Module org.dyn4j

Class Gjk

  • All Implemented Interfaces:
    DistanceDetector, NarrowphaseDetector, RaycastDetector

    public class Gjk
    extends Object
    implements NarrowphaseDetector, DistanceDetector, RaycastDetector
    Implementation of the Gilbert-Johnson-Keerthi (GJK) algorithm for collision detection.

    Gjk is an algorithm used to find minimum distance from one Convex Shape to another, but can also be used to determine whether or not they intersect.

    Gjk uses a specific mathematical construct called the MinkowskiSum. The MinkowskiSum of two Convex Shapes create another Convex Shape. If the shapes are labeled A and B, the MinkowskiSum is the convex hull of adding every point in A to every point in B.

    Now, if we subtract every point in A and every point in B, we still end up with a Convex Shape, but we also get another interesting property. If the two Convex Shapes are penetrating one another (overlapping) then the MinkowskiSum, using the difference operator, will contain the origin.

    Computing the MinkowskiSum directly would not be very efficient and performance would be directly linked to the number of vertices each shape contained (n*m subtractions). In addition, curved shapes have an infinite number of vertices.

    That said, it's not necessary to compute the MinkowskiSum. Instead, to determine whether the origin is contained in the MinkowskiSum we iteratively create a Shape inside the MinkowskiSum that encloses the origin. This is called the simplex and for 2D it will be a triangle. If we can enclose the origin using a shape contained within the MinkowskiSum, then we can conclude that the origin is contained within the MinkowskiSum, and that the two shapes are penetrating. If we cannot, then the shapes are separated.

    To create a shape inside the MinkowskiSum, we use what is called a support function. The support function returns a point on the edge of the MinkowskiSum farthest in a given direction. This can be obtained by taking the farthest point in shape A minus the farthest point in shape B in the opposite direction.

    If the MinkowskiSum is:

     A - B
     
    then the support function would be:
     (farthest point in direction D in A) - (farthest point in direction -D in B)
     
    With this we can obtain a point which is on the edge of the MinkowskiSum shape in any direction. Next we iteratively create these points so that we create a shape (triangle in the 2d case) that encloses the origin.

    Algorithm psuedo-code:

     // get a point farthest in the direction
     // choose some random direction (selection of the initial direction can
     // determine the speed at which the algorithm terminates)
     Point p = support(A, B, direction);
     // add it to the simplex
     simplex.addPoint(p);
     // negate the direction
     direction = -direction;
     // make sure the point we are about to add is actually past the origin
     // if its not past the origin then that means we can never enclose the origin
     // therefore its not in the Minkowski sum and therefore there is no penetration.
     while (p = support(A, B, direction).dot(direction) > 0) {
            // if the point is past the origin then add it to the simplex
            simplex.add(p);
            // then check to see if the simplex contains the origin
            // passing back a new search direction if it does not
            if (check(simplex, direction)) {
                    return true;
            }
     }
     return false;
     
    The last method to discuss is the check method. This method can be implemented in any fashion, however, if the simplex points are stored in a way that we always know what point was added last, many optimizations can be done. For these optimizations please refer to the source documentation on checkSimplex(List, Vector2).

    Once Gjk has found that the two CollisionBodys are penetrating it will exit and hand off the resulting simplex to a MinkowskiPenetrationSolver to find the collision depth and normal.

    Gjk's default MinkowskiPenetrationSolver is Epa.

    The Gjk algorithm's original intent was to find the minimum distance between two Convex Shapes. Refer to distance(Convex, Transform, Convex, Transform, Separation) for details on the implementation.

    Since:
    1.0.0
    Version:
    4.1.0
    Author:
    William Bittle
    See Also:
    Epa, GJK (Gilbert-Johnson-Keerthi), GJK - Distance & Closest Points
    • Field Detail

