Path planning-based drone delivery system

Abstract

The rapid expansion of autonomous drone delivery systems necessitates the development of highly efficient and reliable path planning methodologies. Ensuring flight trajectories are collision-free while minimizing distance and preserving vehicle maneuverability is a critical challenge. This Bachelor Thesis addresses this challenge by proposing and evaluating an enhanced Particle Swarm Optimization (PSO) algorithm for 3D drone path planning. To improve search efficiency and avoid local minima, we implement an adaptive, linearly decreasing inertia weight that balances global exploration and local exploitation. Additionally, a robust penalty-based constraint handling mechanism is integrated for collision avoidance, and Catmull-Rom splines are employed as a post-processing step to smooth discrete waypoint paths into continuous trajectories. The enhanced PSO is systematically compared against a constant-weight baseline PSO across three environments of varying obstacle density (sparse, moderate, and dense). Performance is evaluated using four quantitative metrics: total path length, path smoothness (turn angle sum), collision rate (feasibility), and convergence iteration. Empirical results show that the enhanced PSO consistently generates shorter and smoother paths, reducing mean path length by up to 9% and total curvature by up to 34% in simple scenarios, while improving feasibility from 50% to 90% in dense obstacle fields. The modular Python-based simulation framework developed in this work offers a reproducible methodology for validating and optimizing drone routing in next-generation logistics.