It’s a new year and a new semester, and Team Tactica is ready to get to work! Now that our plan is in place, we will spend this semester working directly on our project. Over the next sixteen weeks, we’ll hopefully have an AI model that demonstrates strategic decision-making in Catan.

We’ve decided to use a simulation different from the one we used for our Prototype Inspection Day (PID). Unlike the previous simulation, this one integrates well with a Python library called Gymnasium, which provides a consistent API for reinforcement learning. Gymnasium is a maintained fork of OpenAI’s Gym library. We can customize the environment, including the number of players, the reward function, the action space, etc. This will allow us to train our models more rapidly. Once our models are trained, we can export the models and reuse them in the testing simulation to see how well they perform.

We plan to switch to a more agile workflow in the next few weeks. We’ll start by creating a backlog of action items to accomplish. This includes modifying our simulation to fit our needs, building our reinforcement-learning policy, and coming up with several reward functions to test. We also need a way to test rapidly since there are a lot of variables we can adjust, including the size of the neural network, the algorithm used in the model, the reward functions, etc.

That’s all for now! See you next week!