My dissertation topic under the title:
Implementing machine learning artificial intelligence (AI) to control an AI agent to evaluate the effects on the player’s gameplay experience compared to existing, more commonplace AI methodologies.
In this project I produced two different artefacts:
1) Playable FPS game with enemy AI
2) Proof of concept for a machine learning AI cube
I used these two different artefacts to conduct research and come to conclusions on whether machine learning could be a more beneficial method in which to create AI agents within games.
FPS Enemy AI
The enemy AI uses a custom behaviour tree to switch between different behaviours, including patrol, combat and flee, depending on various factors in the gameplay.
This video was produced for study participants so they fully understood how the AI would act in the demo.
Machine Learning Cube
The cube was taught how to move to the location of the cube by using reinforcement learning.
The cube was rewarded if it reached the goal and punished if it collided with the walls or did not reach the goal.
After a period of training, the cube could consistently reach the yellow sphere.
In addition to the development of the two demos, the project also required background reading into relevant literature.
Individuals who participated in the study were required to play/watch the two AI demos and then answered a subsequent questionnaire. The results of which, alongside the knowledge gained from the background reading, informed a conclusion to the dissertation project.
The project received a first-degree grade.
