IX Congreso Nacional de I+D en Defensa y Seguridad - DESEi+d 2022



News and Events 

IX Congreso Nacional de I+D en Defensa y Seguridad - DESEi+d 2022

The IX National Congress of R&D in Defence and Security (DESEi+d 2022) took place at the General Morillo Base of the Spanish Army, located in Pontevedra, on 15, 16 and 17 November 2022.

In line with previous editions, this Congress was presented as a forum and meeting point for all the agents related to R&D in the field of Defence and Security, where there was the opportunity to present and disseminate the results of the latest research and work carried out in some of the thematic areas related to Defence and Security.

Daniel Fuertes, PhD student at GTI, attended the DESEi+d 2022 where he presented his work entitled "Multi-drone route planning using Transformer deep neural networks".


Poster DESEID2022


Abstract: One of the most critical stages of drone (UAV/RPAS) control and navigation is route planning. Today, this task is essential in search and rescue applications or in the Future Combat Air System, where a drone needs to plan a route that minimises the flight distance between a set of locations in order to maximise the number of areas visited, all subject to operational constraints, such as battery or fuel usage. This already complex task is further complicated in the case of multiple drones, where the cooperation and coordination of an entire fleet is necessary. In this paper we propose an automatic route planning system for multiple drones using deep learning and reinforcement learning techniques. The system divides the routing problem into two phases: Initial Planning and Mission Execution. During Initial Planning, a clustering of the regions to be visited is performed following a distance criterion, followed by an assignment of these clusters to each drone. In the mission execution phase, the best route for each agent is estimated using a Transformer, a state-of-the-art neural network architecture based on attention mechanisms, which is trained using deep reinforcement learning techniques. This architecture is able to obtain accurate and much faster solutions than conventional optimisation algorithms. To show the benefits of the proposed solution, several tests and comparisons with other Combinatorial Optimisation algorithms, including cooperative and non-cooperative scenarios, have been performed. More info