/p/ascii-wilderness.html check out "ghoul.lua"

I thought I would share my success at getting my deer to herd in my roguelike, I did it by doing this: This article relates to the LUA programming language.

The deer herd together by making a sound each turn, and then when they move to another square, they value those squares according to how much sound there is (but only checking sound for a fraction of the turns that it moves) and move onto the square with the most sound from it's "friends"

"makesound" currently gets called once per turn per creature, soundscape gets nilled out each game turn and re-initailized with [x][y][z] values of the map, prepared to take new entries., and moveenemytowards selects a square based on sound and distance to destination information. the strength of herding is mostly controlled by monster.herdingtendency. Sound propogates out from a square, dropping information about itself. The function is as follows:

Intensity is generally an integer. At intensity 20, the sound will travel outwards 20 squares and use a lot more processor than intensity 5, per se.

function funcs.makesound(intensity,activator,x,y,z)
	local objectloc = data.objectloc --contains [x][y][z][object with internal data]
	local working, future = {}, { {x,y,z} }
	local soundscape = data.soundscape --where your sound data is stored, per tile [x][y][z][the object][thesound - arbitrary key][various keys like strength of the sound]
	local finished = {}
	local insert = table.insert
	local tiles = data.tiles
	local checkforedgeofmap = funcs.checkforedgeofmap --so sound doesnt go off the edge of your limited world.
	finished[x] = {}
	finished[x][y] = {}
	finished[x][y][z] = 1
	 soundscape[x][y][z][activator] = soundscape[x][y][z][activator] or {} --for moveenemytowards
	 insert(soundscape[x][y][z][activator],{["strength"] = intensity})
	local counter = intensity
	while(counter <= intensity and counter >= 1) do

	counter = counter - 1 --this keeps track of how strong the sound is when it gets to the square.
	working,future = future, {}
	local x,y,z
	for k,v in pairs(working) do
		for x = v[1]- 1,v[1] + 1 do
			finished[x] = finished[x] or {}
		for y = v[2] -1,v[2] + 1 do
			finished[x][y] = finished[x][y] or {}
		for z = v[3] - 1,v[3] + 1 do --now we're iterating over the nine squares around the center square, on the first iteration.
			if z < data.amountofzlevels and z >= 1 and (not checkforedgeofmap(x,y)) and (not finished[x][y][z]) and 
                        (not tiles[x][y][z]["filled"]) and (not tiles[x][y][z]["open"]) then
				finished[x][y][z] = 1
				insert(future,{x,y,z})
				soundscape[x][y][z][activator] = soundscape[x][y][z][activator] or {} --for moveenemytowards
				insert(soundscape[x][y][z][activator],{["strength"] = counter})
			end
		end
		end
		end
	end
	end
end

---

Data about the "soundscape" is stored, for now, just with the data of who made the sound and how intensly it was heard at the square. soundscape[x][y][z][the object that activated it][irreleveant key][strengthofsound] and any other pertinent variables about the noise would also be stored there.

To use this data in pathfinding to get your denizens to herd, simply do the following:

1/x times that the creature activates pathfinding to choose a square with the shortest path, simply weight the RELATIVE VALUE of those squares. Squares with more noise generated by creatures who are the same species as me should be weighted with an amount relative to the total noise on that square from those animals except myself, weighted to be MORE valuable, but only one out of x times - this fractional usage allows the creature to break from the herd and stop herding behavior once it cant hear any sound at all from those of its own species. Its a simple matter of using a function like "total_sound_at_square_except_me(me,x,y,z).

My code for moving the denizens is at follows. the table "Distances" is a table of data in the form distances[x][y][z] = amount, where distances[x][y][z] is the distance of a particular square to the destination square. It's a little hokey but it's pretty simple too:

All I do is subtract sound from distance, and select the square with the "lowest" relative value to move the denizen to.

function funcs.totalsoundfromob(ob,x,y,z)

	local soundscape = data.soundscape

	local totalstr = 0

	for k2,v2 in pairs(soundscape[x][y][z][ob]) do

		totalstr = totalstr + v2["strength"]

	end

	return totalstr

end


function funcs.totalsoundatspot(ob,x,y,z) --totals sounds at spot from "my species"

	local soundscape = data.soundscape

	local sound

	local totalstrength = 0

	local numfriends = 0

	for k,v in pairs(soundscape[x][y][z]) do

		if ob["species"] == k["species"] and k ~= ob and (not k["isplayer"]) then --not k isplayer for sanity purposes.

