Finding the bordering regions

If you look at this screenshot you will see multiple regions separated by a brown line, I need to know the surrounding regions of a current region.

Here is another picture of what I’m attempting to find out, the green one will be our selected region and the red ones will be the ones that I want to find out about.

I have over 250 regions and I don’t really want to go through and attach each region to the ones that surround it, so would it be possible to make a system for this?

1 Like

Did you create these regions yourself? Is there an underlying datastructure we can use or are these just parts?

Possible Solution


You’ll probably have to run a geometrical algorithm of getting the smallest geometrical shape from center points of each region, while checking that these points create a geometry that contain the green center.

There were visualisations of them here: Algorithm Showcase - Roblox by @sircfenner. I think this is the correct link, I totally forgot if it was it, because last time I checked I lost track of the link and also I don’t have access on my primary device yet.

1 Like

All of these were custom made and hand placed.

You can use a Region3 check (the Region3 would probably be centered at the region you want to check with a size that’s twice the size of the bounding box of the region) to get a list of all all possible neighbors.

For each possible neighbor, check with a raycast if there’s a direct line of sight between it and the original region. If there is, then it’s an actual neighbor. Otherwise, it isn’t.

What a nasty problem! Do you have access to the points of the parts? Are these actual parts, unions, or mesh parts? To make the problem more complicated, these arn’t even convex shapes…

If you have the points, I’ll probably write the script for you

1 Like

Each of these parts happen to be a union. I wonder if it’d be possible to union each part, get all of the unioned parts and there corners, then write a general shape from that.

Okay, I have the bulk of the code complete but untested. To make it work, findAdjacent needs to be called with a table whose keys are arrays of the Vector2 points of each region. Let me know if you need help doing that. This code is pretty close to the solution and should only have minor bugs, if any. Also lmk if you have any trouble debugging it.

It returns a table whose keys (key 1) is each region, and the value is a table with the keys (key 2) equal to all the regions adjacent to key 1. Value 2 is just true. It should make sense when you look at the code.

local function union(from, to)
	for key, value in next, from do
		to[key] = value
	end
end

-- We can cache centers since regions are immutable
local centerCache = setmetatable({}, {__key = 'k'})
local function getCenter(region)
	if centerCache[region] then
		return centerCache[region]
	end

	local center = Vector2.new()
	for i, point in next, region do
		center = center + point
	end
	center = center / #region

	centerCache[region] = center
	return center
end

-- Line point a, line point b, center c, point p
local function isInside(a, b, c, p)
	local ab = b - a
	return ab:Cross(c - a) * ab:Cross(p - a) > 0
end

local function findConvexParts(originalRegion)
	local region = union(originalRegion, {})
	local convexParts = {[region] = true}

	local center = getCenter(region)

	local i = 1
	local a = region[1]
	local b = region[2]
	local c = region[3]
	while c do
		if isInside(a, c, center, b) then
			local prev
			if i == 1 then
				prev = region[#region]
			else
				prev = region[i - 1]
			end

			-- Triangle are garenteed to be convex, being the 2D simplex
			convexParts[{b, a, prev}] = true

			-- Bye-bye point a
			table.remove(region, i)
			center = center - a / #region
		else
			i = i + 1
		end

		a = b
		b = c
		c = region[(i+1 % #region) + 1]
	end

	return convexParts
end

local function getOutwardNormal(a, b, center)
	local line = b - a
	return line:Cross(line:Cross(center - a)).Unit
end

local maxGap = 0.5
local function expand(originalRegion)
	local region = {}
	local center = getCenter(originalRegion)

	local prevI = #region
	for i, point in next, originalRegion do
		local prev = originalRegion[prevI]
		local next = originalRegion[(i % #region) + 1]
		local normA = getOutwardNormal(point, prev, center)
		local normB = getOutwardNormal(point, next, center)
		region[i] = point + maxGap * (normA + normB)
	end

	return region
end

local function contains(region, other, center)
	for i, a in next, region do
		local b = region[(i % #region) + 1]
		for j, p in next, other do
			if not isInside(a, b, center, p) then
				other[j] = nil
			end
		end
	end

	if next(other) then
		return true
	end
end

-- 2D intersection algorithm
-- For 2D shapes to intersect, one must have a point inside another
-- Since they are convex the seperating line must be the line between them
local function isIntersecting(region1Original, region2Original)
	local region1 = union(region1Original, {})
	local region2 = union(region2Original, {})
	return contains(region1Original, region2, getCenter(region1))
		or contains(region2Original, region1, getCenter(region2))
end

local function getBoundingCircle(region)
	local center = getCenter(region)

	local maxRadius = 0
	for i, point in next, region do
		local radius = (point - center).Magnitude
		if radius > maxRadius then
			maxRadius = radius
		end
	end

	return {
		center = center;
		radius = maxRadius;
	}
end

local function circlesIntersect(circle, otherCircle)
	return (circle.center - otherCircle.center).Magnitude < circle.radius + otherCircle.radius
end

local function findAdjacent(originalRegions)
	local toOriginal = {}
	local adjacent = {}
	local circles = {}

	for region in next, originalRegions do
		adjacent[region] = {}
		for part in next, findConvexParts(expand(region)) do
			toOriginal[part] = region
			circles[part] = getBoundingCircle(region)
		end
	end

