Project Introduction

My project investigates the suitability of the use of machine learning techniques,
in particular Reinforcement Learning (RL), for producing realistic gameplay in
simulation-based games. A scaled-down version of a standard Real-Time Strat-
egy (RTS) game was constructed and a RL-driven opponent was developed. The
RL opponent was trained against artificial & human opponents, with varying
control architectures, differing degrees of complexity.

Once trained, these players were tested against one another to determine each
agent's ability to perform against their opponents, including hand-coded, ma-
chine learning agents and a human opponent. Additionally, the artificial com-
petitors are assessed for realism in their gameplay using criteria such as
level of predictability, the level of challenge ordered and the learning agent's abil-
ity to identify and exploit weaknesses in their opposition.

Next