A simple collision approach

Collision is one of the core elements in game-development. Without collision we weren’t able to bring the most simple games to life. But there are a LOT of ways to start dealing with this particular problem since it’s not as simple as it sounds.

Possible complications:

  • We have to figure out who is colliding with each other
  • We have to specify who is allowed to interact with each other
  • We need a system which is not to complicated to expand
  • How detailed will our collision be?
  • How much performance can we sacrifice for convenience?

These are just some of the problems and questions you might face if you are dealing with this particular problem.

The big engines solved this very well with some interesting approaches, but this article is going to show, how one COULD!!! implement an own 2D collision approach.

I will try to focus on the theoretical part of the idea and how the idea works instead of coding everything into the last detail.

There will be a bit of code, some explanation, some code… aaand so on.

Ok let’s start. First of all I will implement a little helper class to deal with coordinates:

struct Vector2
{
 Vector2(int x = 0, int y = 0)
 : X(x), Y(y)
 {}

 int X;
 int Y;
};

static const Vector2 Origin;

Vector2 contains two int values, our X and our Y position. This class can be used to represent points as well as motions in a 2D space (I am going to presuppose that you guys are familiar with vectors in math). Our Origin is our base-Vector2 with a position of (0|0).

Now we need something which holds the info where our “collidable” begins/ends. We could save this information directly in our Actor (you will see later) but I prefer to outsource such kind of functionality to other components. We will implement a BaseCollisionBody from which we can inherit different CollisionBodies (like squares/circles) so we can easily change the collision body for our later Actors.

class CollisionBody
{
public:
 virtual bool IsCollidingWith(CollisionBody* target) = 0;
 virtual bool Contains(Vector2 pos)= 0;

 std::vector<Vector2>& GetBorderPositions()
 {
  return m_Points;
 }

 virtual ~CollisionBody(){}

 virtual void Move(Vector2 motion)
 {
  for (auto point : m_Points)
  {
   point.X += motion.X;
   point.Y += motion.Y;
  }
 }

private:
 std::vector<Vector2> m_Points;
};

This is our BaseBody. We got an IsCollidingWith() pure virtual function which will allow us to check if two Bodies are colliding with each other. We got a dynamic array which stores our border points so we can implement different forms with a different number of points. Futhermore we got a Contains() function which will later on implement the functionality to check if a point is contained by our body. Last but not least we got a Move() function which allows us to move our body.

Now we need at least one specified CollisionBody to cover the form of a sprite where we want to add collision to. I decided to go with a square.

class SquareCollisionBody : public CollisionBody
{
public:
 SquareCollisionBody(Vector2 position, int sideLength)
 {
 Vector2 pointA = position;

 pointA.X -= sideLength / 2;
 pointA.Y -= sideLength / 2;

 Vector2 pointB = position;

 pointB.X += sideLength / 2;
 pointB.Y -= sideLength / 2;

 Vector2 pointC = position;

 pointC.X += sideLength / 2;
 pointC.Y += sideLength / 2;

 Vector2 pointD = position;

 pointD.X -= sideLength / 2;
 pointD.Y += sideLength / 2;

 GetBorderPositions().push_back(pointA);
 GetBorderPositions().push_back(pointB);
 GetBorderPositions().push_back(pointC);
 GetBorderPositions().push_back(pointD);
 }

 bool IsCollidingWith(CollisionBody* target)
 {
  std::vector<Vector2>& targetPoints = target->GetBorderPositions();

  bool areColliding = false;

  for (unsigned int currentPoint = 0; targetPoints.size() > currentPoint; ++currentPoint)
  {
   Vector2 point = targetPoints.at(currentPoint);

   if (Contains(point)
   {
    areColliding = true;
    break;
   }
  }

  if(!areColliding)
  {
    bool condition1 = target->Contains(pointA) || target->Contains(pointB);
    bool condition2 = target->Contains(pointC) || target->Contains(pointD);

    areColliding = condition1 || condition2;
  }

  return areColliding;
 }

 bool Contains(Vector2 point)
 {
  Vector2 pointA = GetBorderPositions().at(0);
  Vector2 pointB = GetBorderPositions().at(1);
  Vector2 pointC = GetBorderPositions().at(2);
  Vector2 pointD = GetBorderPositions().at(3);

  if (point.X >= pointA.X && point.X <= pointB.X && point.Y >= pointA.Y && point.Y <= pointD.Y)
  {
   return true;
  }

  return false;
 }

private:
};

