threejs-core-raycaster

Quick Reference

Raycaster Constructor

import * as THREE from 'three';

const raycaster = new THREE.Raycaster(
  origin,     // Vector3 — default: (0, 0, 0)
  direction,  // Vector3 — MUST be normalized, default: (0, 0, -1)
  near,       // number — minimum distance, default: 0
  far         // number — maximum distance, default: Infinity
);

Properties

Params Defaults

raycaster.params = {
  Mesh: {},
  Line: { threshold: 1 },    // world units distance for line hits
  LOD: {},
  Points: { threshold: 1 },  // world units distance for point hits
  Sprite: {}
};

Critical Warnings

NEVER raycast on every mousemove event without throttling -- this causes severe frame drops on complex scenes. ALWAYS throttle to the render loop via requestAnimationFrame or limit to 30-60 checks per second.

NEVER pass recursive: true on large scene graphs when you have a flat array of target objects -- ALWAYS maintain a flat array of selectable objects and pass recursive: false to avoid unnecessary tree traversal.

NEVER forget to normalize the direction vector when using raycaster.set() -- an unnormalized direction produces incorrect distance values in all intersection results.

NEVER allocate a new results array every frame -- ALWAYS reuse an array via the optionalTarget parameter and clear it with array.length = 0 after processing.

NEVER raycast against the entire scene.children when only a subset of objects is interactive -- ALWAYS maintain a separate array of selectable objects or use layers for filtering.

NEVER omit raycaster.camera when raycasting against Sprite objects -- the raycaster requires the camera reference to compute sprite screen-space bounds.

Core Methods

set(origin, direction)

Manually sets the ray origin and direction. Direction MUST be normalized.

const origin = new THREE.Vector3(0, 1, 0);
const direction = new THREE.Vector3(0, -1, 0); // already normalized
raycaster.set(origin, direction);

setFromCamera(coords, camera)

Sets the ray from normalized device coordinates (NDC) and a camera. This is the standard mouse-picking method.

• coords: Vector2 with x and y in range [-1, +1]
• camera: PerspectiveCamera produces a ray from the camera through the point; OrthographicCamera produces a parallel ray

const mouse = new THREE.Vector2();
mouse.x = (event.clientX / window.innerWidth) * 2 - 1;
mouse.y = -(event.clientY / window.innerHeight) * 2 + 1;
raycaster.setFromCamera(mouse, camera);

intersectObject(object, recursive?, optionalTarget?)

Tests a single object (and optionally its descendants) for intersections.

• recursive (default: true): when true, tests all descendants
• optionalTarget: reusable array to avoid allocation
• Returns: Intersection[] ALWAYS sorted by distance (nearest first)

intersectObjects(objects, recursive?, optionalTarget?)

Tests an array of objects for intersections.

• recursive (default: true): when true, tests descendants of each object
• optionalTarget: reusable array to avoid allocation
• Returns: Intersection[] ALWAYS sorted by distance (nearest first)

Intersection Object Format

Every intersection in the returned array has this structure:

interface Intersection {
  distance: number;       // distance from ray origin to hit point
  point: Vector3;         // hit point in world space
  face: Face | null;      // hit face ({a, b, c} vertex indices + normal)
  faceIndex: number;      // index of hit face in the geometry
  object: Object3D;       // the intersected object reference
  uv?: Vector2;           // UV coordinates at intersection point
  uv1?: Vector2;          // second UV set (if available)
  normal?: Vector3;       // interpolated surface normal at hit point
  instanceId?: number;    // instance index (only for InstancedMesh)
}

ALWAYS check intersects.length > 0 before accessing intersects[0]. The array is empty when no objects are hit.

NDC Conversion (Normalized Device Coordinates)

Converting DOM mouse coordinates to NDC is required for setFromCamera. The NDC space ranges from -1 (left/bottom) to +1 (right/top).

