Towards Case-Based Reasoning with k-d Trees for a Computer Game of Soccer

Real soccer tracking data can potentially be used for building a learning by observation-based game AI system, possessing human-like decision making traits. Such a system should be able to find matching example cases for the current game situation and act accordingly. This can be a challenging issue due to high dimensionality of soccer data and the requirements to detect approximate matches. We show how a simple k-d tree-based search can accomplish this task with modest space and time requirements, making it a feasible approach for a practical game system.