Thèse Essaims de Drones Résilients et Économes en Énergie pour la Surveillance des Feux de Forêt H/F Doctorat.Gouv.Fr

Établissement : Ecole Nationale de l'Aviation Civile École doctorale : SYSTEMES Laboratoire de recherche : ENAC-LAB - Laboratoire de Recherche ENAC Direction de la thèse : Guthemberg DA SILVA SILVESTRE ORCID 0000000202737212 Début de la thèse : 2027-09-01 Date limite de candidature : 2026-11-23T23:59:59 Le changement climatique accroît la fréquence et la gravité des conditions propices aux feux de forêt à travers l'Europe, notamment les vagues de chaleur extrême, la sécheresse et les périodes prolongées de risque élevé d'incendie. Par conséquent, une surveillance et une télédétection rapides et fiables sont nécessaires pour permettre une intervention d'urgence efficace. Les drones équipés de capteurs thermiques, visuels et environnementaux constituent une solution prometteuse, car ils peuvent accéder rapidement à des zones dangereuses ou difficiles d'accès. La surveillance des feux de forêt étant une activité critique pour la sécurité, il ne suffit pas de s'appuyer sur un seul drone. Au contraire, un essaim de drones, composé de plusieurs drones coordonnant leurs actions, peut offrir une tolérance aux pannes et permettre à la surveillance de se poursuivre même en cas de défaillance de certains drones. Ce projet explore comment ces technologies émergentes peuvent améliorer la fiabilité et la résilience de la surveillance des feux de forêt. The increasing frequency and severity of wildfires require timely and reliable monitoring to support effective emergency response. UAVs equipped with thermal, visual, and environmental sensors provide a promising solution, as they can rapidly access dangerous or difficult-to-reach areas. Since wildfire monitoring is safety-critical, relying on a single UAV is insufficient. Instead, a UAV swarm, consisting of multiple UAVs that coordinate their actions, can provide fault tolerance and allow monitoring to continue even if some UAVs fail.

Using a UAV swarm, however, raises the challenge of how to coordinate the placement and sensing activities of its members so as to maximize the monitored area while preserving the required level of fault tolerance. Recent research has explored UAV swarms for wildfire monitoring, including decentralized monitoring, fault-tolerant navigation, and coverage planning [PDM+ 24, HNC+ 22, SCB24]. However, these works mainly focus on specific algorithms and operational settings, while the fundamental relationship between fault tolerance and the amount of area that can be reliably monitored remains insufficiently characterized.

This relationship is influenced by the communication requirements of the swarm. UAVs can communicate directly only when they are within communication range, forming a communication network whose structure depends on their positions. To remain connected despite up to f UAV failures, this network must provide sufficient redundant communication paths. Stronger fault-tolerance requirements therefore constrain how far the UAVs can spread apart, which in turn limits the area that can be monitored. This leads to our first research question:

RQ1. Given a swarm of n UAVs and a requirement to tolerate up to f UAV failures, what is the maximum area that can be monitored while preserving the required level of fault
tolerance?
A second limitation of UAV swarms is their finite energy. Achieving fault tolerance requires sufficient redundancy so that monitoring can continue after failures. Maintaining this redundancy may require additional sensing, communication, movement, or standby capacity, all of which consume energy and reduce the operational lifetime of the swarm. Recent work has considered battery levels together with failures when assigning tasks to UAVs [NA26]. However, the fundamental energy cost of providing a given level of fault-tolerant monitoring remains insufficiently characterized. This leads
to our second research question:

RQ2. What is the minimum energy required to achieve a given level of fault-tolerant monitoring, and how can the operational lifetime of the swarm be maximized?
Together, these questions capture the relationship between the amount of area that can be monitored, the level of fault tolerance that can be guaranteed, and the lifetime of the UAV swarm. The project has three main scientific goals:

1. Characterize the fundamental trade-off between fault tolerance and the maximum area that can
be monitored by a UAV swarm.
2. Characterize the minimum energy required to achieve a given level of fault-tolerant monitoring
and determine its impact on the operational lifetime of the swarm.
3. Design distributed algorithms that coordinate UAV placement and redundancy while approaching
these fundamental limits.

To address these goals, we will first develop formal models capturing UAV positions, communication range, failures, monitored area, and energy consumption. A fault-tolerant distributed swarm must maintain sufficient communication redundancy so that information can still propagate through the system even when some UAVs fail. At the same time, each UAV should preferably rely only on local information, since maintaining complete knowledge of a large and dynamic swarm may be costly and impractical.

