Implementing efficient joint beliefs on multi-robot teams

In this report, I investigate the problem of coordinating multiple robots in a cooperative team activity. Many different approaches have been suggested for team coordination, ranging from swarm techniques to symbolic methods. I will propose a class of coordination protocols called broadcast-an-aggregate mechanisms and will describe a specific instantiation of this approach on a multi-robot control architecture called HIVEMind (Highly Interconnected Very Efficient Mind) which operates in the following way: (1) Each robot periodically broadcasts all its team-relevant data to its teammates. (2) Team members base control decisions on shared situational awareness formed by aggregating all broadcast data from the team.; HIVEMind allows near-instantaneous sharing of data between all team members. This lets the team respond in real-time to contingencies sensed by individual robots. Furthermore, HIVEMind robots can use role-passing inference to support limited real-time symbolic reasoning over synchronized knowledgebases. This allows the use of more structured representations than conventional behavior-based systems, which in turn allows the team to be tasked and monitored in natural ways. All this is accomplished while using surprisingly little bandwidth. I will show that given the assumptions: (1) Relevant aspects of the environment change relatively quickly. (2) All team members must be informed of these changes in bounded time. (3) All team members must be able to detect if they are failing to receiving data in a timely manner from their teammates. The HIVEMind data sharing method is optimal in the sense that any communication scheme would have to transmit at least as many packets as it would. I also will describe an implementation of HIVEMind used to coordinate a team of robots in variety of tasks involving systematic search of physical space.