Using complex network effects for communication decisions in large multi-robot teams

Sharing information is critical to multi-robot team coordination when robots are widely deployed in a dynamic and partially observable environment. To be efficient, robots should balance well between broadcasting information and reserving limited bandwidth so that only the right information should be broadcast to the interested receivers. Robots' communication decision is normally modeled as a multi-agent decision theoretical problem. However, when the team expands to very large, the solution is classified as NEXP-COMPLETE. In this paper, in addition to building heuristic approaches to solve the decision theoretical problem based on the information context to be broadcast, we put forward a novel context-free decision model that allows fast communication decision by considering complex network attributes in large teams. Similar to human society, information should be broadcast if the action can make a good information coverage in the team. We analyze how complex network attributes can improve communication in a broadcast network. By putting forward a heuristic model to estimate those complex network attributes from robots' local view, we can build decision models either from robots' experiences or from their local incoming communications. Finally, we incorporate our algorithm in well-known information sharing algorithms and the results manifest the feasibility of our design.