At the moment I have built a custom Vector2 class and have a character object that has an acceleration vector (modified by vector forces), a velocity, and a position as well as bounding boxes for every entity within the scene.

Right now my physics works quite well and the character falls and moves with gravity and player input and is able to detect when it collides with another rectangle bounding box shown here:

    PhysicsEngine::TimeStepProcess(float deltaTime, 
        Entity* character, std::vector<Walls*> walls, 
        double screenWidth, double screenHeight,
        int timeStep) 
        double processedDeltaTime = deltaTime / timeStep;

        //Process entire physics engine multiple times per step with
        //a smaller deltaTime for more accuracy
        for (int i = 0; i < timeStep; ++i) {
            //Add forces back per frame
            //ApplyForce(character, gravity);
            ApplyForce(character, character->GetForce());

            //Calculate and resolve collisions
            ResolveCollisions(character, walls, processedDeltaTime);

            //Set angular acceleration
            angularAcceleration = character->GetTorque() / character->GetIntertia();

            //Add Velocity per frame for linear and angular
            Vector2::ScaleMult(*character->GetVelocity(), *linearAcceleration, processedDeltaTime, Vector2::ADDITON);
            *character->GetAngularVelocity() += angularAcceleration * processedDeltaTime;

            //Add displacement per frame for linear and angular
            Vector2::ScaleMult(*character->GetPos(), *character->GetVelocity(), processedDeltaTime, Vector2::ADDITON);
            *character->GetAngle() += *character->GetAngularVelocity() * processedDeltaTime;

            //Add friction
            CalculateFriction(character, processedDeltaTime);

            //Clear forces per frame
            Vector2::ScaleMult(*linearAcceleration, *linearAcceleration, 0, Vector2::EQUAL);


With the collision being detected quite accurately here:

PhysicsEngine::ResolveCollisions(Entity* character, std::vector<Walls*> walls, double deltaTime)
    double minDistanceFromObject = 1; //pixels

    //Iterate through all walls
    vector<Walls*>::iterator wallIter;
    for (wallIter = walls.begin(); wallIter != walls.end(); ++wallIter) {
        if (character->GetCollisionBox()->CheckCollision((*wallIter)->GetCollisionBox())) {
            //Resolve collision

//Does this rectangle contain other rectangle
Rectangle::CheckCollision(Rectangle* other) {
    //The sides of the rectangles
    int leftA, leftB;
    int rightA, rightB;
    int topA, topB;
    int bottomA, bottomB;

    //Calculate the sides of rectangle A
    leftA   = static_cast<int>(this->X());
    rightA  = static_cast<int>(this->X() + this->W());
    topA    = static_cast<int>(this->Y());
    bottomA = static_cast<int>(this->Y() + this->H());

    //Calculate the sides of rectangle B
    leftB   = static_cast<int>(other->X());
    rightB  = static_cast<int>(other->X() + other->W());
    topB    = static_cast<int>(other->Y());
    bottomB = static_cast<int>(other->Y() + other->H());

    //If any of the sides from rectangle A are outside of B
    //Then there is no possible way the two boxes are colliding
    if (bottomA <= topB) {
        return false;
    if (topA >= bottomB) {
        return false;
    if (rightA <= leftB) {
        return false;
    if (leftA >= rightB) {
        return false;

    //If none of the sides of the primary rectangle are outside of other
    //Then there must be a collision of some sort
    return true;

But my vector math is quite shaky and I have no idea how to resolve these collisions and stop the character moving through surfaces. I want to be able to walk along floors and collide with any surfaces realistically. I have now been trying to accomplish this for over a week and have gone through at least 20-30 posts, comments, articles and questions about this and cannot understand how it works or how to implement it.

Can someone please help me with the coding aspect of this?


I get this question a lot at my school's physics club. As an immediate answer, check out the source code from Box2D-lite, which is an educational engine designed exactly for this purpose, and is provided by Erin Catto, one of the makers of Box2D. According to him, he recently updated his box2d lite implementation. What you'll find in here is the result of an incredible amount of research and planning rolled up into a small, easy to follow codebase. While you're reading that, I highly encourage you to examine the following presentations which outline the mathematical details pertaining to your question, as I could not explain it all in one go.

Having tried collision resolution before, I can tell you what happens when trying to solve the problem of resolution on your own using kinematic equations alongside the collision equations; you will wind up with tolerable results at first, but you will find that your boxes do not stack, and multibody collisions are a nightmare, and when you get to apply the rotational aspects of collision resolution it will get even harder.

