Skip to content

Tilemaps

The @yagejs/tilemap package loads and renders maps created with the Tiled map editor. It supports tile layer rendering, object layer extraction, and physics collision shape generation.

Worth knowing up front so you don’t reach for something that isn’t there.

FeatureStatus
Orthogonal Tiled JSON mapsSupported
Multiple tile layers per mapSupported
Object layers (spawns, triggers, collision shapes)Supported
Custom properties on objectsSupported
Object-reference resolution (Tiled object refs)Supported
Collision-shape extraction (rect / ellipse / capsule / polygon → polyline)Supported
toPhysicsColliders() adapter for RapierSupported
Tileset-image and collection-of-images tilesetsSupported
Animated tilesNot yet
Infinite / chunked mapsNot yet
Isometric / hexagonal / staggered orientationsNot planned
Dynamic tile editing at runtimeNot yet
Parallax / background scroll layersNot built-in (use a regular sprite layer)

For a worked example of loading a Tiled map and extracting collision shapes, see the tilemap example — it parses a Tiled JSON, renders the layers, calls getCollisionShapes("walls"), and visualizes the resulting rectangles through the debug overlay. Two distinct steps make up the Tiled-collision-to-Rapier workflow:

  1. Extractiontilemap.getCollisionShapes("walls") returns raw shape data in Tiled’s coordinate system (top-left-origin rect / circle / capsule / polygon / polyline configs).
  2. Physics conversiontoPhysicsColliders(shapes) adapts those raw shapes into Rapier ColliderConfigs with the offsets baked in. Bounding-box shapes (rect, circle, capsule) become center-origin configs (offset = x + width/2, y + height/2); polygons and polylines keep their original vertices with a top-left-based offset (offset = { x, y }), since vertex-defined shapes don’t need centering.

The wiring (extracted shapes → toPhysicsColliders() → static RigidBodyComponent + per-shape ColliderComponent) is shown inline in the Collision Extraction section below.

Tiled’s “Export As JSON” dumps the tileset wherever you point it, but the JSON itself still references its source PNG by path-relative-to- the-JSON. The loader resolves it as dirname(tilesetSrc) + tileset.image. Tiled writes that field exactly as it was authored — typically a path that walks out of the project ("../../Downloads/tiled-projects/dungeon/spr_tileset.png"). The moment you copy the JSON into public/, that path no longer resolves in the browser and you get a silent 404: the tileset object loads, the texture doesn’t, and tiles render as blank rectangles.

The fix is mechanical — rewrite image to a sibling-relative path whenever you stage a tileset:

public/assets/maps/dungeon.tsj
{
"tilewidth": 16,
"tileheight": 16,
"tilecount": 256,
"image": "../../Downloads/tiled-projects/dungeon/spr_tileset.png",
"image": "spr_tileset.png",
...
}

…and drop spr_tileset.png next to dungeon.tsj in public/assets/maps/. The same rule applies to embedded tilesets inside a map JSON: the image field is resolved relative to the map file’s directory, so the PNG needs to sit next to the map (or at any sibling path you reference). Watching for this saves a frustrating “why is my map blank?” debugging cycle.

import { TilemapPlugin } from "@yagejs/tilemap";
engine.use(new TilemapPlugin());

The plugin depends on @yagejs/renderer.

Use the tiledMap() asset factory and pair it with a spritesheet texture:

import { tiledMap } from "@yagejs/tilemap";
import { renderAsset } from "@yagejs/renderer";
const MapData = tiledMap("assets/level.json");
const Tileset = renderAsset("assets/tileset.png");

Add both to your scene’s preload array. The tileset texture must load before the map so tile rendering can resolve texture frames:

class LevelScene extends Scene {
readonly preload = [Tileset, MapData];
}

Add a TilemapComponent to an entity to render the map:

import { Scene, Transform } from "@yagejs/core";
import { TilemapComponent } from "@yagejs/tilemap";
import type { LayerDef } from "@yagejs/renderer";
class LevelScene extends Scene {
readonly name = "level";
readonly preload = [Tileset, MapData];
// Declare the scene's layers — the renderer plugin materializes them
// automatically when the scene is pushed.
readonly layers: readonly LayerDef[] = [
{ name: "map", order: -10 },
];
onEnter(): void {
const map = this.spawn("map");
map.add(new Transform());
map.add(new TilemapComponent({
source: MapData, // asset handle — preferred form
layers: ["ground", "walls"], // which tile layers to render (omit for all)
layer: "map", // render layer name
}));
}
}

source takes the same AssetHandle you passed to preload. Internally the component captures both the parsed map data and the asset path, so the same constructor argument enables save/load and the Tiled-derived auto-keys covered below. If you don’t have a handle in scope, you can still pass mapKey as a plain string.

