This project was created as part of Jessica Casey's dissertation (module code CSC3094) at Newcastle University.
This project aims to create a solution that automates user feedback testing by training a reinforcement learning agent to make decisions in a turn-based combat game based on its perception of any hints provided. The agent's performance under a given hint system should be compared to that when no hints are provided to assess the effectiveness of a hint system.
This project contains three versions of a turn-based combat game. They are almost identical, with one key difference. One scene displays no hints, while the remaining scenes utilise different methods of offering hints.
The agent was trained using Unity's ML-Agents Toolkit, an open-source library that enables training intelligent agents via communication between the Unity game engine and a Python API.