			sound = funcs.totalsoundfromob(k,x,y,z)

			totalstrength = totalstrength + sound

			numfriends = numfriends + 1

		end

	end

	return totalstrength, numfriends

	end

		
--ob contains ob.x and ob.y and ob.z, targ contains targ.x,y,z etc.
--if called without "targ" object, it moves to a random square.

function funcs.moveenemytowards(ob,targ)

	local numfriends

	local soundonhomesquare, numfriends = funcs.totalsoundatspot(ob,ob.x,ob.y,ob.z)

	local distances

	if targ and targ.name then print("Targ name "..targ.name) end

	if targ then funcs.pathfindfromto(ob,targ) distances = ob.distances end --gets a path... checks for a new path every so 
often.

	local soundscape = data.soundscape

	local totalsound

	local oldx,oldy,oldz = ob.x,ob.y,ob.z

	local x local y

	local z

	local lowestdistance = {distance = 1000000 ,x=oldx,y=oldy,z=oldz} --HARDCODED LIMIT

	local iterations = 0



	local movementintelligence

	if ob.movementintelligence then movementintelligence = ob.movementintelligence

	else movementintelligence = 20 end



	while(iterations < movementintelligence) do

		iterations = iterations + 1

		x = math.random(ob["x"]-1,ob["x"]+1)

		y = math.random(ob["y"]-1,ob["y"]+1)

		z = math.random(ob["z"]-1,ob["z"]+1)

			if ((not targ) and (not funcs.checkforblockpassageofpath(x,y,z))) or (targ) then

			local relativevalue = 0

			if ob.activateherding then totalsound = funcs.totalsoundatspot(ob,x,y,z) end

			if targ and not distances then return end --at destination already

			if ((not targ) or (distances and distances[x] and distances[x][y] and distances[x][y][z])) and not 
funcs.checkforedgeofmap(x,y) then --was and distances[x][y][z]

				if targ then relativevalue = distances[x][y][z] + relativevalue end

				local soundvariable

				local usesoundvariable = true

				if ob.tightgroups then soundvariable = ob.prefersoundlevel * numfriends end

				if ob.loosegroups then soundvariable = ob.prefersoundlevel end

				if ob.prefersoundlevel == 0 then usesoundvariable = false end 


				if (not targ) and ob.activateherding and soundonhomesquare == 0 then --if i dont hear anything and 
have no target, choose a random square.

				relativevalue = relativevalue - (math.random(1,1000) * 10)

				elseif ob.activateherding and (usesoundvariable and (soundonhomesquare < (soundvariable)) or not 
usesoundvariable) and soundonhomesquare > 0 then --if sound is lower than threshhold, herd a FRACTION of the time..

				if math.random(1,ob.herdingtendency) == 1 then relativevalue = relativevalue - (totalsound * 10) end 
     --herd according to a fraction of times ob.herdingtendency should be 10 or 20 or 5 or whatever. If I'm beyond the sound range 
     --by not herding, I won't herd from that point until I hear more sound..

				elseif (not targ) then --if I don't care about sound, move randomly.

				relativevalue = relativevalue - (math.random(1,1000) * 10) --was +

			 	end

				if relativevalue < lowestdistance["distance"] then --go towards high strength sound

				lowestdistance = {["distance"] = relativevalue,["x"]=x,["y"]=y,["z"]=z}

				end

			  end

		  end --ends if not targ...


	end --ends iterations over intelligence...

	funcs.moveobject(ob,lowestdistance["x"],lowestdistance["y"],lowestdistance["z"])

end

---

--It uses "soundonhomesquare" to prevent a bug that caused creatures to prefer to stay a certain distance AWAY from each other. LOL. --so as you can see the "relative value" of the square gets changed depending on how much sound is detected nearby.

Hope this helps someone! Check out my game Ascii Wilderness, which is open source LUA.

http://asciiwilderness.blogspot.com/p/ascii-wilderness.html

In this game, the deer properly herd together based on their internal variables. This code also depends on some internal herding variables from the participants such as "activateherding = 1" and "tightgroups" or "loosegroups" = 1 and "prefersoundlevel" = amount and also "herdingtendency" = 3- 10

Using this code, my ghouls will wait up for other nearby ghouls before closing in for the attack. Sweet!

By the way, this is processor hungry, and a good solution would be some kind of "auditory memory" for your denizens, so sounds wouldnt have to be "generated" each turn - which is where your processor gets used up, if you have lots of denizens all generating sounds, thats a lot of squares to check and values to assign. I'm planning some kind of auditory memory now for Ascii Wilderness.