Collision Detection

Collision detection is needed in most games to detect and act upon two components intersecting each other. For example an arrow hitting an enemy or the player picking up a coin.

In most collision detection systems you use something called hitboxes to create more precise bounding boxes of your components. In Flame the hitboxes are areas of the component that can react to collisions (and make gesture input) more accurate.

The collision detection system supports three different types of shapes that you can build hitboxes from, these shapes are Polygon, Rectangle and Circle. Multiple hitbox can be added to a component to form the area which can be used to either detect collisions or whether it contains a point or not, the latter is very useful for accurate gesture detection. The collision detection does not handle what should happen when two hitboxes collide, so it is up to the user to implement what will happen when for example two PositionComponents have intersecting hitboxes.

Do note that the built-in collision detection system does not take collisions between two hitboxes that overshoot each other into account, this could happen when they either move very fast or update being called with a large delta time (for example if your app is not in the foreground). This behavior is called tunneling, if you want to read more about it.

Also note that the collision detection system has a limitation that makes it not work properly if you have certain types of combinations of flips and scales of the ancestors of the hitboxes.

Mixins

HasCollisionDetection

If you want to use collision detection in your game you have to add the HasCollisionDetection mixin to your game so that it can keep track of the components that can collide.

Example:

class MyGame extends FlameGame with HasCollisionDetection {
  // ...
}

Now when you add ShapeHitboxs to components that are then added to the game, they will automatically be checked for collisions.

You can also add HasCollisionDetection directly to another Component instead of the FlameGame, for example to the World that is used for the CameraComponent. If that is done, hitboxes that are added in that component’s tree will only be compared to other hitboxes in that subtree, which makes it possible to have several worlds with collision detection within one FlameGame.

Example:

class CollisionDetectionWorld extends World with HasCollisionDetection {}

Note

Hitboxes will only be connected to one collision detection system and that is the closest parent that has the HasCollisionDetection mixin.

CollisionCallbacks

To react to a collision you should add the CollisionCallbacks mixin to your component. Example:

collision_detection.dart
 1import 'package:doc_flame_examples/ember.dart';
 2import 'package:flame/collisions.dart';
 3import 'package:flame/components.dart';
 4import 'package:flame/effects.dart';
 5import 'package:flame/game.dart';
 6import 'package:flutter/material.dart' hide Image;
 7
 8class CollisionDetectionGame extends FlameGame with HasCollisionDetection {
 9  @override
10  Future<void> onLoad() async {
11    final emberPlayer = EmberPlayer(
12      position: Vector2(10, (size.y / 2) - 20),
13      size: Vector2.all(40),
14      onTap: (emberPlayer) {
15        emberPlayer.add(
16          MoveEffect.to(
17            Vector2(size.x - 40, (size.y / 2) - 20),
18            EffectController(
19              duration: 5,
20              reverseDuration: 5,
21              repeatCount: 1,
22              curve: Curves.easeOut,
23            ),
24          ),
25        );
26      },
27    );
28    add(emberPlayer);
29    add(RectangleCollidable(canvasSize / 2));
30  }
31}
32
33class RectangleCollidable extends PositionComponent with CollisionCallbacks {
34  final _collisionStartColor = Colors.amber;
35  final _defaultColor = Colors.cyan;
36  late ShapeHitbox hitbox;
37
38  RectangleCollidable(Vector2 position)
39      : super(
40          position: position,
41          size: Vector2.all(50),
42          anchor: Anchor.center,
43        );
44
45  @override
46  Future<void> onLoad() async {
47    final defaultPaint = Paint()
48      ..color = _defaultColor
49      ..style = PaintingStyle.stroke;
50    hitbox = RectangleHitbox()
51      ..paint = defaultPaint
52      ..renderShape = true;
53    add(hitbox);
54  }
55
56  @override
57  void onCollisionStart(
58    Set<Vector2> intersectionPoints,
59    PositionComponent other,
60  ) {
61    super.onCollisionStart(intersectionPoints, other);
62    hitbox.paint.color = _collisionStartColor;
63  }
64
65  @override
66  void onCollisionEnd(PositionComponent other) {
67    super.onCollisionEnd(other);
68    if (!isColliding) {
69      hitbox.paint.color = _defaultColor;
70    }
71  }
72}

class MyCollidable extends PositionComponent with CollisionCallbacks {
  @override
  void onCollision(Set<Vector2> points, PositionComponent other) {
    if (other is ScreenHitbox) {
      //...
    } else if (other is YourOtherComponent) {
      //...
    }
  }

  @override
  void onCollisionEnd(PositionComponent other) {
    if (other is ScreenHitbox) {
      //...
    } else if (other is YourOtherComponent) {
      //...
    }
  }
}

In this example we use Dart’s is keyword to check what kind of component we collided with. The set of points is where the edges of the hitboxes intersect.