Query map dimensions from the component:

const tilemap = map.get(TilemapComponent);
tilemap.widthPx; // total width in pixels
tilemap.heightPx; // total height in pixels
tilemap.tileWidth; // single tile width
tilemap.tileHeight; // single tile height

TilemapComponent is @serializable, but the live parsed map it wraps is not — the parsed TiledMapData carries PixiJS textures, which don’t survive JSON.stringify. The constructor therefore accepts three alternative inputs: a source: AssetHandle<TiledMapData> (preferred — captures both the parsed data and the asset path in one argument, which doubles as the prefix for object auto-keys), an in-memory map object for the fast-path case where you already have the parsed map, and a mapKey asset path that can be re-resolved from the AssetManager after a reload.

interface TilemapComponentOptions {
/** Asset handle — preferred. Captures both data and path in one place. */
source?: AssetHandle<TiledMapData>;
/** Parsed Tiled map data — not serializable, no auto-keys. */
map?: TiledMapData;
/** Asset path string — equivalent to `source` when you don't have the handle. */
mapKey?: string;
/** Which tile layers to render. Omit to render all. */
layers?: string[];
/** Render layer name. Default: "default". */
layer?: string;
/** Override prefix for object auto-keys. Defaults to mapKey. */
keyPrefix?: string;
}

For any game that uses the save/load system, pass source (or mapKey) rather than map. When a TilemapComponent constructed with an inline map gets serialized, serialize() emits a warning and the snapshot cannot round-trip through afterRestore(). The asset-path form lets the save system store just the reference and re-look-up the texture-backed map on restore:

// Recommended: save/load friendly + auto-keys for Tiled objects
map.add(new TilemapComponent({
source: MapData,
layers: ["ground", "walls"],
layer: "map",
}));
// One-shot prototypes only — will NOT survive a save/load cycle and
// auto-keys are unavailable.
map.add(new TilemapComponent({
map: parsedTiledJson,
layer: "map",
}));

The snapshot shape itself (TilemapComponentData) is just { mapKey, layers?, layer, keyPrefix? } — the serialized entity stores only the asset key, layer selection, and any explicit keyPrefix override, and reconstructs the live map on load.

Sometimes you need to reach past the renderer and read raw tile data — for building a pathfinding grid, counting tile types, or driving custom gameplay off tile positions. TilemapComponent exposes a format-agnostic snapshot of the parsed map on its data property:

const tilemap = map.get(TilemapComponent);
const data = tilemap.data; // TilemapData
console.log(`${data.width} x ${data.height} tiles`);
console.log(`${data.tileLayers.length} tile layers`);

The TilemapData shape is deliberately decoupled from the Tiled JSON format — you can build your own parser for a different editor and reuse the same downstream consumers.

interface TilemapData {
width: number; // tiles wide
height: number; // tiles tall
tileWidth: number; // pixel width of one tile
tileHeight: number;
tileLayers: TileLayerData[];
objectLayers: ObjectLayerData[];
}
interface TileLayerData {
name: string;
data: number[]; // flat, row-major tile GIDs (0 = empty)
width: number;
height: number;
visible: boolean;
}
interface ObjectLayerData {
name: string;
objects: MapObject[];
visible: boolean;
}

TileLayerData.data is a single flat array in row-major order. To read the tile at tile-coordinate (tx, ty), index with data[ty * width + tx]. A GID of 0 means “no tile at this position”. That flat layout makes it cheap to iterate for pathfinding or damage maps:

// Count all solid tiles in the "walls" layer
const walls = data.tileLayers.find((l) => l.name === "walls")!;
let solidCount = 0;
for (const gid of walls.data) {
if (gid !== 0) solidCount++;
}

MapObject (the element type in objectLayers[i].objects) carries id, name, optional class, position and size, rotation, an optional point flag, an optional polygon, and an optional properties: MapObjectProperty[] array of Tiled custom properties — covered in detail below.

