The Body That Isn't There
You crouch behind a crate. You hear the shot. You're already dead.
Or you swing a sword that passes through empty air, and the enemy just crumples. You've filed both complaints mentally, probably while saying something your mother wouldn't appreciate. Neither moment is a bug. It's a deliberate, carefully chosen lie, and the engineers who built it are not sorry.
Game engines separate two things that look like one: the visual mesh you see on screen, and the collision shape (often called a hitbox or collider) that the physics system actually uses for calculations. Same space, completely different jobs. Why they're different tells you something fundamental about how real-time simulation holds itself together.
Polygons Are Expensive Neighbors
A modern character model in a AAA game can carry 100,000 polygons or more. Each polygon is a triangle with three vertices in 3D space. Checking whether a bullet's path intersects one specific triangle is a solved math problem. Checking it against 100,000 triangles, for every character, every projectile, every frame, sixty times a second, is a catastrophe.
Run the numbers. Twelve enemies on screen, each with 80,000 polygons, ten active projectiles per frame: that's potentially 9.6 million triangle-intersection tests every single frame. At 60 frames per second, you're looking at 576 million tests per second just for bullet collision. The CPU would be doing nothing else. Nothing.
So engines cheat. Beautifully.
Instead of testing against the visual mesh, the physics engine tests against a proxy shape, something geometrically simple that approximates the character's volume. The most common options are capsules (a cylinder with hemispherical caps), axis-aligned boxes, and convex hulls. A capsule test reduces those 80,000 triangle checks to a single mathematical operation: find the closest point on a line segment to another point, then check if the distance is less than a radius. The math resolves in nanoseconds.
A capsule collider for a human character might be roughly 1.8 metres tall and 0.5 metres in diameter. The actual model inside it has detailed shoulders, a narrow waist, hair drifting past the silhouette. None of that matters to the physics engine. It's living inside an invisible pill, which is honestly a more accurate description of game development than most studios would choose for their press kits.
Why Capsules Specifically (and Not Boxes)
A box sounds simpler, and in some ways it is. But boxes have corners, and corners snag. A character walking along uneven terrain can catch on a tiny geometry edge and stutter or stop dead. Capsules don't have that problem. The rounded base slides over small bumps the way a well-worn river stone skips over gravel, which is why virtually every humanoid character in every major engine from Unity to Unreal defaults to a capsule for movement.
Boxes get used for rigid objects where snag doesn't matter: crates, walls, vehicles. Spheres handle rolling objects and some projectiles. Convex hulls, tight polygon cages automatically generated to wrap an irregular shape, get used when you need more accuracy than a capsule but can't afford the full mesh. A hand grenade, say, or a dropped weapon with an awkward profile.
The trade-off is always identical: fidelity costs time, simplicity costs accuracy. Every designer is negotiating that exchange with every object in the scene, on every frame.
The Gap Between What You See and What Gets Hit
This is where the frustration lives.
Take a cover-based shooter. You're crouching behind a low wall, your character's head visually below the top edge. But the capsule collider is still tall enough that its upper hemisphere pokes above the wall's collision geometry. You get shot. The game is technically correct. It feels completely wrong, and that distinction matters more than most engineers like to admit.
This is a tuning problem, not a structural flaw. Studios spend real time shrinking hitboxes on player characters to make the experience feel fair, and expanding them slightly on enemies to make shooting feel satisfying. Counter-Strike has historically made player hitboxes slightly smaller than the visual model, specifically so near-misses register as misses. It rewards precision. The lie becomes a design tool, which is either elegant or mildly unsettling depending on your disposition.
Consider two players: Marcus and Priya, same game, same hardware. Marcus plays a character in flowing robes. Priya plays an armored knight. Visually, Marcus looks bigger because the fabric billows. But if the designer gave both characters the same capsule collider, they're identical targets despite looking nothing alike. Whether that's fair is a design question. Whether it's technically necessary is a physics question. The answers pull in opposite directions, and someone has to make a call.
What People Consistently Misread About This
The widespread assumption is that visible geometry and physical geometry should always match, and any gap is laziness. It isn't. It's a considered exchange, and the people making it usually know exactly what they're giving up.
Perfect mesh collision exists and gets used in specific contexts: slow-moving puzzle games, architectural walkthroughs, some VR experiences where precise physical contact matters more than frame rate headroom. The moment you add crowds, projectiles, and real-time destruction, the full-mesh approach collapses under its own weight.
There's also a subtler point most players miss. The visual mesh itself isn't reality. It's already a simplification of a concept. The collision shape is a simplification of a simplification. Games are nested approximations, each layer optimized for a different purpose. Expecting physical precision from a visual representation is like expecting your shadow to stop a door from closing.
Found yourself raging at a hitbox lately? Check the game's competitive community. For most titles, the hitbox geometry is documented and debated in obsessive detail. You'll often find that what felt like a miss was geometrically a hit, or vice versa, and understanding that changes how you read the whole system.
The invisible capsule isn't a failure of craft. It's what makes craft possible at the speed games demand, and the beautiful mesh wrapped around it could never do that job alone.