Module Directory

This module covers a range of Artificial Intelligence techniques employed in games, and teaches how games are and can be used for research in Artificial Intelligence.

The module explores algorithms for creating agents that play classical board games (such as chess or checkers) and real-time games (Mario or PacMan), including applying deep learning and reinforcement learning to play games.

The course also covers Procedural Content Generation, and explores the techniques used to simulate intelligence in the latest videogames.

Students should either be strong programmers or have some previous Python programming experience to take this module.  The module starts assuming students have proficiency in Python programming from the very first week's lab exercise.

The aims of this module are:

• To provide the student with the core knowledge of the techniques behind the most popular algorithms for Artificial Intelligence.


• To program the agents that play a range of different games, either by themselves or as opponents the player must face.


• To teach how automatic generation of content (complete games, rules, levels and gameplay elements) can be achieved by means of Artificial Intelligence.


• To teach the basics of deep learning and reinforcement learning, and how these can be used to create agents which are capable of playing a range of different games.

By the end of this module, students will be expected to be able to:

1. Understand the techniques behind algorithms that play real-time video games and adversarial board games.


2. Design and program agents that are able to play any given real-time or adversarial board game.


3. Design and program algorithms that generate content for games procedurally, according to some predefined goals.


4. Design and program non-player character behaviours for video games that create the illusion of intelligence.

Syllabus

• Basics of Reinforcement Learning: Value Iteration, Policy iteration, Monte Carlo, Temporal Difference Learning, Q-Learning. The Multi-Armed Bandit Problem, Upper Confidence Bound for Trees, Monte Carlo Tree Search.


• Nature Inspired Intelligence: Neural Networks and Genetic Algorithms Adversarial Games: Minimax, Alpha-Beta Search, Monte Carlo Tree Search.


• A Case Study: Deep Learning in Game Playing Steering Behaviours for Autonomous Characters.


• Path-finding: A*, Dijkstra's algorithm, Waypoint-based path-finding, Navigational Meshes.


• Agent Behaviour: Finite State Machines, Hierarchical FSMs, Behaviour Trees.


• Simulating Intelligence: Sensory Systems, Animations, Goal-Oriented Action Planning. Procedural Content Generation for Games. Refresher information on Vectors, Probability and Calculus will be provided during this module. The course will be taught via lectures and extensive laboratory exercises, using Python as the main programming language. Note that students should either be strong programmers, or have previous Python programming experience, to take this module, as the Python programming will be fairly demanding as soon as the module starts.

This module will be delivered via:

• Eight 2-hour lectures.
• Ten 2-hour labs.
• Two 1-hour progress tests (in weeks 6 and 11).
• Weekly Moodle Quizzes.
• Additional video lectures.