Note that the onCollision method will be called on both PositionComponents if they have both implemented the onCollision method, and also on both hitboxes. The same goes for the onCollisionStart and onCollisionEnd methods, which are called when two components and hitboxes starts or stops colliding with each other.

When a PositionComponent (and hitbox) starts to collide with another PositionComponent both onCollisionStart and onCollision are called, so if you don’t need to do something specific when a collision starts you only need to override onCollision, and vice versa.

If you want to check collisions with the screen edges, as we do in the example above, you can use the predefined ScreenHitbox class.

By default all hitboxes are hollow, this means that one hitbox can be fully enclosed by another hitbox without triggering a collision. If you want to set your hitboxes to be solid you can set isSolid = true. A hollow hitbox inside of a solid hitbox will trigger a collision, but not the other way around. If there are no intersections with the edges on a solid hitbox the center position is instead returned.

Collision order

If a Hitbox collides with more than one other Hitbox within a given time step, then the onCollision callbacks will be called in an essentially random order. In some cases this can be a problem, such as in a bouncing ball game where the trajectory of the ball can differ depending on which other object was hit first. To help resolve this the collisionsCompletedNotifier listener can be used - this triggers at the end of the collision detection process.

An example of how this might be used is to add a local variable in your PositionComponent to save the other components with which it’s colliding: List<PositionComponent> collisionComponents = [];. The onCollision callback is then used to save all the other PositionComponents to this list:

@override
void onCollision(Set<Vector2> intersectionPoints, PositionComponent other) {
  collisionComponents.add(other);
  super.onCollision(intersectionPoints, other);
}

Finally, one adds a listener to the onLoad method of the PositionComponent to call a function which will resolve how the collisions should be dealt with:

(game as HasCollisionDetection)
    .collisionDetection
    .collisionsCompletedNotifier
    .addListener(() {
  resolveCollisions();
});

The list collisionComponents would need to be cleared in each call to update.

ShapeHitbox

The ShapeHitboxs are normal components, so you add them to the component that you want to add hitboxes to just like any other component:

class MyComponent extends PositionComponent {
  @override
  void onLoad() {
    add(RectangleHitbox());
  }
}

If you don’t add any arguments to the hitbox, like above, the hitbox will try to fill its parent as much as possible. Except for having the hitboxes trying to fill their parents, there are two ways to initiate hitboxes and it is with the normal constructor where you define the hitbox by itself, with a size and a position etc. The other way is to use the relative constructor which defines the hitbox in relation to the size of its intended parent.

In some specific cases you might want to handle collisions only between hitboxes, without propagating onCollision* events to the hitbox’s parent component. For example, a vehicle could have a body hitbox to control collisions and side hitboxes to check the possibility to turn left or right. So, colliding with a body hitbox means colliding with the component itself, whereas colliding with a side hitbox does not mean a real collision and should not be propagated to hitbox’s parent. For this case you can set triggersParentCollision variable to false:

class MyComponent extends PositionComponent {

  late final MySpecialHitbox utilityHitbox;

  @override
  void onLoad() {
    utilityHitbox = MySpecialHitbox();
    add(utilityHitbox);
  }

  void update(double dt) {
    if (utilityHitbox.isColliding) {
      // do some specific things if hitbox is colliding
    }
  }
// component's onCollision* functions, ignoring MySpecialHitbox collisions.
}

class MySpecialHitbox extends RectangleHitbox {
  MySpecialHitbox() {
    triggersParentCollision = false;
  }

// hitbox specific onCollision* functions

}

You can read more about how the different shapes are defined in the ShapeComponents section.

