Distributed control of multi-agent systems over wireless networks, such as robot swarms and wireless sensor networks, provide modularity and robustness to single point of failure over classic centralized controller, but open the door to cyberattacks through increased attack surface. An adversary may introduce the system from any point in the networks, exploiting inter-agent couplings through network interconnections to disrupt a global task. For example, cooperative coverage with a multi-robot swarm may be disrupted by spoofing fake robots at a certain region in space, which is a standard attack strategy. At the same time, standard cybersecurity mechanisms, such as encryption, are ill-suited to real-time control due to their communication and computational overhead.

A promising way to overcome this issue and recover performance with normally behaving robots is to implement a soft, lightweight detection based on physical layer security. In words, physical features of the wireless channel are processed on-the-fly to infer whether the sender can be trusted. This approach has been studied for single links, but how to effectively use this approach for multi-agent control is still underexplored. In this thesis, you will develop and implement algorithms of physical layer security and authentication for multi-agent tasks, with relevant examples including multi-robot coverage and distributed learning; however, you are welcome to propose a different application.

This thesis may involve a (possibly remote) collaboration with Harvard University. Please ask the supervisor for further information and potential funding.