Automatic State Abstraction for Pathfinding in Real-Time Video Games

Real-time video games are a unique domain for pathfinding and search. Traditional approaches to search have usually assumed static worlds with a single agent. But, in real-time video games there are multiple cooperative and adversarial agents. While the search space in most games is relatively small, algorithms are expected to plan in mere milliseconds. Thus, techniques such as abstraction are needed to effectively reason and act in these worlds. We provide an overview of the research we have completed in this area, as well as areas of current and future work.