Dynamic pathfinding for a swarm intelligence based UAV control model using particle swarm optimisation

Craig R. Stark*, Lewis M Pyke

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

1 Downloads (Pure)

Abstract

In recent years unmanned aerial vehicles (UAVs) have become smaller, cheaper, and more efficient, enabling the use of multiple autonomous drones where previously a single, human-operated drone would have been used. This likely includes crisis response and search and rescue missions. These systems will need a method of navigating unknown and dynamic environments. Typically, this would require an incremental heuristic search algorithm, however, these algorithms become increasingly computationally and memory intensive as the environment size increases. This paper used two different Swarm Intelligence (SI) algorithms: Particle Swarm Optimisation and Reynolds flocking to propose an overall system for controlling and navigating groups of autonomous drones through unknown and dynamic environments. This paper proposes Particle Swarm Optimisation Pathfinding (PSOP): a dynamic, cooperative algorithm; and, Drone Flock Control (DFC): a modular model for controlling systems of agents, in 3D environments, such that collisions are minimised. Using the Unity game engine, a realtime application, simulation environment, and data collection apparatus were developed and the performances of DFC-controlled drones—navigating with either the PSOP algorithm or a D* Lite implementation—were compared. The simulations do not consider UAV dynamics. The drones were tasked with navigating to a given target position in environments of varying size and quantitative data on pathfinding performance, computational and memory performance, and usability were collected. Using this data, the advantages of PSO-based pathfinding were demonstrated. PSOP was shown to be more memory efficient, more successful in the creation of high quality, accurate paths, more usable and as computationally efficient as a typical incremental heuristic search algorithm when used as part of a SI-based drone control model. This study demonstrated the capabilities of SI approaches as a means of controlling multi-agent UAV systems in a simple simulation environment. Future research may look to apply the DFC model, with the PSOP algorithm, to more advanced simulations which considered environment factors like atmospheric pressure and turbulence, or to real-world UAVs in a controlled environment.
Original languageEnglish
Article number744955
Number of pages16
JournalFrontiers in Applied Mathematics and Statistics
Volume7
Early online date22 Nov 2021
DOIs
Publication statusPublished - 22 Nov 2021

Keywords

  • Dynamic pathfinding
  • Swarm intelligence
  • Particle swarm optimisation
  • Flocking
  • Multi-agent systems and autonomous agents

Cite this