The Pathfinding Benchmark (Extended Abstract)

Abstract

Pathfinding problems are found throughout chemistry, mathematics, robotics, and computing. We present the Pathfinding Benchmark (PB), a work-in-progress benchmark of diverse and challenging pathfinding problems to evaluate the ability of state-of-the-art pathfinding algorithms and inspire new pathfinding algorithms. PB contains pathfinding problems from chemical reaction mechanism pathfinding, quantum circuit synthesis, theorem proving, multi-agent pathfinding, and combinatorial puzzles. Furthermore, PB pushes the boundaries of what should be posed as a pathfinding problem by posing language modeling, image generation, and training neural networks as pathfinding problems.