Abstract

Settlers of Catan is a modern board game, where 3–4 players collect resources to build and trade with each other, in order to be the first to reach a set number of points. This game presents interesting difficulties for an AI, including randomness, hidden information, negotiation and trading. In this talk, I shall discuss the implementations presented to handle these considerations. We will begin with the AI implementation in the JSettlers game and the later improvements to this agent, and follow by reviewing the benefits of using Monte-Carlo Tree Search or a Reinforcement Learning approach in this context, ending with a closer look at the trading AI. By following the contributions by researchers towards a competent Settlers AI, we shall explore the concepts behind various AI algorithms and techniques, and the process of improving upon established work.