When would you prefer a finite state machine over a behavior tree for NPC AI?

When would you prefer a finite state machine over a behavior tree for NPC AI?

• A. FSMs are more scalable and modular.
• B. Behavior trees are simple and deterministic.
• C. FSMs are best for simple deterministic enemies; behavior trees are more modular and scalable.
• D. Behavior trees are only for environmental storytelling.

Answer: (C)

Finite state machines are most effective when the NPC has a small, well-defined set of behaviors and transitions are clear and deterministic. If an enemy simply patrols, notices the player, chases, attacks, or retreats with straightforward rules, a finite number of states and explicit transitions keeps the logic easy to reason about. Each state represents a behavior, and swapping between states happens under specific conditions, so you can predict exactly how the NPC will respond in a given situation. The limit comes when behavior becomes more complex. Adding more actions and nuanced responses causes the state space to grow rapidly, leading to a tangled web of states and transitions that becomes hard to maintain. Behavior trees address this by organizing tasks hierarchically into modular, reusable components. They let you compose behaviors from smaller parts, add new capabilities without exploding the number of states, and scale more gracefully as AI demands increase. That’s why, for simple deterministic enemies, a finite state machine is the best fit, while behavior trees excel when you need modularity and scalability. The other statements don’t fit because FSMs aren’t inherently more scalable or modular than BTs, BTs aren’t confined to simplicity or determinism, and BTs aren’t solely for environmental storytelling.

Finite state machines are most effective when the NPC has a small, well-defined set of behaviors and transitions are clear and deterministic. If an enemy simply patrols, notices the player, chases, attacks, or retreats with straightforward rules, a finite number of states and explicit transitions keeps the logic easy to reason about. Each state represents a behavior, and swapping between states happens under specific conditions, so you can predict exactly how the NPC will respond in a given situation.

The limit comes when behavior becomes more complex. Adding more actions and nuanced responses causes the state space to grow rapidly, leading to a tangled web of states and transitions that becomes hard to maintain. Behavior trees address this by organizing tasks hierarchically into modular, reusable components. They let you compose behaviors from smaller parts, add new capabilities without exploding the number of states, and scale more gracefully as AI demands increase.

That’s why, for simple deterministic enemies, a finite state machine is the best fit, while behavior trees excel when you need modularity and scalability. The other statements don’t fit because FSMs aren’t inherently more scalable or modular than BTs, BTs aren’t confined to simplicity or determinism, and BTs aren’t solely for environmental storytelling.