Remember that you can add as many ShapeHitboxs as you want to your PositionComponent to make up more complex areas. For example a snowman with a hat could be represented by three CircleHitboxs and two RectangleHitboxs as its hat.

A hitbox can be used either for collision detection or for making gesture detection more accurate on top of components, see more regarding the latter in the section about the GestureHitboxes mixin.

CollisionType

The hitboxes have a field called collisionType which defines when a hitbox should collide with another. Usually you want to set as many hitboxes as possible to CollisionType.passive to make the collision detection more performant. By default the CollisionType is active.

The CollisionType enum contains the following values:

  • active collides with other Hitboxes of type active or passive

  • passive collides with other Hitboxes of type active

  • inactive will not collide with any other Hitboxes

So if you have hitboxes that you don’t need to check collisions against each other you can mark them as passive by setting collisionType: CollisionType.passive in the constructor, this could for example be ground components or maybe your enemies don’t need to check collisions between each other, then they could be marked as passive too.

Imagine a game where there are a lot of bullets, that can’t collide with each other, flying towards the player, then the player would be set to CollisionType.active and the bullets would be set to CollisionType.passive.

Then we have the inactive type which simply doesn’t get checked at all in the collision detection. This could be used for example if you have components outside of the screen that you don’t care about at the moment but that might later come back in to view so they are not completely removed from the game.

These are just examples of how you could use these types, there will be a lot more use cases for them so don’t doubt to use them even if your use case isn’t listed here.

PolygonHitbox

It should be noted that if you want to use collision detection or containsPoint on the Polygon, the polygon needs to be convex. So always use convex polygons or you will most likely run into problems if you don’t really know what you are doing.

The other hitbox shapes don’t have any mandatory constructor, that is because they can have a default calculated from the size of the collidable that they are attached to, but since a polygon can be made in an infinite number of ways inside of a bounding box you have to add the definition in the constructor for this shape.

The PolygonHitbox has the same constructors as the PolygonComponent, see that section for documentation regarding those.

RectangleHitbox

The RectangleHitbox has the same constructors as the RectangleComponent, see that section for documentation regarding those.

CircleHitbox

The CircleHitbox has the same constructors as the CircleComponent, see that section for documentation regarding those.

ScreenHitbox

ScreenHitbox is a component which represents the edges of your viewport/screen. If you add a ScreenHitbox to your game your other components with hitboxes will be notified when they collide with the edges. It doesn’t take any arguments, it only depends on the size of the game that it is added to. To add it you can just do add(ScreenHitbox()) in your game, if you don’t want the ScreenHitbox itself to be notified when something collides with it. Since ScreenHitbox has the CollisionCallbacks mixin you can add your own onCollisionCallback, onStartCollisionCallback and onEndCollisionCallback functions to that object if needed.

CompositeHitbox

In the CompositeHitbox you can add multiple hitboxes so that they emulate being one joined hitbox.

If you want to form a hat for example you might want to use two RectangleHitboxs to follow that hat’s edges properly, then you can add those hitboxes to an instance of this class and react to collisions to the whole hat, instead of for just each hitbox separately.

Broad phase

If your game field isn’t huge and does not have a lot of collidable components - you don’t have to worry about the broad phase system that is used, so if the standard implementation is performant enough for you, you probably don’t have to read this section.

A broad phase is the first step of collision detection where potential collisions are calculated. Calculating these potential collisions is faster than to checking the intersections exactly, and it removes the need to check all hitboxes against each other and therefore avoiding O(n²).

The broad phase produces a set of potential collisions (a set of CollisionProspects). This set is then used to check the exact intersections between hitboxes (sometimes called “narrow phase”).

By default, Flame’s collision detection is using a sweep and prune broadphase step. If your game requires another type of broadphase you can write your own broadphase by extending Broadphase and manually setting the collision detection system that should be used.

For example, if you have implemented a broadphase built on a magic algorithm instead of the standard sweep and prune, then you would do the following:

class MyGame extends FlameGame with HasCollisionDetection {
  MyGame() : super() {
    collisionDetection =
        StandardCollisionDetection(broadphase: MagicAlgorithmBroadphase());
  }
}

Quad Tree broad phase

If your game field is large and the game contains a lot of collidable components (more than a hundred), standard sweep and prune can become inefficient. If it does, you can try to use the quad tree broad phase.

