Ground & aerial patrol
The dragon patrols on the ground, transitions into flight at path boundaries and continues along aerial movement logic.
01 / AI STATES
Ground, air, combat
The goal was to build a boss that changes behavior rather than simply cycling attacks.
Dynamic attack selection
Attacks are selected from data assets using range, priority and cooldown conditions instead of hard-coded branches.
Target awareness
AI Perception drives target detection and re-engagement while keeping state changes centralized.
Target-loss behavior
When sight is lost, the boss searches a configurable area before either re-acquiring the player or returning to patrol.
02 / C++ NODES
Behavior Tree architecture
I extended Unreal’s Behavior Tree with custom C++ tasks, services and decorators so the boss logic could stay modular and readable.
Selection, flight, investigation
Custom tasks choose attacks, move through aerial patrol points and execute investigation behavior.
Continuous state updates
Services update flying state, monitor target loss and maintain attack cooldown information.
Context validation
Decorators validate attack conditions, visibility and patrol milestones before branches are entered.
UAttackDataAsset
Attack tuning lives in data instead of code, making damage, cooldowns and animation references easier to iterate.
03 / DESIGN
Technical highlights
- ABossController: Owns perception, high-level state changes and attack selection.
- ABossDragonCharacter: Handles movement, attack execution, animation, VFX and cooldown interaction.
- Modular AI: New attacks or behaviors can be introduced without rewriting the full boss state flow.
- Full integration: Animation, audio-visual feedback and gameplay state are triggered from the same underlying AI decisions.