We propose to capture these requirements using the notion of k-One Sink Reducibility (k-OSR). At a high level, k-OSR describes a communication structure in which nodes (UAVs) can reach a sufficiently connected part of the network through multiple independent paths, allowing distributed tasks to remain solvable despite failures and limited knowledge of the overall network. In this setting, each UAV knows and communicates only with a subset of other UAVs, rather than maintaining complete knowledge of the entire swarm. The notion was originally introduced in [GT07], and more recent work has extended and generalized these requirements to broader fault models and weaker assumptions about network knowledge [HVB24].

Building on this theoretical analysis, we will design distributed algorithms that allow the UAVs to determine their placement using local information and to adapt their configuration when failures occur or energy levels change. We will also investigate strategies for managing redundancy while minimizing energy consumption and maximizing the operational lifetime of the UAV swarm.

The proposed algorithms will first be evaluated through simulation using realistic wildfire scenarios and models of UAV mobility, communication, failures, and energy consumption. We also plan to implement and experimentally evaluate the resulting algorithms using physical UAVs. For this purpose, we plan to use the UAV experimentation facilities available at ENAC, which provides a controlled environment for UAV experiments and accurate tracking of UAV movements.

In summary, the expected contribution is a theoretical and practical framework for characterizing the relationship between fault tolerance, monitored area, and energy consumption in UAV swarms, together with distributed algorithms that approach these limits in realistic wildfire-monitoring missions. The PhD project will be organized into four work packages covering the theoretical, algorithmic, and experimental aspects of the research.

WP1 - Modelling and theoretical foundations (Months 1-8). The student will first study the state of the art on fault-tolerant distributed systems, UAV swarms, coverage optimization, and energy-aware coordination. Based on this study, the student will formalize the system model, including UAV positions, communication range, failures, monitored area, local knowledge, and energy constraints. Particular attention will be given to adapting k-OSR and related knowledge-connectivity concepts to geometric UAV networks. The main objective of this work package is to establish a formal relationship between the communication structure required for fault tolerance and the spatial organization of the UAV swarm.

*Expected outcome*: formal system model, precise problem definitions, and initial theoretical results on fault-tolerant coverage.

*Expected outcome*: formal system model, precise problem definitions, and initial theoretical results on fault-tolerant coverage.

*WP2* - Fault tolerance and monitored area (Months 6-18). This work package will address RQ1. The student will investigate lower and upper bounds on the maximum area that can be monitored while guaranteeing a given level of fault tolerance. Based on these results, distributed placement algorithms will be designed in which each UAV relies only on local information about the swarm. Theoretical analysis will establish the correctness and fault-tolerance properties of the proposed algorithms and quantify how closely they approach the fundamental coverage bounds.

*Expected outcome*: theoretical characterization of the fault-tolerance-coverage trade-off and distributed algorithms for fault-tolerant UAV placement.

*WP3* - Energy-aware fault tolerance and mission lifetime (Months 16-28). This work package will address RQ2. The system model will be extended to include the energy consumed by UAV movement, sensing, communication, and redundancy. The student will investigate the minimum energy required to maintain different levels of fault-tolerant monitoring and how this requirement affects the operational lifetime of the swarm.
Based on this analysis, energy-aware distributed algorithms will be developed to adapt UAV place-
ment and redundancy according to remaining energy and observed failures.

*Expected outcome*: theoretical bounds on the energy cost of fault tolerance and distributed algorithms for extending mission lifetime while maintaining monitoring and resilience guarantees.

*WP4* - Implementation and experimental validation (Months 24-36). The algorithms developed in WP2 and WP3 will first be evaluated through simulation using realistic wildfire scenarios and models of UAV mobility, communication, failures, and energy consumption. The most promising algorithms will subsequently be implemented and evaluated using physical UAVs.
The student will benefit from the UAV experimentation facilities available at ENAC, which provide a controlled environment for testing UAV coordination and accurately tracking UAV movements. This stage will allow the theoretical assumptions and guarantees developed during the PhD to be evaluated under realistic communication, mobility, and energy conditions.

*Expected outcome*: an integrated prototype, experimental evaluation of the proposed algorithms, and a comparison between theoretical predictions, simulation results, and physical experiments.

Throughout the PhD, the student will receive interdisciplinary training in distributed computing, fault tolerance, UAV systems, and experimental research. The results of the different work packages are expected to lead to publications in international conferences and journals in distributed systems, autonomous systems, and UAV research.

The PhD thesis will integrate the theoretical and experimental results into a unified framework for understanding and optimizing the trade-off between fault tolerance, monitored area, energy consumption, and mission lifetime in UAV swarms.