ETD

• coordination
• decentralized control
• drone
• flocking
• parameter optimization
• performance evaluation
• PX4
• ROS2
• search and rescue
• simulation
• stigmergy
• swarm Intelligence
• UAV

This thesis work is part of the 4DDS (Four-Domain Drone System) project, an advanced platform designed for the coordination of heterogeneous autonomous vehicles operating across four distinct domains: aerial, terrestrial, water surface, and underwater. Specifically, the research focuses on the problem of decentralized swarm coordination, proposing a control architecture that synergistically combines two classic bio-inspired mechanisms: Reynolds' kinematic flocking rules and indirect communication via stigmergy. The first phase of the work focused on reviewing and validating the individual algorithms, to align them with established models in the scientific literature while simultaneously ensuring greater adherence to the real flight dynamics of a drone. Subsequently, for the integration of the two mechanisms, a control architecture was developed that implements a priority logic alongside a context-dependent weight-switching configurations.
From an architectural standpoint, the system is developed on the PX4, ROS2, and Gazebo technology stack, whose components interact synergistically. ROS2 acts as the decentralized communication infrastructure managing the data exchange between software modules, while Gazebo operates as the multi-domain 3D simulator, reproducing the environmental physics and injecting virtual sensor data into the ROS2 topics. Meanwhile, the PX4 Autopilot manages the low-level flight control by interfacing with ROS2 to receive movement commands and publish telemetry. At the application level, the swarm coordination is structured around three main ROS2 nodes: the BaseStation, responsible for processing the kinematic flocking interactions for the entire fleet; the Pheromone node, dedicated to the mapping and management of the virtual pheromone field; and the Drone node, which processes inputs locally on the individual vehicle, applying its own autonomous movement logic.
To evaluate the proposed architecture, an extensive experimental validation was conducted across four scenarios characterized by profoundly different topologies and environmental conditions. The simulations included Amatrice, representing a portion of the urban area hit by the 2016 earthquake; Posidonia, a submarine region featuring oceanic Posidonia and sand dunes; Wildfire, a large forest with a central mountain and active fires; and Illegal Dumps, a rural area with waste abandoned among the trees. The simulations were conducted by varying the fleet size, using 8 and 16 drones, and modulating the key parameters of flocking and stigmergy.
The methodology involved the execution of a preliminary grid search, followed by a detailed statistical analysis to isolate and evaluate the optimal configurations. Finally, the performance of the proposed architecture was compared with a deterministic squadron-based system utilizing pre-planned trajectories, highlighting the operational advantages and the limitations of the swarm approach for each scenario explored.