Mixed-Criticality Swarm Robotics Through Real-Time Wireless Networks

Communication will be a vital part of real-world applications of swarm robotics. Prior work has shown the existing swarm robotics prototypes are often designed with unrealistic assumptions on network performance and are therefore prone to failure under real world network conditions. This thesis argues for an introduction of a mixed criticality approach from the real-time systems domain to swarm robotics systems, beginning at the network layer. Mixed criticality communication protocols allow the system designer to reason a priori about the network behaviour and provide graceful degradation in the presence of network faults. This more predictable network behaviour is shown to translate into more predictable robot behaviour at the application level.

Since robot behaviour can influence the network conditions, for example due to robot mobility, the application and network components of a swarm robotics system are mutually dependent. Redefining criticality modes from an internal parameter of the network to an interface between the application and network layers allows the application component to adapt its own behaviour in response to the network conditions. This is shown to result in an overall system being better able to adapt to changing conditions. Raising the concept of criticality further towards the application component of a swarm robotics systems through a proposed mixed criticality wireless communication protocol that make criticality mode more meaningful to an application component make it more feasible to provide application level behaviour guarantees, while avoiding the centralised control of other protocols that is antithetical to swarm robotics systems.