UE5 Reinforcement Learning

Reinforcement Learning AI systems implemented in Unreal Engine 5 (C++), exploring both classic Q-Learning and Deep Q-Network (DQN) approaches for adaptive enemy behavior in real-time combat.

Note: This is a reupload of academic projects completed in 2024.
Not under active development, but preserved to demonstrate applied reinforcement learning techniques.
Source directories contain only the AI-related C++ and Python files — not full Unreal Engine project data.

---

Overview

These projects demonstrate how tabular and neural reinforcement learning methods can be applied to enemy AI in Unreal Engine 5.
Agents learn through reward signals to attack, dodge, guard, and wait based on real-time parameters.

---

Repository Organization

UE5_ReinforcementLearning_AI/  
├── QLearning/  
│ ├── ue5_source/ # Referenced UE5 C++ source files for Q-Learning logic  
│ └── report/     # Presentation and brief summary for Q-Learning project  
│  
└── DQN/  
├── ue5_source/   # UE5-side C++ code interacting with the Python server  
├── dqn_source/   # Python server code (model, training loop, comm layer)  
└── report/       # Presentation and brief summary for DQN project  

---

Algorithms Implemented

• Q-Learning: Classic temporal-difference learning using discrete state-action pairs.
• DQN: Neural approximation of Q-values with experience replay and epsilon-greedy exploration.
• TD-Update: Both versions use Q(s,a) ← Q(s,a) + α [r + γ max Q(s′, a′) − Q(s,a)].

---

Reports & Slides

Detailed summaries and presentations for each project are included under their respective report/ folders.

---

Author

George Kotti — B.S. in Computer Science & Engineering (AI + Systems)
Developed as part of academic research on Reinforcement Learning for Game AI.