Decentralized Cooperative Search

Abstract—We describe a mechanism for decentralizedcooperative control of a robot team for a geographic searchtask. The approach is behavior-based: robots are attractedto discrete probability distributions in the form of virtualparticles of probability that represent hypotheses about theexistence and identity of objects of interest in a geographicarea. Reliance on dependable, high-bandwidth communi-cation is reduced by modeling the movements of otherteam members and the objects of interest between periodicupdate messages. We combine ideas from multi-hypothesistracking, behavior-based control, and decentralized taskallocation.Index Terms—Cooperative systems, distributed control,distributed decision-making, resource management. I. I NTRODUCTION /M OTIVATION Robots are increasingly employed for dull, dirtyand dangerous tasks. Agencies are starting to lookfor ways to control teams of two to twenty robots oncooperative missions, with minimal required humansupervision (e.g., [3]). Teams of robots can be usedunderwater to comb bays for mines and to monitorthe ocean chemistry for changes. Future CombatSystems (FCS) plans to use them on land and inthe air to fight battles. Decentralized Control ofAutoma-Teams (DCAT), Mixed-Initiative Controlof Automa-teams (MICA), Unmanned Combat AirRotorcraft (UCAR) and Multisensor Command andControl Constellation (MC2C) are recent air pro-grams that require control of multiple vehicles onjoint missions.The problem is to design a scalable, robust con-trol mechanism that will allocate tasks, deconflictmotion vectors, and make decisions based on joint