      • DEFAULT_MAX_ITERATIONS

        public static final int DEFAULT_MAX_ITERATIONS
        The default Gjk maximum iterations
        See Also:
        Constant Field Values
      • DEFAULT_DETECT_EPSILON

        public static final double DEFAULT_DETECT_EPSILON
        The default epsilon in meters for collision detection
        See Also:
        Constant Field Values
      • DEFAULT_DISTANCE_EPSILON

        public static final double DEFAULT_DISTANCE_EPSILON
        The default epsilon in meters for distance checks
      • DEFAULT_RAYCAST_EPSILON

        public static final double DEFAULT_RAYCAST_EPSILON
        The default epsilon in meters for raycast checks
      • maxDetectIterations

        protected int maxDetectIterations
        The maximum number of collision detection iterations
      • maxDistanceIterations

        protected int maxDistanceIterations
        The maximum number of distance check iterations
      • maxRaycastIterations

        protected int maxRaycastIterations
        The maximum number of raycast iterations
      • detectEpsilon

        protected double detectEpsilon
        The collision detection epsilon in meters
      • distanceEpsilon

        protected double distanceEpsilon
        The distance check epsilon in meters
      • raycastEpsilon

        protected double raycastEpsilon
        The raycast check epsilon in meters
    • Method Detail

      • getInitialDirection

        protected Vector2 getInitialDirection​(Convex convex1,
                                              Transform transform1,
                                              Convex convex2,
                                              Transform transform2)
        Returns a vector for the initial direction for the GJK algorithm in world coordinates.

        This implementation returns the vector from the center of the first convex to the center of the second.

        Parameters:
        convex1 - the first convex
        transform1 - the first convex's transform
        convex2 - the second convex
        transform2 - the second convex's transform
        Returns:
        Vector2
      • detect

        protected boolean detect​(MinkowskiSum ms,
                                 List<Vector2> simplex,
                                 Vector2 d)
        The main Gjk algorithm loop.

        Returns true if a collision was detected and false otherwise.

        The simplex and direction parameters will reflect the state of the algorithm at termination, whether a collision was found or not. This is useful for subsequent algorithms that use the GJK simplex to find the collision information (Epa for example).

        Parameters:
        ms - the MinkowskiSum
        simplex - the simplex; should be an empty list
        d - the initial direction
        Returns:
        boolean
      • checkSimplex

        protected boolean checkSimplex​(List<Vector2> simplex,
                                       Vector2 direction)
        Determines whether the given simplex contains the origin. If it does contain the origin, then this method will return true. If it does not, this method will update both the given simplex and also the given search direction.

        This method only handles the line segment and triangle simplex cases, however, these two cases should be the only ones needed for 2 dimensional Gjk. The single point case is handled in detect(MinkowskiSum, List, Vector2).

        This method also assumes that the last point in the simplex is the most recently added point. This matters because optimizations are available when you know this information.

        Parameters:
        simplex - the simplex
        direction - the search direction
        Returns:
        boolean true if the simplex contains the origin
      • findClosestPoints

        protected void findClosestPoints​(MinkowskiSumPoint a,
                                         MinkowskiSumPoint b,
                                         Separation separation)
        Finds the closest points on A and B given the termination simplex and places them into point1 and point2 of the given Separation object.

        The support points used to obtain a and b are not good enough since the support methods of Convex Shapes only return the farthest vertex, not necessarily the farthest point.

        Parameters:
        a - the first simplex point
        b - the second simplex point
        separation - the Separation object to populate
        See Also:
        GJK - Distance & Closest Points
      • containsOrigin

        protected boolean containsOrigin​(Vector2 a,
                                         Vector2 b,
                                         Vector2 c)
        Returns true if the origin is within the triangle given by a, b, and c.

        If the origin lies on the same side of all the points then we know that the origin is in the triangle.

         sign(location(origin, a, b)) == sign(location(origin, b, c)) == sign(location(origin, c, a))
        The Segment.getLocation(Vector2, Vector2, Vector2) method can be simplified because we are using the origin as the search point:
         = (b.x - a.x) * (origin.y - a.y) - (origin.x - a.x) * (b.y - a.y)
         = (b.x - a.x) * (-a.y) - (-a.x) * (b.y - a.y)
         = -a.y * b.x + a.y * a.x + a.x * b.y - a.x * a.y
         = -a.y * b.x + a.x * b.y
         = a.x * b.y - a.y * b.x
         = a.cross(b)
        Parameters:
        a - the first point
        b - the second point
        c - the third point
        Returns:
        boolean
      • raycast

        public boolean raycast​(Ray ray,
                               double maxLength,
                               Convex convex,
                               Transform transform,
                               Raycast raycast)
        Description copied from interface: RaycastDetector
        Performs a ray cast given a Ray and a Convex Shape returning true if the ray passes through the convex shape.

        The raycast parameter is used to stored the results of the raycast when returning true.

        Returns false if the start position of the ray lies inside the given convex.

