« Retourner aux résultats de recherche

• Citer

Swarm Oracle: Towards Trustless Self-Organization in Robot Swarms using Blockchain Technologies

par Melo Pacheco, Alexandre
Président du jury Birattari, Mauro
Promoteur Dorigo, Marco
Publication Non publié, 2026-04-24

Thèse de doctorat

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

The integration of blockchain technology with robotics offers a promising foundation for coordinating and securing decentralized multi-robot systems in real-world deployments. By maintaining a distributed record and a shared execution state, blockchains enable robots to enforce collective rules, synchronize data, log events, and detect inconsistencies in peer behavior without relying on centralized control.A fundamental limitation, however, hinders the practical deployment of blockchain-enabled swarms: the oracle problem. Blockchain consensus follows a trustless design in which all participants independently verify and execute state transitions to maintain a consistent global state. While this design allows blockchains to operate in open environments with untrusted participants, it inherently restricts them to deterministic computations.In contrast, robotic systems rely on local perception derived from noisy and uncertain sensor data, making it difficult to directly integrate such information into blockchain-enabled decision processes. Blockchain oracles supply external data to the system, but many designs rely on trusted parties or centralized components, undermining decentralization and fault-tolerance.This thesis introduces Swarm Oracle, a distributed oracle network composed of mobile robots that collectively validate uncertain environmental observations without assuming trust or cooperation among robots. Each robot submits locally perceived data, which is aggregated and processed through Byzantine fault-tolerant protocols executed on-chain. The consensus outcome can then be used to support the enactment of on-chain collective decisions. In this way, Swarm Oracle closes the loop between swarm-level perception and planning, supporting a new blockchain-enabled paradigm for swarm coordination.Within this paradigm, self-organization emerges not only from individual robot behaviors but also from globally validated information that drives swarm-wide action policies. This information can, for example, regulate resource allocation, update shared databases, trigger behavioral adaptations, and refine learned models. By distributing validation across independent robots, Swarm Oracle ensures reliable operation even when a significant portion of the swarm behaves maliciously. This trustless design makes it suitable for multi-stakeholder deployments, in which robots operated by different parties can collaborate without needing to trust one another.The system is evaluated on a collective perception task in which robots estimate environmental colors using biased and noisy sensors. Swarm Oracle achieves correct agreement even when up to one-third of the robots collude to manipulate consensus outcomes. Furthermore, a token-based reputation mechanism enables the system to progressively mitigate the influence of faulty or malicious robots. Through practical application scenarios—including collaborative mapping, decentralized neural network training, and collective foraging—this thesis demonstrates how Swarm Oracle mechanisms can be applied to support coordination and self-organization in realistic scenarios.Although these mechanisms can be deployed on existing blockchain infrastructure, this thesis focuses on the case where the blockchain is maintained directly by the robots themselves. The feasibility of this approach is assessed through experiments with up to 24 physical robots and 120 simulated robots. Results show that storage requirements scale linearly with swarm size and mission duration, while bandwidth, computation, and memory usage remain stable over extended operations.Overall, this thesis establishes Swarm Oracle as a principled blockchain-enabled mechanism for trustless coordination in robot swarms, providing a foundation for secure, scalable, and self-organizing multi-robot systems in realistic deployments.