Collaborative Agent Gameplay in the Pandemic Board Game

The Rolling Horizon Evolutionary Algorithm (RHEA) provides a robust framework for navigating the high stochasticity and complex decision spaces inherent in the collaborative board game Pandemic. By employing macro-actions—sequences of moves designed to achieve specific objectives like disease treatment or cure discovery—the system effectively manages the game’s significant branching factor. This approach utilizes a forward model to evaluate potential game states, balancing optimistic pursuit of victory with pessimistic risk mitigation to ensure long-term survival against the game’s escalating threats.

Experimental results confirm that this evolutionary approach substantially outperforms baseline default policies, yielding an average win rate improvement of 302%. The algorithm demonstrates a superior ability to prioritize critical actions, such as sharing knowledge between players, which is essential for success in the game’s cooperative environment. By integrating a default policy as both a performance benchmark and a mechanism for strategy repair, the agent maintains consistency even when faced with the high levels of randomness characteristic of the Pandemic board game.

While the current implementation proves highly effective across various difficulty levels, its performance remains sensitive to extreme randomization and complex game states. Future advancements are required to incorporate nuanced elements such as player-specific inventories and special event cards to further enhance strategic depth. Ultimately, this research establishes a viable foundation for autonomous agents in cooperative gaming, though further development is necessary to facilitate seamless human-AI collaboration and address more intricate board game mechanics.