Thèse de doctorat
Résumé : The swarm robotics research field addresses the challenge of controlling a large number of robots that act in a self-organized and decentralized way. These robot swarms have the potential to operate in highly complex and unknown environments. For years, robot swarms were thought to be inherently secure thanks to their redundant and decentralized design. This thesis challenges this assumption by providing evidence that even a single non-cooperating robot can completely disrupt the swarm's behavior in existing algorithms. We call these robots, which do not behave according to the specified protocol, Byzantine robots.The management of Byzantine agents in decentralized systems is addressed in another research field: blockchain technology. Blockchain technology, which was initially developed for the peer-to-peer currency Bitcoin, enables mutually distrusting agents to maintain a secure shared ledger. However, the concept of a blockchain is not limited to storing financial transactions: the Ethereum framework made it possible to deploy computer programs to a blockchain. These tamper-proof computer programs are called blockchain-based smart contracts and enable a decentralized network to reach an agreement on the outcome of their code.In this thesis, we investigate how blockchain technology can securely coordinate a robot swarm. In experiments with both physical robots and large-scale simulations, we demonstrate that the robots are able to maintain blockchain networks and that they can coordinate their behavior via blockchain-based smart contracts. In particular, we show that blockchain-based smart contracts can neutralize Byzantine robots in a fully decentralized way. We demonstrate how a blockchain can be used in robot swarms as a shared knowledge medium, computing platform, reputation management system, consensus agreement protocol, and economical platform.Experiments and simulations combined, this thesis lays the foundation for blockchain-based swarm robotics. This foundation can pave the way for a wide range of secure swarm robotics applications, such as task-allocation scenarios, collective mapping, and lightweight machine-learning algorithms.