DSPH: Distributed Swarm Navigation in Dynamic Complex Environments via Smoothed Particle Hydrodynamics

Abstract

UAV swarm navigation in dynamic complex environments must balance collective coherence and local collision safety. Existing formation-control and obstacle-avoidance methods are often developed independently, and their direct combination may produce conflicting commands, disrupt swarm structure, or induce overly conservative motion, especially when moving obstacles create rapidly changing collision risks. This paper presents a distributed navigation framework that integrates Smoothed Particle Hydrodynamics (SPH) with a Gradient Velocity Obstacle (GVO) method. The framework decouples inter-agent coordination from agent-obstacle interaction at both the perception and algorithmic levels. SPH models local interactions among neighboring UAVs to preserve cooperative motion and formation consistency, while GVO generates smooth, gradient-based responses to dynamic obstacles. Each UAV can therefore maintain swarm-level coordination while adapting independently to local collision threats. Extensive simulations and real-world flight experiments in cluttered, dynamic environments show that the proposed method outperforms representative baselines in navigation success rate, motion efficiency, and formation consistency. The results further demonstrate strong robustness and scalability, supporting practical deployment in UAV swarm navigation tasks. Note to Practitioners—This paper studies autonomous UAV swarm navigation in dynamic complex environments, focusing on balancing collision safety and swarm consistency under limited onboard computation, local perception, and constrained communication. The proposed framework only requires task commands, such as target positions, to achieve swarm self-organization, dynamic obstacle avoidance, and target navigation. Each UAV coordinates with neighboring agents via UDP and avoids dynamic obstacles using onboard perception, maintaining a high degree of individual independence. The method is suitable for parallel swarm tasks such as inspection, search and rescue, and cooperative monitoring. Simulations and real-world experiments show improved task success rate and swarm consistency in complex environments. For deployment, parameter validation and adjustment are recommended when platform scale, sensing range, or obstacle density changes significantly.