Pathfinding for Autonomous Drones

Pathfinding research conducted at the Governor’s School of New Jersey in collaboration with Lockheed Martin, comparing A* and RRT* across four classes of maze environments designed to mirror real-world scenarios: single-solution, high-density, low-density, and ladder. Each algorithm was evaluated on four metrics that matter for autonomous flight: flight time, flight distance, compute time, and path error.

The results were not symmetric. A* was faster to compute across every maze type, but RRT* produced shorter and faster flight paths in three of the four (low-density, high-density, and ladder), with the two performing comparably in single-solution mazes. The takeaway is that algorithm choice depends on what you’re optimizing for at the mission level, not just which is “better” in the abstract. The paper was published in IEEE Proceedings and presented at MIT’s Undergraduate Research Technology Conference.