Within the specialized domain of computational geometry and game physics, the GJK algorithm stands as a cornerstone for efficient collision detection. Named after its creators Gilbert, Johnson, and Keerthi, this method determines whether two convex shapes intersect without requiring the computationally expensive examination of every polygon edge. Its primary appeal lies in the iterative construction of a simplex, a geometric structure that probes the Minkowski Difference of the two objects to find the closest approach.
Foundational Mechanics of the GJK Algorithm
The algorithm operates on a fundamental geometric principle: if two convex objects overlap, the origin of the coordinate system must lie within the Minkowski Difference of those objects. Instead of calculating the difference explicitly, GJK uses a support function to sample the farthest point in a given direction. By iteratively querying these points, it builds a simplex—a triangle in 2D or a tetrahedron in 3D—that attempts to enclose the origin. The core loop refines this simplex, checking if the new point allows the shape to enclose the origin or if the search direction has reversed, indicating a miss.
The Role of the Support Function
The support function is the engine of the GJK algorithm, providing the necessary directional information without heavy computation. Given a direction vector, it returns the farthest point on the shape along that vector. Crucially, when applied to the Minkowski Difference, the support function can be computed by subtracting the support points of the individual shapes. This property allows the algorithm to work with complex mesh geometries by leveraging the vertices of the collision shapes, maintaining high performance even with detailed models.
Simplex Evolution and Convergence
The evolution of the simplex is a dynamic process that dictates the path to a solution. A line segment indicates the objects are approaching but not yet intersecting; a triangle suggests a potential edge collision in 2D; and a tetrahedron implies a volume collision in 3D. At each iteration, the algorithm identifies the closest feature of the simplex to the origin and updates the search direction to head toward the origin. If the new support point fails to extend beyond the simplex in the current search direction, the algorithm concludes that the objects are separated, and the penetration vector can be derived from the closest point on the simplex.
While GJK excels at determining intersection, it lacks the ability to calculate the collision response data required for physics simulation. This is where the Expanding Polytope Algorithm (EPA) becomes necessary. Once GJK confirms a collision by finding a simplex enclosing the origin, EPA uses that simplex as a starting point. It iteratively expands the polytope by finding the face closest to the origin and adding the supporting direction to push the shape outward. The result is the minimum translation vector, which provides both the depth and direction needed to resolve the collision physically.
Performance and Practical Considerations
The efficiency of the GJK algorithm makes it ideal for real-time applications, often converging in just a few iterations. However, its performance is sensitive to the initial direction and the quality of the input geometry. Poorly scaled objects or shapes with concave regions can challenge the core logic, necessitating pre-processing or decomposition into convex sub-shapes. Furthermore, degenerate cases, such as when the simplex contains the origin exactly, require robust numerical handling to avoid division by zero or floating-point instability, ensuring the algorithm remains reliable across diverse simulation scenarios.