To do this, add the HasQuadTreeCollisionDetection mixin to your game instead of HasCollisionDetection and call the initializeCollisionDetection function on game load:

class MyGame extends FlameGame with HasQuadTreeCollisionDetection {
  @override
  void onLoad() {
    initializeCollisionDetection(
      mapDimensions: const Rect.fromLTWH(0, 0, mapWidth, mapHeight),
      minimumDistance: 10,
    );
  }
}

When calling initializeCollisionDetection you should pass it the correct map dimensions, to make the quad tree algorithm to work properly. There are also additional parameters to make the system more efficient:

  • minimumDistance: minimum distance between objects to consider them as possibly colliding. If null - the check is disabled, it is default behavior

  • maxObjects: maximum objects count in one quadrant. Default to 25.

  • maxDepth: maximum nesting levels inside quadrant. Default to 10

If you use the quad tree system, you can make it even more efficient by implementing the onComponentTypeCheck function of the CollisionCallbacks mixin in your components. It is useful if you need to prevent collisions of items of different types. The result of the calculation is cached so you should not check any dynamic parameters here, the function is intended to be used as a pure type checker:

class Bullet extends PositionComponent with CollisionCallbacks {

  @override
  bool onComponentTypeCheck(PositionComponent other) {
    if (other is Player || other is Water) {
      // do NOT collide with Player or Water
      return false;
    }
    // Just return true if you're not interested in the parent's type check result.
    // Or call super and you will be able to override the result with the parent's
    // result.
    return super.onComponentTypeCheck(other);
  }

  @override
  void onCollisionStart(
    Set<Vector2> intersectionPoints,
    PositionComponent other,
  ) {
    // Removes the component when it comes in contact with a Brick.
    // Neither Player nor Water would be passed to this function
    // because these classes are filtered out by [onComponentTypeCheck]
    // in an earlier stage.
    if (other is Brick) {
      removeFromParent();
    }
    super.onCollisionStart(intersectionPoints, other);
  }
}

After intensive gameplay a map could become over-clusterized with a lot of empty quadrants. Run QuadTree.optimize() to perform a cleanup of empty quadrants:

class QuadTreeExample extends FlameGame
        with HasQuadTreeCollisionDetection {

  /// A function called when intensive gameplay session is over
  /// It also might be scheduled, but no need to run it on every update.
  /// Use right interval depending on your game circumstances
  onGameIdle() {
    (collisionDetection as QuadTreeCollisionDetection)
            .quadBroadphase
            .tree
            .optimize();
  }
}

Note

Always experiment with different collision detection approaches and check how they perform on your game. It is not unheard of that QuadTreeBroadphase is significantly slower than the default. Don’t assume that the more sophisticated approach is always faster.

Ray casting and Ray tracing

Ray casting and ray tracing are methods for sending out rays from a point in your game and being able to see what these rays collide with and how they reflect after hitting something.

For all of the following methods, if there are any hitboxes that you wish to ignore, you can add the ignoreHitboxes argument which is a list of the hitboxes that you wish to disregard for the call. This can be quite useful for example if you are casting rays from within a hitbox, which could be on your player or NPC; or if you don’t want a ray to bounce off a ScreenHitbox.

Ray casting

Ray casting is the operation of casting out one or more rays from a point and see if they hit anything, in Flame’s case, hitboxes.

We provide two methods for doing so, raycast and raycastAll. The first one just casts out a single ray and gets back a result with information about what and where the ray hit, and some extra information like the distance, the normal and the reflection ray. The second one, raycastAll, works similarly but sends out multiple rays uniformly around the origin, or within an angle centered at the origin.

By default, raycast and raycastAll scan for the nearest hit irrespective of how far it lies from the ray origin. But in some use cases, it might be interesting to find hits only within a certain range. For such cases, an optional maxDistance can be provided.

To use the ray casting functionality you have to have the HasCollisionDetection mixin on your game. After you have added that, you can call collisionDetection.raycast(...) on your game class, or with the HasGameReference Mixin from other components as well.