This is why I recommend these resources first; the task of realtime rigid body simulation in games is a non-trivial problem, and there are many many ways of going about solving them. There are tried and true methods such as sequential impulse solvers (which I believe Erin Catto came up with and demonstrates in his engine) or projected gauss-siedel solvers which are industry standards when it comes to this. Currently PhysX, for example, employes a mysterious solver that we're all waiting for documentation on called the Temporal Gauss-Sidel solver, which appears to have made some stability improvements.


Here is the file containing the code you're looking for, but the setup for an iterative solver like this requires a bit of extra design in addition to what you've got. Most importantly caching contacts so they persist over frames is tricky, but for most comprehensive implementations, it's a must.











(i believe Erwin Coumans did a lot of work regarding Featherstone, possibly with Rockstar games in developing their fabulous ragdoll implementation alongside the Euphoria engine) http://box2d.org/files/GDC2014/ErwinCoumans_ExploringMLCPSolversAndFeatherstone.pdf

  • \$\begingroup\$ This is... probably the most detailed answer I've ever gotten. Thank you so so much. A couple of those articles I've actually already read, but Randy Gaul's Game Programming Blog actually gives me a lot of the answers I need. I'm trying to keep my solution simple so as to avoid coding a full engine as this is just a small uni project and what I was really really looking for is how to find the direction of a collision and resolve a collision simple by reversing the velocity direction. But this answer is fantastic and really does answer quite a few questions I've had. \$\endgroup\$
    – Aquaphor
    Sep 11 '19 at 13:03
  • \$\begingroup\$ The only thing I'll need to do is just clean up my vector class cause there's a lot of very shaky vector math in it right now. But again, thank you so much for all these articles and info. This gives me great detail about how to best approach this. Again, I'm trying to keep it very simple. The only thing I'm moving are the entities that collide with the stationary "flooring/surfaces" so hopefully now I can get it working \$\endgroup\$
    – Aquaphor
    Sep 11 '19 at 13:06
  • \$\begingroup\$ Thanks, I put some oomf into it cause I've been exactly where you're at, spending a few months trying to do it all from scratch without any guidance. This is why I started physics club. Incidentally, the struggle is how I came to love physics programming and geometry. Addictive problem solving I guess. Please feel free to hit me up any time, I have a folder full of the collective resources we've been able to accrue over the last few years on the topic plus the means to ask questions to others who might be able to help. \$\endgroup\$
    – Jon
    Sep 11 '19 at 13:16

In particular, for as simple as a possible solution, I could just use this code and make sure that the second object has 0 velocity and infinite mass or has a restitution coefficient of 0.

// j = -(1 + epsilon) * N dot Vrel
// -------------------------------
//        invM_1 + invM_2
void Manifold::ApplyImpulse( void )
  Vec2 rv = B->m_velocity - A->m_velocity;
  real contactVel = Vec2::DotProduct( rv, normal );

  // Do not resolve if velocities are separating
  if(contactVel > 0)

  // Calculate restitution
  real e = std::min(A->m_material.restitution, B->m_material.restitution);

  // Calculate impulse scalar
  real j = -(1.0f + e) * contactVel;
  j /= A->m_massdata.inv_mass + B->m_massdata.inv_mass;

  // Apply impulse
  Vec2 impulse = j * normal;
  A->m_velocity -= A->m_massdata.inv_mass * impulse;
  B->m_velocity += B->m_massdata.inv_mass * impulse;
  • \$\begingroup\$ This Randy Gaul's old code? Just as an FYI, that site is good for conceptual understanding and his explanations, but I would take that code with a grain of salt. It's a bit of an amalgamation. I believe this is where he also employes positional correction, a technique commonly referred to as non-linear gauss siedel. Randy gaul has a more recent implementation of simplified 2D physics covering niche techniques and tricks which he presented to physics club last fall, his demo and slideshow are available on his github github.com/RandyGaul \$\endgroup\$
    – Jon
    Sep 11 '19 at 13:49
  • \$\begingroup\$ @JonKoelzer This is perfect. His slides and demo are exactly what I needed. Thank you so much. \$\endgroup\$
    – Aquaphor
    Sep 12 '19 at 3:26
  • \$\begingroup\$ Happy to help, friend \$\endgroup\$
    – Jon
    Sep 12 '19 at 3:46

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .