Gameplay-driven terrain generation in Scorched Earth

In this paper, we propose a hybrid method to procedurally generate terrains in Scorch Earth. It features both the fast computing speed of random techniques and the wide variety of details of fractal techniques. We indirectly evaluate terrains by analyzing the gameplay data. We create an AI agent to model players' behaviors to produce gameplay data. The agent is capable to shoot and move in different scenarios. Finally we introduce metrics for gameplays to select good terrains.