Game AI Pro 3

28

Pitfalls and Solutions When Using Monte Carlo Tree Search for Strategy and Tactical Games

Gijs-Jan Roelofs

28.1    Introduction

28.2    The Problem

28.3    Expansion

28.4    Simulation

28.5    Backpropagation

28.6    Selection

28.7    Conclusion

References

28.1  Introduction

The biggest challenge when building AI for modern strategy games is dealing with their complexity. Utility- or expert-based techniques provide good bang for their buck in the sense that they get the AI to a reasonable, even decent level of play quickly. Problems start to arise, however, when an AI programmer is faced with situations that are not straightforward or too time-consuming to encode in heuristics. Or when he or she is faced with above average gamers who demand ...