AI & NPC behavior

Hytale ‘s artificial intelligence system is based on a modular , fast , and highly dynamic architecture , designed to allow the creatures of Orbis to adapt their behavior in real time.

This AI aims for one central objective: to produce credible and lifelike NPCs , capable of reacting like a human player would.

Each NPC is capable of:

• understanding one’s environment,
• interpreting his social relationships,
• adjust his decisions in combat,
• interact with players in a believable way,
• produce unpredictable emergent behaviors.

This system makes Orbis feel alive , responsive , and deeply immersive , with behaviors that seem genuinely natural.

Technical Foundations

Entity Component System (ECS)

Hytale ‘s AI uses a modern ECS , notably via the Flecs library , offering:

• A total decoupling of behaviors : each function (pathfinding, perception, decision, animation…) is a modular component.
• A natural parallelization , essential for managing dozens of creatures simultaneously.
• Superior responsiveness : behavioral calculations no longer depend on a centralized Java loop.
• An advanced configuration : each NPC can combine a multitude of components (fear, aggression, hunger, detected threat, emotional state…).

Modular decision-making

AI continuously evaluates:

• his internal state (health, fatigue, morale, posture),
• its environment (obstacles, light, terrain, weather),
• its relationships (allies, prey, predators, factions),
• its immediate stimuli (noise, movement, smell, bait).

This set forms the decision-making basis of the two main systems:
the Utility System and the Combat Action Evaluator .

Utility System: Choose the best action at any time

The Utility System is a mathematical system that assigns a value to each possible action. The best action is automatically selected based on:

• the present danger,
• the opportunity (loot, food, vulnerable target),
• the role of the creature (predator, prey, protector),
• its interspecies relationships,
• their emotional state (calm, fear, aggression, curiosity).

Examples of uses

The system allows for unscripted but consistent behaviors , varying according to individuals and contexts.

Combat Action Evaluator: Tactical Decisions in PvE

The Combat Action Evaluator is the system dedicated to advanced combat decisions.
Thanks to the ECS, it can now:

• analyze several inputs in parallel (target distance, obstacles, angle of attack, health, endurance…),
• recalculating decisions several times per second,
• using feigned behavior, hesitations and margins of error (“fuzziness”),
• produce less predictable attacks.

What NPCs can do in combat

• Change strategy (distance, scrum, retreat).
• Choose a target intelligently (isolated player, main threat).
• Execute or delay an attack depending on the opportune moment.
• Change weapons if the arsenal allows it (e.g. Trork berserker).
• Flee if the situation becomes unfavorable .

The objective: a fight that seems improvised , human, spontaneous.

Perception & Stimuli: How NPCs “read” the world

Perception in Hytale combines several specialized systems:

• Vision : detection of silhouettes, player outlines, projectiles.
• Hearing : sounds of footsteps, impacts, sounds of breaking blocks.
• Odor / Chemical stimuli : mainly for animals (food, danger).
• Ground pressure : vibrations or nearby movements (some sensitive creatures).
• Light & temperature : some species flee from or seek warmth.

These stimuli immediately trigger state transitions via ECS components.

Interspecies relations

Orbis species are not isolated: they follow ecological behavioral patterns :

Types of interactions

• Predator/prey
• Protective symbiosis
• Territorial rivalry
• Temporary alliances (e.g., a friendly NPC helping to repel a common enemy)

Complex example: Fen Stalkers & frogs

The Fen Stalkers:

1. They imitate frogs (jumping, posture, curiosity)
2. They actively protect them
3. They attack any predator that approaches a frog
4. Make hunting risky for the player

This is an example of an interspecies relationship coded via the Utility System + ECS , without a static script.

Behavioral scripts & animation states

NPCs change state according to modular scripts connected to the ECS.

main behavioral states

The animations are not decorative: they reflect the internal state
and can influence decision-making (fatigue, recovery, pride after a workout, etc.).

Case studies: creatures & specific behaviors

Fen Stalkers

• advanced behavioral imitations
• active protection of frogs
• conditioned aggression only in the event of a threat
• use of a comprehensive social/ecological model

Trorks

Mice & Bait

• default avoidance behavior
• strongly attracted to cheese
• Several mice can eat simultaneously
• A mouse that is eating becomes less fearful for a few seconds.
• Grouping around chests/dungeons: natural environmental clues

Navigation, terrain & contextual behaviors

AI takes into account:

• terrain (jumps, detours, slope)
• Destructible obstacles (Goblins using bombs to clear the way)
• water, mud, fire, toxic areas
• weather and day/night cycles (many creatures change their behavior at night)

Pathfinding is asynchronous and parallel , therefore much more reactive.
Some creatures even adapt their strategy according to the biome or temperature .

Tier system: classifying enemies by difficulty

This system enriches the progression and diversity of the bestiary.