Introduction
One of the main challenges of AI is creating general intelligence; this is an agent that excels at many tasks and not just one. In the area of gaming, research into General Video Game Playing (GVGP) has become very popular especially with the advent of frameworks like GVGAI (General Video Game Artificial Intelligence). For my final year project I presented a new Statistical Forward Planning method known as Rolling Horizon NeuroEvolution of Augmenting Topologies or rhNEAT for short. This algorithm combines features from two existing algorithms namely, Rolling Horizon Evolutionary Algorithm (RHEA) and NeuroEvolution of Augmenting Topologies (NEAT). In my paper I discussed my finding from analysing different variants of rhNEAT as well as its performance compared to other methods such as RHEA and Monte Carlo Tree Search (MCTS).
For more details about the specifics you can read my full dissertation here alterantively you can read the condensed published version here.
This project was completed using GVGAI a Java based framework. The framework provided 20 games to test my AI agents on as well as the interface for agents to choose which actions to take in said games.
It also includes a forward model allowing for simulation of future states. Furthermore, the framework has been extensively used to test other SFP methods such as RHEA.
The source code for the rhNEAT agent can be found here: source
Accomplishments
I was given the oppurtunity to work on a condensed version of my dissertation with my supervisor Diego Perez-Liebana as part of the IEEE 2020 Conference on Games. The paper was accepted for publication and will be made available on IEEE Xplore from August 2020. This was one of the greatest accomplishments from conducting this project. I also achieved 90.6% on the final report something which I am very proud of.