Ok, here we go, that’s our SquareBody implementation. In the constructor we are setting four points A(upper left), B(upper right), C(lower right) and D(lower left) and then we add them to our BaseBody::m_Points vector. We only need those four points to recognize if our square is colliding with something since we can simply check if one of the points is contained in our square. This leads to our simple two functions we are implementing for our BaseClass, Contains & IsCollidingWith. I don’t think I will have to explain those since they are quite simple. (And yes one could do this better, but this will suit our needs for this example)

The next class we will create is our Actor. I added some functions to the actor as placeholder which will give you a little view of how the class might work later. I won’t deal with these placeholders. Our Actor class will serve as base class for our other interactive classes like Character (Actors which have animations or can be moved around or which will be possessed by the AI later), EnviromentActors (Stones, Barriers etc…) or even projectiles.

class Actor
{
public:
 enum Type
 {
  actor,
  character,
  projectile,
  enviroment,
  background
 };

 class CollisionProperties
 {
 public:
  CollisionProperties(Actor::Type myType, CollisionBody* body, int n, ...)
  : m_CollisionBody(body)
  {
   va_list params;
   va_start(params, n);
 
   for (int currentArg = 0; currentArg < n; ++currentArg)
   {
     m_TypesToInteract.push_back(va_arg(params, Actor::Type));
   }
   va_end(params);
  }

  ~CollisionProperties()
  {
   if (m_CollisionBody != nullptr)
   {
    delete
    m_CollisionBody = nullptr;
   }
  }

  bool CanCollideWith(Type collisionType) const
  {
   return m_TypesToInteract.end() != std::find(m_TypesToInteract.begin(), m_TypesToInteract.end(), collisionType);
  }

  bool IsColliding(Actor* target) const
  {
   if (m_CollisionBody != nullptr)
   {
    const Actor::CollisionProperties& props = target->GetCollisionProps();

    if (CanCollideWith(props.m_CollisionType))
    {
     return
    }
   }
   return false;
  }

  void Move(Vector2 motion)
  {
   m_CollisionBody->Move(motion);
  }

private:
  Actor::Type m_CollisionType;
  CollisionBody* m_CollisionBody;
  std::vector<Actor::Type> m_TypesToInteract;
 };

 Actor(Actor::CollisionProperties* props, Vector2 position)
 : m_CollisionProps(props), m_Position(position)
 {}

 virtual ~Actor()
 {
  if (m_CollisionProps != nullptr)
  {
   delete m_CollisionProps;
   m_CollisionProps = nullptr;
  }
 }

 virtual bool IsColliding(Actor* actor)
 {
  return m_CollisionProps->IsColliding(actor);
 }

 virtual void Move(Vector2 motion)
 {
  m_CollisionProps->Move(motion);
  m_Position.X += motion.X;
  m_Position.Y += motion.Y;
 }

 virtual void Update()
 {
 //Some Stuff...
 }

 virtual void Draw()
 {
 //Some Stuff...
 }

 virtual const CollisionProperties& GetCollisionProps() const
 {
  return *m_CollisionProps;
 }

private:
 //...
 //Some Stuff...
 //...

 CollisionProperties* m_CollisionProps;
 Vector2              m_Position;
};

So this is our actor. Our actor contains a subclass CollisionProperties which will handle the collision for the actor. The properties class contains an Actor::Type member which specifies, which kind of actor we got (environment etc. see the enum) as well as the CollisionBody and a vector which contains the collision types, our actor can interact with. Our Actor is basically calling the functions we already implemented in our CollisionBody with the little difference that we can specify in our m_TypesToInteract which kind of collisions shall be ignored by our later implemented inherited subactors.

Let’s take a character for instance:

class Character : public Actor
{
 Character(const std::string& name, CollisionBody* body)
 : Actor(new Actor::CollisionProperties(Actor::character, body, 3, Actor::character, Actor::enviroment, Actor::projectile), Origin), m_CharacterName(name)
 {}

 std::string m_CharacterName; 
};

This class get’s all its features like update, draw and move from our Actor class, but it specifies which kind of actor(Actor::character) we got and which classes can collide with our character (other characters, environment actors and projectiles). This class got close to all functionalities a character needs (motion, collision, updates etc) but it remains pretty slim by now.