// For fullscreen canvas (canvas fills entire window)
mouse.x = (event.clientX / window.innerWidth) * 2 - 1;
mouse.y = -(event.clientY / window.innerHeight) * 2 + 1;

// For non-fullscreen canvas (canvas has offset within page)
const rect = renderer.domElement.getBoundingClientRect();
mouse.x = ((event.clientX - rect.left) / rect.width) * 2 - 1;
mouse.y = -((event.clientY - rect.top) / rect.height) * 2 + 1;

ALWAYS use getBoundingClientRect() when the canvas does NOT fill the entire viewport. Using window.innerWidth/Height on a partial canvas produces incorrect NDC values and misaligned picking.

InstancedMesh Picking

When raycasting against InstancedMesh, the intersection result includes instanceId identifying which instance was hit:

import * as THREE from 'three';

const intersects = raycaster.intersectObject(instancedMesh);
if (intersects.length > 0) {
  const instanceId = intersects[0].instanceId;
  // Retrieve the instance's transform matrix
  const matrix = new THREE.Matrix4();
  instancedMesh.getMatrixAt(instanceId, matrix);

  // Example: change the color of the hit instance
  const color = new THREE.Color();
  instancedMesh.getColorAt(instanceId, color);
  instancedMesh.setColorAt(instanceId, new THREE.Color(0xff0000));
  instancedMesh.instanceColor.needsUpdate = true;
}

ALWAYS check that instanceId !== undefined before using it -- non-InstancedMesh objects do not have this property.

Layer Filtering

Raycaster respects the Layers system. Only objects whose layers overlap with raycaster.layers are tested.

// Assign objects to layers
selectableMesh.layers.set(1);        // exclusively on layer 1
decorationMesh.layers.set(2);        // exclusively on layer 2

// Configure raycaster to only test layer 1
raycaster.layers.set(1);
// Now intersectObjects skips all objects NOT on layer 1

// Enable multiple layers
raycaster.layers.enable(1);
raycaster.layers.enable(2);
// Now tests objects on layer 1 OR layer 2

// Reset to default (layer 0 only)
raycaster.layers.set(0);

Use layers to create picking groups: interactive objects on layer 1, non-interactive decoration on layer 2, helpers/gizmos on layer 3.

Performance Guidelines

1. Throttle mousemove raycasting -- NEVER raycast on every mousemove event. Use a flag checked in the render loop:

   javascriptCopy

   let needsRaycast = false;
   canvas.addEventListener('pointermove', (event) => {
     mouse.x = (event.clientX / window.innerWidth) * 2 - 1;
     mouse.y = -(event.clientY / window.innerHeight) * 2 + 1;
     needsRaycast = true;
   });
   
   function animate() {
     if (needsRaycast) {
       raycaster.setFromCamera(mouse, camera);
       // perform intersection tests
       needsRaycast = false;
     }
     renderer.render(scene, camera);
     requestAnimationFrame(animate);
   }
   

2. Use layers -- assign interactive objects to a specific layer and set raycaster.layers accordingly to skip non-interactive geometry entirely.

3. Maintain a flat selectable array -- instead of raycasting against scene with recursive: true, keep a separate selectableObjects[] array and pass recursive: false.

4. Reuse the intersections array -- pass optionalTarget to avoid garbage collection:

   javascriptCopy

   const intersections = [];
   raycaster.intersectObjects(selectableObjects, false, intersections);
   // process results...
   intersections.length = 0;
   

5. Bounding sphere pre-check -- Three.js automatically tests bounding spheres before triangle intersection. Ensure geometry.computeBoundingSphere() has been called (it is called automatically on first render, but manual call is needed if raycasting before first render).

6. Limit near/far -- narrow the ray range when you know the expected intersection distance to skip distant objects early.

Reference Links

• references/methods.md -- Complete Raycaster API signatures
• references/examples.md -- Working code examples for picking, hover, and InstancedMesh
• references/anti-patterns.md -- What NOT to do with raycasting

Official Sources

• https://threejs.org/docs/#api/en/core/Raycaster
• https://threejs.org/docs/#api/en/core/Layers