	-- Use very quick bounding circle tests to cut down number of checks
	local checkList = {}
	for part, circle in next, circles do
		for other, otherCircle in next, circles, part do
			if toOriginal[part] ~= toOriginal[other] and circlesIntersect(circle, otherCircle) then
				checkList[{part, other}] = true
			end
		end
	end

	-- Perform exact 2D intersection tests on near parts
	for set in next, checkList do
		local part1 = set[1]
		local part2 = set[2]
		local original1 = toOriginal[part1]
		local original2 = toOriginal[part2]

		-- We'll need to see if the regions the parts are from
		-- have already been found to be adjacent
		if not adjacent[original1][original2]
			and isIntersecting(part1, part2)
		then
			adjacent[original1][original2] = true
			adjacent[original2][original1] = true
		end
	end

	return adjacent
end
1 Like

I am a bit confused on initializing this so it works, could you explain what you mean by…

Would that just be the position of the union, or the position of all of the parts inside the union?

1 Like

That would be the points that define the boundry of each region, the edge points / corners. The first point is any point on the edge of the region and the next point is a point adjacent to it. The first and last points are also adjacent, forming a closed shaped. I’m not sure how you’ll get the points… it depends on how the unions were created. If you post some more information, maybe I could help with that too.

1 Like

Each of the parts are just unions, let me get a screenshot of a few of them unioned for you.

Cool, if all of the regions are put into a single group and are flat on the XZ plane I’ll write a script that goes through each one, breaks it into its parts, finds the points, and then calls the previous function. I’ll start on it in an hour or two.

Awesome man, I’ll union them up while you work on that. Thank you so much for your help!

One option is to practically brute force it, taking advantage of the semi-accurate union physics.

  1. Raycast a bunch of grid points in a square, only whitelisting a single region part. This will let you find which points hit and which points don’t hit, letting you find an approximate border. RegionApx
  2. Raycast a bunch of grid points a few grid positions away from the approximate border, whitelisting only every other region. This will let you find bordering regions. Bordering

Since this only needs to be ran once in Studio, the overall efficiency is not very important, but the dev time you put in is. Writing a simple brute-force solution like this will save you time.

I could do something like this, it was also my original idea, however @IdiomicLanguage seemed to come up with a potential solution that would work miles better then a bruteforce would it not?

This entirely depends on where and how you will use your solution. Both should give you similar results.

We can draft out two scenerios:

  1. You need to calculate this data in real-time in a live game. You want computational efficiency at the expense of longer dev time.
  2. You need to pre-calculate this in Studio. You want short dev time at the expense of computational efficiency.

To my understanding, you have all these regions set up in Studio and you won’t be generating any new ones in the game. Because of this, you can pre-calculate the surrounding regions, save it, and use that saved data in a live game. Your scenario is #2, then, not #1, right?

IdiomicLanguage’s solution certainly sounds more computational efficient and correct, but if – for example – you can spend 1 hour writing a brute force solution and 1 hour pre-generating data, that’s better than 4 hours writing an efficient solution that you’ll only use once.

How would I actually save a module full of the required data though? Can’t you only write to a modulescript/script via plugin? I’m not really looking to make a plugin for this haha.

You can give each region an id and convert the adjacency matrix into JSON using the IDs instead of tables for keys. You’d just have to name the regions the tostring of their ID. Just starting.

If you trigger the code from the command bar, then you can write to a module’s source with it. Lua scripts you save as .lua and run with the Run Script button also have write access to the Source property.

If you don’t want to do that, you can just save it to a StringValue then copy the Value to a modulescript. If you want an easy way to output approximate Lua code for a table, this module will format non-recursive tables with normal types like Lua syntax.

Another option is to save it as JSON and decode the JSON at runtime.

Here’s an example format you can save that data in:

return {
	['001'] = { -- region name
		'002', '003', '004', '007', '023', -- bordering region names
	},
	['002'] = {
		'001', '003', -- etc.
	}
}
1 Like

I am a tad bit confused on the usage of the module you provided. There doesn’t seem to be any actual examples for it, unless I’m looking in the wrong places?

Was looking in the wrong place, I’ll look in to this now.