Okay that was a whole lot of stuff. Now we got close to all components we need for our collision together. There are only a few little things missing:

A little event class and an inherited collision event.

class Event
{
public:
 enum Type
 {
 collision
 //...
 //Some Others...
 //
 };

Event(Type type)
 : m_EventType(type)
 {}

Type GetEventType()
 {
 return m_EventType;
 }

private:
 Type m_EventType;
};

class CollisionEvent : public Event
{
public:
 CollisionEvent(Actor* a, Actor* b)
 : Event(Event::collision), m_ParticipantA(a), m_ParticipantB(b)
 {}

Actor* GetPA()
 {
 return m_ParticipantA;
 }

Actor* GetPB()
 {
 return m_ParticipantB;
 }

private:
 Actor* m_ParticipantA;
 Actor* m_ParticipantB;
};

Since you are pretty smart guys, I am pretty sure I won’t have to explain this classes to you because they are actually pretty simple so I will go on with a simple usage example of the collision eco system we just created. I will leave or actor/character creation and focus on the collection of collision events per frame.

class Engine
{
public:
 void Play()
 {
 while (m_Running)
 {
 Collision();
 Update();
 Draw();
 Present();
 }
 }
private:
 void Collision()
 {
 for (unsigned int currentActorA = 0; m_ActorPipeline.size() > currentActorA; ++currentActorA)
 {
 if (m_ActorPipeline.size() < 2)
 {
 return;
 }

for (unsigned int currentActorB = currentActorA; m_ActorPipeline.size() > currentActorB; ++currentActorB)
 {
 Actor* a = m_ActorPipeline.at(currentActorA);
 Actor* b = m_ActorPipeline.at(currentActorB);

if (a != b && a->IsColliding(b))
 {
 m_EventPipeline.push_back(new CollisionEvent(a, b));
 }
 }
 }
 }

void Update()
 {
 std::for_each(m_ActorPipeline.begin(), m_ActorPipeline.end(), [](Actor* actor)->void{actor->Update(); });
 }

void Draw()
 {
 std::for_each(m_ActorPipeline.begin(), m_ActorPipeline.end(), [](Actor* actor)->void{actor->Draw(); });
 }

void Present()
 {
 //Some Stuff...
 }

 //...
 //Some Stuff...
 //...

 std::vector<Actor*> m_ActorPipeline;
 std::vector<Event*> m_EventPipeline;

bool m_Running;
};

Okay, what do we got here. An Update() function, a draw function and our present function. Pretty basic if you ask me so I am going to leave them out. Collision().. well, that’s the interesting part. In this function we are handling our collision and there we will create our CollisionEvents and add them to our EventPipeline.

Basically we are iterating through out actor pipeline and we check our actors for collisions. If we detect a collision (remember, we specify which kind of actor may collide with which other types.), we will save a CollisionEvent with the two participating actors to our EventPipeline. What you will do with them from there on is no longer my concern since that is not part of this tutorial/article 😉

Puh, that was a whole lot of info I pushed out there. I hope some of you can get some helpful information out of this article. But hey, I know there is a lot of optimization potential in this code but optimal performance wasn’t the goal I wanted to archive with that.

The thing I like most about this approach is the little effort you will have to put into changing this one into a concurrent environment. Basically you have two ways of doing this:

First (and less effective way): You could create a task for every collision check and hand this task over to a thread pool to collect the results in future objects and process them later.

Second: You could add a pipeline for every type of collision object that you have a pipeline for environment actors, one for projectile actors and for each pipeline an own EventPipeline and so on. This way you could process one pipelines per thread and merge the events together in the end. (That’s how I would handle this, but I would handle the collision threads in a pool anyway.)

I hope you liked today’s post, that’s all for today. Have a nice day/evening whatever 😉

Advertisements

Author: mango2go

I am a software developer for a big pharmaceutical trading company. I am interested in C++ and C# programming in general and in game development in particular (only as a hobby atm). My experience about game development is restricted to what I tought me myself so I am still in the process of learning. My goal is to become a professional game developer and find my name in the credits of a popular game one day.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s