Example:

ray_cast.dart
 1import 'package:flame/collisions.dart';
 2import 'package:flame/components.dart';
 3import 'package:flame/game.dart';
 4import 'package:flame/geometry.dart';
 5import 'package:flame/palette.dart';
 6import 'package:flutter/material.dart';
 7
 8class RayCastExample extends FlameGame with HasCollisionDetection {
 9  final origin = Vector2(20, 20);
10
11  final direction = Vector2(1, 0);
12
13  final velocity = 60;
14  double get resetPosition => -canvasSize.y;
15
16  Paint paint = Paint()..color = Colors.red.withValues(alpha: 0.6);
17
18  RaycastResult<ShapeHitbox>? result;
19
20  @override
21  Future<void> onLoad() async {
22    final paint = BasicPalette.gray.paint()
23      ..style = PaintingStyle.stroke
24      ..strokeWidth = 2.0;
25
26    add(ScreenHitbox());
27
28    add(
29      CircleComponent(
30        position: canvasSize / 2,
31        radius: 30,
32        paint: paint,
33        children: [CircleHitbox()],
34      ),
35    );
36  }
37
38  @override
39  void update(double dt) {
40    super.update(dt);
41    final ray = Ray2(
42      origin: origin,
43      direction: direction,
44    );
45    result = collisionDetection.raycast(ray);
46
47    origin.y += velocity * dt;
48
49    if (origin.y > canvasSize.y) {
50      origin.y += resetPosition;
51    }
52  }
53
54  @override
55  void render(Canvas canvas) {
56    super.render(canvas);
57
58    if (result != null && result!.isActive) {
59      final originOffset = origin.toOffset();
60      final intersectionPoint = result!.intersectionPoint!.toOffset();
61      canvas.drawLine(
62        originOffset,
63        intersectionPoint,
64        paint,
65      );
66
67      canvas.drawCircle(originOffset, 10, paint);
68    }
69  }
70}

class MyGame extends FlameGame with HasCollisionDetection {
  @override
  void update(double dt) {
    super.update(dt);
    final ray = Ray2(
        origin: Vector2(0, 100),
        direction: Vector2(1, 0),
    );
    final result = collisionDetection.raycast(ray);
  }
}

In this example one can see that the Ray2 class is being used, this class defines a ray from an origin position and a direction (which are both defined by Vector2s). This particular ray starts from 0, 100 and shoots a ray straight to the right.

The result from this operation will either be null if the ray didn’t hit anything, or a RaycastResult which contains:

  • Which hitbox the ray hit

  • The intersection point of the collision

  • The reflection ray, i.e. how the ray would reflect on the hitbox that it hix

  • The normal of the collision, i.e. a vector perpendicular to the face of the hitbox that it hits

If you are concerned about performance you can pre create a RaycastResult object that you send in to the method with the out argument, this will make it possible for the method to reuse this object instead of creating a new one for each iteration. This can be good if you do a lot of ray casting in your update methods.

raycastAll

Sometimes you want to send out rays in all, or a limited range, of directions from an origin. This can have a lot of applications, for example you could calculate the field of view of a player or enemy, or it can also be used to create light sources.

Example:

class MyGame extends FlameGame with HasCollisionDetection {
  @override
  void update(double dt) {
    super.update(dt);
    final origin = Vector2(200, 200);
    final result = collisionDetection.raycastAll(
      origin,
      numberOfRays: 100,
    );
  }
}

In this example we would send out 100 rays from (200, 200) uniformly spread in all directions.

If you want to limit the directions you can use the startAngle and the sweepAngle arguments. Where the startAngle (counting from straight up) is where the rays will start and then the rays will end at startAngle + sweepAngle.

If you are concerned about performance you can re-use the RaycastResult objects that are created by the function by sending them in as a list with the out argument.

Ray tracing

Ray tracing is similar to ray casting, but instead of just checking what the ray hits you can continue to trace the ray and see what its reflection ray (the ray bouncing off the hitbox) will hit and then what that casted reflection ray’s reflection ray will hit and so on, until you decide that you have traced the ray for long enough. If you imagine how a pool ball would bounce on a pool table for example, that information could be retrieved with the help of ray tracing.

