« Retourner aux résultats de recherche
par Oguz, Sinan 
Président du jury Birattari, Mauro
Promoteur Dorigo, Marco
Co-Promoteur Garone, Emanuele
Publication Non publié, 2025-12-15
Président du jury Birattari, Mauro
Promoteur Dorigo, Marco
Co-Promoteur Garone, Emanuele
Publication Non publié, 2025-12-15
Thèse de doctorat
| Résumé : | In recent years, swarms of aerial robots have attracted growing attention in both scientific research and practical applications, driven by their potential to collectively perform tasks that would otherwise overwhelm the limitations of individual robots. Unlike a single unit acting in isolation, a swarm leverages coordination—each robot makes decisions based on simple rules and local perceptions, yet together they can generate complex, coordinated behaviors. This emergent cooperation enables the group to tackle diverse tasks with adaptability and resilience. While robot swarms demonstrate remarkable potential for collective intelligence, reliably translating this promise into practice presents sizable technical challenges. For aerial swarms specifically, these challenges include: (i) the scarcity of open, swarm-ready UAV platforms that can support research into new aerial swarm behav- iors in laboratory environments; (ii) the di"culty of developing self-organized UAV coordination algorithms that are both provably reliable—with formal guarantees for stability, convergence, and robustness—and practical to deploy under real-world con- straints such as limited computation, communication, and energy; and (iii) a lack of existing self-organized approaches to handle intermittent sensing and communication disturbances in aerial swarms—including temporally correlated dropouts, delays, and noise—that, while initially non-catastrophic, can cumulatively degrade coordination quality and system-level performance and may eventually manifest as permanent faults, thereby compromising long-term swarm reliability. This thesis advances resilient coordination in aerial robot swarms through contri- butions spanning hardware, theory, and algorithms. The first contribution is the S- drone, a fully open-source UAV platform and simulation stack that supports onboard sensing and cooperative inter-robot tracking, enabling demonstrations of swarm be- haviors (e.g., formation flight, cooperative transport, wide-area coverage) without external infrastructure. The second is the Self-organizing Nervous System (SoNS), which contributes self-organizing hierarchy for robot swarms. Using the SoNS, robots self-organize into temporary locally centralized networks that fuse information and coordinate actions; these networks can split or merge and any robot acting as a temporary leader can be replaced on the fly, allowing the swarm to reconfigure as conditions change, while maintaining scalability and fault tolerance against the loss of arbitrary robots. The third is Hierarchical Henneberg Construction (HHC), which contributes a graph-theoretic formalization of the SoNS for aerial swarms, based on bearing (angle of arrival). The HHC algorithms yield analyzable procedures to con- struct, split, merge, and rebuild self-organized frameworks while preserving rigidity and hierarchy. Finally, the fourth contribution is a proactive–reactive fault-tolerance strategy for intermittent faults in aerial swarms, based on the introduction of Adap- tive Biased Minimum Consensus (ABMC). The ABMC is a distributed protocol that constructs and maintains low-cost backup paths to leaders in self-organized hierar- chical swarms. The paths constructed using ABMC support one-shot likelihood-ratio tests to detect faulty information originating in a portion of the swarm, then sup- ply the backup paths that enable quick rerouting to sustain information flow and stable formation control. Together, these contributions provide an open experimental platform, a reconfigurable hierarchical control paradigm, and provable resilience tools—bridging the gap between theory and practice in resilient coordination in aerial robot swarms. |
