civrealm

CivRealm is an interactive environment designed for the open-source strategy game Freeciv-web, providing a platform for decision-making agents to engage in complex strategic gameplay. It supports two primary types of agents: tensor-based reinforcement learning agents compatible with the Gymnasium API and language-based agents powered by advanced language models. CivRealm allows these agents to navigate game mechanics, make decisions, and learn from their interactions in a dynamic environment inspired by the Civilization series.

The environment includes a robust API that facilitates the development of custom agents, enabling functionalities such as selecting valid actions based on the game’s current state. For instance, an agent can choose the first valid actor to act within a turn and generate actions like building a city or moving, enhancing its strategic capabilities. CivRealm offers comprehensive interfaces for experimentation, allowing researchers and developers to explore various decision-making strategies and reinforce learning through gameplay scenarios.