Collision & Hitboxes: The Lie That Feels Fair

Collision detection is where your game decides what touched what — and almost every good game lies about it on purpose. The hitbox (the invisible shape used for collisions) is almost never the same as the sprite the player sees. That gap is one of the most important tools you have.

The two shapes you're always juggling:

• ▸The sprite — what the player sees.
• ▸The hitbox — the rectangle (or circle) the physics actually checks. Usually an AABB: Axis-Aligned Bounding Box, the cheapest, fastest collision shape there is.

The rule that makes games feel fair: Make the player's hurtbox slightly smaller than their sprite, and the player's attack hitbox slightly larger than the visual. Enemies get the opposite treatment. The result: hits the player lands feel generous; hits the player takes feel deserved. Real-world examples ship this everywhere — Super Mario's hurtbox is famously forgiving near the edges; bullet-hell games (Touhou, Ikaruga) shrink the player's hitbox to a single pixel at the center so near-misses read as skill, not luck.

Why this isn't cheating: The player's brain judges fairness by what it sees just before a hit. A pixel-perfect simulation produces deaths the player swears they avoided — that's the simulation lying to their perception. A forgiving hitbox makes the visible truth and the mechanical truth agree. (This is the same forgiveness principle behind coyote time, applied to space instead of time.)

The practical foundation: For a 2D platformer, you almost never need pixel-perfect or polygon collision. AABB-vs-AABB overlap tests are enough for 95% of games and are trivially fast. Reach for circles for projectiles and round things; reach for polygon colliders only when geometry genuinely demands it.