Flying enemy AI (Unreal Engine 5)

A Lyra-based Unreal Engine 5 project, a bullet hell with Returnal as the reference, unreleased. Developer: Nvizzio Creations. I worked as gameplay and AI programmer. The work breaks into five parts.

Octree flight

• Enemies had been on the standard navmesh. I moved them onto an octree so they fly through volume instead of over a surface.
• The navigation is multithreaded C++ — a reusable native FlyTo task, not a wrapper around MoveTo.
• One task, three callers: StateTree, Behavior Trees and Animation Blueprints all drive it.
• Roughly 100 flying enemies at 60 FPS in the target scenario.

The movement-pattern framework — the core, and entirely mine

• Patterns are fully data-driven assets. Blueprint override points sit on C++ entry points.
• C++ carries the weight: the framework, the flights, the searches, the collections, the traversal and the heavy methods.
• It is network-replicated.
• Targets are polymorphic — a player, a direction, a point relative to the enemy, or a point that exists only because PCG made it.
• Example: an enemy flies a diamond, stops at a corner, attacks halfway through the stop, then attacks again at 50% of the next leg. Both triggers are data; the framework fires a GAS ability by reference.

The other three parts

• Niagara pattern replication, prediction and health attributes. I did not own the bullet-hell attack patterns or the Niagara system behind them — a colleague wrote those. I worked on that system directly for replication, prediction and applying the effect to health.
• Do Movement Pattern, one custom C++ node of mine, brought old Behavior-Tree-and-navmesh enemies onto the framework and the octree.
• Gates and wave spawning are mine. Gates open, enemies spawn in waves inside the procedural space, with a difficulty budget per level and a cost per enemy.