Example:

ray_trace.dart
 1import 'dart:math';
 2
 3import 'package:flame/collisions.dart';
 4import 'package:flame/components.dart';
 5import 'package:flame/events.dart';
 6import 'package:flame/game.dart';
 7import 'package:flame/geometry.dart';
 8import 'package:flame/palette.dart';
 9import 'package:flutter/material.dart';
10
11class RayTraceExample extends FlameGame
12    with HasCollisionDetection, TapDetector {
13  Paint paint = Paint()..color = Colors.red.withValues(alpha: 0.6);
14  bool isClicked = false;
15
16  Vector2 get origin => canvasSize / 2;
17
18  RaycastResult<ShapeHitbox>? result;
19
20  final Ray2 _ray = Ray2.zero();
21
22  final boxPaint = BasicPalette.gray.paint()
23    ..style = PaintingStyle.stroke
24    ..strokeWidth = 2.0;
25
26  final List<RaycastResult<ShapeHitbox>> results = [];
27
28  @override
29  Future<void> onLoad() async {
30    add(
31      CircleComponent(
32        radius: min(size.x, size.y) / 2,
33        paint: boxPaint,
34        children: [CircleHitbox()],
35      ),
36    );
37  }
38
39  var _timePassed = 0.0;
40
41  @override
42  void update(double dt) {
43    super.update(dt);
44    if (isClicked) {
45      _timePassed += dt;
46    }
47
48    result = collisionDetection.raycast(_ray);
49
50    _ray.origin.setFrom(origin);
51    _ray.direction
52      ..setValues(1, 1)
53      ..normalize();
54    collisionDetection
55        .raytrace(
56          _ray,
57          maxDepth: min((_timePassed * 8).ceil(), 1000),
58          out: results,
59        )
60        .toList();
61  }
62
63  @override
64  void render(Canvas canvas) {
65    super.render(canvas);
66    var originOffset = origin.toOffset();
67    for (final result in results) {
68      if (!result.isActive) {
69        continue;
70      }
71      final intersectionPoint = result.intersectionPoint!.toOffset();
72      canvas.drawLine(
73        originOffset,
74        intersectionPoint,
75        paint,
76      );
77      originOffset = intersectionPoint;
78    }
79  }
80
81  @override
82  void onTap() {
83    super.onTap();
84    if (!isClicked) {
85      isClicked = true;
86      return;
87    }
88    _timePassed = 0;
89  }
90}

class MyGame extends FlameGame with HasCollisionDetection {
  @override
  void update(double dt) {
    super.update(dt);
    final ray = Ray2(
        origin: Vector2(0, 100),
        direction: Vector2(1, 1)..normalize()
    );
    final results = collisionDetection.raytrace(
      ray,
      maxDepth: 100,
    );
    for (final result in results) {
      if (result.intersectionPoint.distanceTo(ray.origin) > 300) {
        break;
      }
    }
  }
}

In the example above we send out a ray from (0, 100) diagonally down to the right and we say that we want it the bounce on at most 100 hitboxes, it doesn’t necessarily have to get 100 results since at some point one of the reflection rays might not hit a hitbox and then the method is done.

The method is lazy, which means that it will only do the calculations that you ask for, so you have to loop through the iterable that it returns to get the results, or do toList() to directly calculate all the results.

In the for-loop it can be seen how this can be used, in that loop we check whether the current reflection rays intersection point (where the previous ray hit the hitbox) is further away than 300 pixels from the origin of the starting ray, and if it is we don’t care about the rest of the results (and then they don’t have to be calculated either).

If you are concerned about performance you can re-use the RaycastResult objects that are created by the function by sending them in as a list with the out argument.

Comparison to Forge2D

If you want to have a full-blown physics engine in your game we recommend that you use Forge2D by adding flame_forge2d as a dependency. But if you have a simpler use-case and just want to check for collisions of components and improve the accuracy of gestures, Flame’s built-in collision detection will serve you very well.

If you have the following needs you should at least consider to use Forge2D:

  • Interacting realistic forces

  • Particle systems that can interact with other bodies

  • Joints between bodies

It is a good idea to just use the Flame collision detection system if you on the other hand only need some of the following things (since it is simpler to not involve Forge2D):

  • The ability to act on some of your components colliding

  • The ability to act on your components colliding with the screen boundaries

  • Complex shapes to act as a hitbox for your component so that gestures will be more accurate

  • Hitboxes that can tell what part of a component that collided with something

Examples