CBL - Campus del Baix Llobregat

Projecte llegit

Títol: Gestión autónoma de un enjambre de drones de reparto


Estudiants que han llegit aquest projecte:


Director/a: VALERO GARCÍA, MIGUEL

Departament: DAC

Títol: Gestión autónoma de un enjambre de drones de reparto

Data inici oferta: 02-02-2026     Data finalització oferta: 02-10-2026



Estudis d'assignació del projecte:
    GR ENG SIST AEROESP
Tipus: Individual
 
Lloc de realització: EETAC
 
Paraules clau:
Drones de reparto, enjambre de drones, asignación de tareas, algoritmo de Jonker-Volgenant, gestión del espacio aéreo, ArduPilot, SITL, simulación
 
Descripció del contingut i pla d'activitats:
 
Overview (resum en anglès):
This project develops and validates a system that autonomously manages a heterogeneous swarm of delivery drones. It is part of the Drone Engineering Ecosystem of the EETAC and reuses the ecosystem's dronLink library to communicate with ArduPilot autopilots.

The main result is a web application with a portal for customers and a dashboard for the operator. The system receives orders, decides which drone serves each one, coordinates the swarm traffic and executes the flights without human intervention. The assignment cost is the energy of each trip, estimated with a model that is linear in the payload for three drone categories. Two assignment methods are available: a greedy one, which serves the queue in order of arrival, and a global one, which solves the minimum-energy assignment in every cycle with the Jonker-Volgenant algorithm. To keep simultaneous deliveries safe, the system separates traffic into altitude layers according to the flight direction, models the base as a vertiport with reserved pads, staggers take-offs and limits the number of drones at the common transit point. All settings are stored in editable scenarios, so the system can be moved to another location without changing the code.

A simulation study also compares which assignment policy serves customers best. When many orders must be delivered at once, the global method delivers considerably more than the greedy one and comes close to the exact optimum. Over full operating days with a saturated fleet it still delivers more orders on time, but at the expense of the most distant customers. A planner based on integer linear programming with demand forecasting is the only policy that improves both punctuality and the waiting time of the worst-served customers. The study also shows that the service is limited by the fleet composition before the algorithm.

The system has been validated at three levels: automated tests, complete waves over a simulated link and flights against ArduCopter autopilots in SITL. It is deployed in the cloud, where anyone can place an order from a web browser and follow the flight of a simulated autopilot. Four ways of connecting to a real drone have also been studied, including a 4G cellular modem on the onboard Raspberry Pi, tested on the bench.


© CBLTIC Campus del Baix Llobregat - UPC