Grand strategy games such as the Total War series create a very complex execution environment with a large number of objects. This places significant time and memory restrictions on the game AI. The decisions that the AI needs to make are highly dependent upon the current state of the world, requiring a deep and careful analysis of these objects.
The resulting search space is huge. It needs to be investigated with limited memory and time. Therefore, we need to be very careful about what algorithms we select to tackle this space, and we need to incorporate profiling and optimization techniques to ensure the game AI is sufficiently performant.
In this talk we’ll discuss the memory and time constraints of the campaign side of the Total War series, how we approach them and the AI techniques that we employ to perform under these limitations.
We will explain how we use Monte Carlo Tree Search with restricted tree building, branching and pruning for our game AI.
We will also give insights into the profiling techniques we use and how they contribute to increasing the performance of the game AI.