SOUL: data sharing for robot swarms

Interconnected devices and mobile multi-robot systems are increasingly present in many real-life scenarios. To be effective, these systems need to collect large amounts of data from their environment, and often these data need to be aggregated, shared, and distributed. Many multi-robot systems are designed to share state information and commands, but their communication infrastructure is often too limited for significant data transfers. This paper introduces Swarm-Oriented Upload of Labeled data, a mechanism that allows members of a fully distributed system to share data with their peers. We leverage a BitTorrent-like strategy to share data in smaller chunks, or datagrams, with policies that minimize reconstruction time. We performed extensive simulations to study the properties of the system and to demonstrate its scalability. We report experiments conducted with real robots following two realistic deployment scenarios: searching for objects in a scene, and replacing the full identity of a defective robot.