Partial Synthesis of Heuristic Search Algorithms
Gallivan, Matthew · ERA: Education and Research Archive (University of Alberta) · 2021
Heuristic search is a core area of Artificial Intelligence (AI) with numerous applications. In video games it is commonly used to calculate paths of AI-controlled agents. Traditionally, heuristic search algorithms have been designed by humans. Recent work attempted to synthesise heuristic search algorithms by automatically combining elements of published algorithms. In doing so, researchers defined a synthesis space for heuristic search algorithms and then automatically searched through that space. We extend this line of work and make the following contributions. First, we define a richer space of algorithms using a finer set of building blocks. This space is constructed using a context-free grammar. We then show that in the new space we can automatically synthesise higher performing real-time heuristic search algorithms. We evaluate these algorithms over benchmark pathfinding problems taken from video games and show that our synthesis method outperforms existing work.