Look up tile data at a world position:

const gid = tilemap.getTileAt(worldX, worldY, "ground");
// Returns the tile GID or null if empty

Tiled object layers contain spawn points, triggers, and other game data. Extract objects grouped by their class or name:

const objects = tilemap.getObjects("spawns");
// Returns: Record<string, MapObject[]>
// e.g. { "player": [{ x, y, ... }], "enemy": [{ x, y, width, height, ... }] }
for (const obj of objects["enemy"] ?? []) {
this.spawn(EnemyBP, { x: obj.x, y: obj.y });
}

Each MapObject has:

interface MapObject {
id: number;
name: string;
class?: string;
x: number;
y: number;
width: number;
height: number;
rotation: number;
visible: boolean;
point?: boolean;
polygon?: { x: number; y: number }[];
properties?: MapObjectProperty[];
}

The component exposes a few direct accessors so you don’t have to flatten getObjects() for one-off lookups:

tilemap.getAllObjects(); // flat list across every layer
tilemap.findObject(42); // by Tiled id
tilemap.findObjectByName("Player"); // first match across all layers

Spawning Entities from Tiled Objects (auto-keys)

Section titled “Spawning Entities from Tiled Objects (auto-keys)”

Tiled assigns every object a stable per-map id. Combine that id with the map’s asset path to derive a stable entity.key so persistent stores (createSet<string>, createMap<string, T>) can be keyed off authored content without your game inventing its own naming scheme. The component exposes the prefix wiring as objectKey() and forEachObject():

import { tiledObjectKey, TilemapComponent } from "@yagejs/tilemap";
// Standalone helper, format: "<prefix>#object:<id>"
tiledObjectKey("/assets/dungeon.json", 42);
// → "/assets/dungeon.json#object:42"
// On the component — prefix already wired up from `source` / `mapKey`:
tilemap.objectKey(obj);
tilemap.forEachObject("interactables", (obj, key) => {
if (obj.class === "EnemySpawn") {
this.spawn(EnemyEntity, { object: obj }, { key });
}
});

Pass keyPrefix: "level1" to the component constructor when multiple instances of the same map need distinct identity namespaces (instanced dungeons, per-floor layouts).

objectKey and forEachObject throw if the component was constructed from raw map: data without a mapKey, source, or explicit keyPrefix — auto-keys need a stable prefix.

See Persistent World State for the canonical pattern that pairs these auto-keys with createSet / createMap to remember which chests are open, which triggers fired, etc., across save/load.

Read custom Tiled properties from objects. The component-method variants are typed and discoverable; the standalone helpers exist for cases where you’ve already collected an object pool yourself.

// On the component (preferred)
tilemap.getProperty<number>(obj, "speed");
tilemap.getPropertyArray<number>(obj, "point");
tilemap.resolveRef(obj, "target"); // walks every layer
tilemap.resolveRefArray(obj, "spawns");
// Standalone (caller supplies the pool)
import { getProperty, getPropertyArray, resolveObjectRef, resolveObjectRefArray } from "@yagejs/tilemap";
getProperty<number>(obj, "speed");
getPropertyArray<number>(obj, "point");
resolveObjectRef(obj, "target", allObjs);
resolveObjectRefArray(obj, "path", allObjs);

Tiled objects map to physics-agnostic TilemapColliderConfig variants:

Tiled objectEmitted shape
Rectangle{ type: "rect", x, y, width, height }
Ellipse (width === height){ type: "circle", x, y, width, height, radius }
Ellipse (width !== height){ type: "circle", ..., radius: max(w, h) / 2 } + dev warning (Rapier has no real ellipse — author it as a capsule for a true non-circular round shape)
Capsule{ type: "capsule", x, y, width, height, halfHeight, radius, axis } (oriented along the longer axis)
Polygon{ type: "polyline", x, y, vertices } (Tiled polygons are outlines; may be concave)
PointSkipped

Because Tiled polygons are emitted as polylines (chain of line segments), they support concave outlines such as a winding water edge or a U-shaped wall. The trade-off: polylines are static-only — they must be attached to a type: "static" rigid body (no mass/inertia is computed). For dynamic concave bodies, decompose the outline into convex pieces manually.

Wire extracted shapes through toPhysicsColliders() to spawn one static body with one ColliderComponent per shape:

import { toPhysicsColliders } from "@yagejs/tilemap";
import { RigidBodyComponent, ColliderComponent } from "@yagejs/physics";
const walls = this.spawn("walls");
walls.add(new Transform());
walls.add(new RigidBodyComponent({ type: "static" }));
for (const cfg of toPhysicsColliders(tilemap.getCollisionShapes("walls"))) {
walls.add(new ColliderComponent(cfg));
}

The component-method shortcut wraps two standalone functions for the same workflow:

import { extractCollisionShapes, toPhysicsColliders } from "@yagejs/tilemap";

Use the tilemap dimensions to constrain the camera. The camera is an entity you spawn in the scene:

import { CameraEntity } from "@yagejs/renderer";
const cam = this.spawn(CameraEntity, { follow: player.get(Transform) });
cam.bounds = {
minX: 0,
minY: 0,
maxX: tilemap.widthPx,
maxY: tilemap.heightPx,
};

This prevents the camera from scrolling past the map edges.