A multi-algorithm pathfinding method: Exploiting performance variations for enhanced efficiency

This paper presents a performance evaluation of several heuristic search algorithms in the context of pathfinding. Our objective is to assess the performance of these algorithms in various grid-based environments to present how specific domain features influence their efficiency. Additionally, we extend our experiments by incorporating Multi-Agent Path Finding (MAPF) benchmarks, using handcrafted features and features extracted with Convolutional Neural Network (CNN) to characterize the maps. The results of our evaluation were later used to train machine learning models capable of predicting the efficient algorithm for a given pathfinding task based on performance criteria. This multi-algorithm pathfinding method enhances the selection of the best algorithm for different pathfinding problems. Furthermore, we revealed the most important features that impact the selection of the efficient algorithm. We identify the most important characteristics of the grid that affect the selection and performance of the algorithms.