Explaining Synthesized Pathfinding Heuristics via Iterative Visualization and Modification

Heuristic search is widely used for game pathfinding with heuristic functions substantially influencing its pathfinding performance. Recent work used program synthesis to automatically generate high-performance formula-based heuristics. Their compactness and human readability offered a promise of explainability. In this paper we present an automated approach to decompose and visualize formula-based heuristics. To illustrate the explanatory power of the visualization we include it in a human-in-the-loop process to iteratively modify heuristic formulae and improve their search performance. The iterative process is meant to encourage human experimentation with the formula-based heuristics thereby increasing the understanding and trust of a game-AI developer or a heuristic search researcher.