|

Joshua Kraft, Sofie Curtis, Ethan Pangilinan
Recompiled Game Screenshot Recompiled Game Screenshot

A first-person PvE rogue-lite built in Unreal Engine, with an AI Director that adapts enemy encounters, attributes, and rewards based on player behavior.

Project Overview

Recompiled was developed for the course CPSC 586 (Artificial Intelligence in Games). Our defining "AI" feature is a custom AI Director that drives the entire gameplay experience, dynamically adapting difficulty through altered enemy composition, spawning, and rewards based on how the player performs.

The player progresses through a hostile environment by reclaiming checkpoints, surviving adaptive enemy waves, collecting weapons and abilities, and ultimately, collecting enough "cores" to unlock the bomb and destroy the robot factory. Being a rogue-like game, death is somewhat frequent, and expected, allowing players to learn and adapt their strategy on successive runs.


Technologies & Tools

Unreal Engine 5
Blueprint
Behavior Trees
State Trees
Game AI

The AI Director

The AI Director observes the player's combat strategies and associates them with one of four player models: Fighter, Ranged, Stealth, or Balanced. It then formulates a counterstrategy, determining enemy types and spawn locations to consistently challenge the player and force them to adapt.

Calculating a Player Model

During non-challenge waves, the Director tracks:

These percenteges are weighted against preset importance weights for each model to determine the best match. Ties or near-ties result in a balanced classification.

Director

Strategic Spawning

The Director uses the player model to spawn enemies that target weaknesses:

Spawn elevations are also factored in - ranged enemies typically spawn on higher ground, while melee enemies spawn level with the player

Challenges use many fixed wave sizes and subwave patterns for more-consistent difficulty, while free-roam spawns vary based on detected nearby spawners.

Enemies

Challenge Rewards

After completing a challenge, the Director calculates ability buffs (HP or movement speed) based on performance:

This probability-based system adds variation while rewarding skilled play. Occasional lucky boosts are by design - adding an element of unpredictability while giving struggling players occasional breaks.

Rewards system

Blueprint Deep Dive

Some of the blueprints that help enable core gameplay mechanics. Ctrl + Drag to pan, scroll or pinch to zoom.

Enemy Challenges

Manages checkpoint challenge initiation and enemy wave configuration.

Spawn Next Wave

Controls enemy spawning logic, triggering subsequent waves dynamically.

Challenge Reward Completion

Calculates and distributes rewards based on a variety of challenge performance metrics.


Videos

Trailer

Gameplay


Reflection

This project was a valuable learning experience for me, beyond just the technical aspects of game development. I learned a lot about the importance of actively understanding the components of a project and how they interact, especially when different team members are responsible for different parts. Constant communication was key to ensuring the project came together the way it did. I also learned the importance of balancing ambition with scope - we had to make many tough decisions about which features to prioritize and which to cut in order to deliver a polished deliverable within the project's timeline. Overall, I'm super proud of what we accomplished and the things I learned throughout the process.