        Specified by:
        raycast in interface RaycastDetector
        Parameters:
        ray - the Ray
        maxLength - the maximum length of the ray; 0 for infinite length
        convex - the Convex Shape
        transform - the Convex Shape's Transform
        raycast - the ray cast result
        Returns:
        boolean true if the Ray passes through the Convex Shape
      • getMaxDetectIterations

        public int getMaxDetectIterations()
        Returns the maximum number of iterations the Gjk collision detection algorithm will perform before returning that two convex shapes are not overlapping.
        Returns:
        int the number of Gjk detect iterations
        Since:
        3.3.0
        See Also:
        setMaxDetectIterations(int)
      • setMaxDetectIterations

        public void setMaxDetectIterations​(int maxIterations)
        Sets the maximum number of iterations the Gjk collision detection algorithm will perform when before return that two convex shapes are not overlapping.

        Valid values are in the range [5, ∞].

        Parameters:
        maxIterations - the maximum number of Gjk detect iterations
        Throws:
        IllegalArgumentException - if maxIterations is less than 5
        Since:
        3.3.0
      • getMaxDistanceIterations

        public int getMaxDistanceIterations()
        Returns the maximum number of iterations the Gjk distance algorithm will perform before returning the distance between two separated convex shapes.
        Returns:
        int the number of Gjk distance iterations
        Since:
        3.3.0
        See Also:
        setMaxDistanceIterations(int)
      • setMaxDistanceIterations

        public void setMaxDistanceIterations​(int maxIterations)
        Sets the maximum number of iterations the Gjk distance algorithm will perform when determining the distance between two convex shapes.

        Valid values are in the range [5, ∞].

        Parameters:
        maxIterations - the maximum number of Gjk distance iterations
        Throws:
        IllegalArgumentException - if maxIterations is less than 5
        Since:
        3.3.0
      • getMaxRaycastIterations

        public int getMaxRaycastIterations()
        Returns the maximum number of iterations the Gjk raycast algorithm will perform before returning that the ray and the convex are not overlapping.
        Returns:
        int the number of Gjk raycast iterations
        Since:
        3.3.0
        See Also:
        setMaxRaycastIterations(int)
      • setMaxRaycastIterations

        public void setMaxRaycastIterations​(int maxIterations)
        Sets the maximum number of iterations the Gjk raycast algorithm will perform when checking whether the ray intersects the convex.

        Valid values are in the range [5, ∞].

        Parameters:
        maxIterations - the maximum number of Gjk raycast iterations
        Throws:
        IllegalArgumentException - if maxIterations is less than 5
        Since:
        3.3.0
      • getDetectEpsilon

        public double getDetectEpsilon()
        Returns the Gjk detect epsilon.
        Returns:
        double the Gjk detect epsilon
        Since:
        3.3.0
        See Also:
        setDetectEpsilon(double)
      • setDetectEpsilon

        public void setDetectEpsilon​(double detectEpsilon)
        The minimum distance to determine that two shapes are not colliding.

        Valid values are in the range [0, ∞].

        Parameters:
        detectEpsilon - the Gjk detect epsilon
        Throws:
        IllegalArgumentException - if detectEpsilon is less than zero
        Since:
        3.3.0
      • getRaycastEpsilon

        public double getRaycastEpsilon()
        Returns the Gjk raycast epsilon.
        Returns:
        double the Gjk raycast epsilon
        Since:
        3.3.0
        See Also:
        setRaycastEpsilon(double)
      • setRaycastEpsilon

        public void setRaycastEpsilon​(double raycastEpsilon)
        The minimum distance between the ray and convex for the Gjk raycast algorithm.

        Valid values are in the range (0, ∞].

        Parameters:
        raycastEpsilon - the Gjk raycast epsilon
        Throws:
        IllegalArgumentException - if raycastEpsilon is less than or equal to zero
        Since:
        3.3.0
      • getDistanceEpsilon

        public double getDistanceEpsilon()
        Returns the Gjk distance epsilon.
        Returns:
        double the Gjk distance epsilon
        See Also:
        setDistanceEpsilon(double)
      • setDistanceEpsilon

        public void setDistanceEpsilon​(double distanceEpsilon)
        The minimum distance between two iterations of the Gjk distance algorithm.

        Valid values are in the range (0, ∞].

        Parameters:
        distanceEpsilon - the Gjk distance epsilon
        Throws:
        IllegalArgumentException - if distanceEpsilon is less than or equal to zero