Game AI Design

Problem

NPC logic is often written as “see the target, move straight there,” which causes agents to jitter against walls, get stuck on corners, or overwrite each other’s state. The harder part is decoupling decision, routing, and motion: what to do, where to route, and how to move should not be tangled together.

How It Works

• Decision model: use an FSM for a small number of clear states, a behavior tree for priorities and interruption, or utility scoring for continuous preferences.
• Pathfinding: choose a grid, waypoint graph, or navmesh that fits the level; generate a path with A*, and avoid recomputing every frame.
• Steering: follow the next waypoint with seek / arrive, rather than aiming directly at the goal, so agents round corners correctly.
• Verification: check whether the agent reaches the goal, sticks to corners, or oscillates between states; draw the path and current state while tuning.

Boundaries

This fits engine-neutral NPC behavior architecture. Use the relevant Unity, Unreal, or Godot skill for concrete navmesh and agent APIs. Movement feel, collision radius, and tower-defense spawn lanes are out of scope.