Task and Role Allocation Within Multi-Agent and Robotics Research

In the past couple of decades, multi-agent researchers and roboticists have focused on designing systems containing multiple, autonomous agents that work together to accomplish a common objective. In natural systems, we have seen how groups of animals can solve problems that could not be solved by solitary individuals [1, 27]. Because of the emergent behaviors within multi-agent systems, the whole is sometimes greater than the sum of their parts. Roboticists are interested in multi-robot cooperation because if intelligent teamwork can be seen amongst a group of robots, then in the future they can probably cooperate with humans also [7]. Also, because of the high level of redundancy and use of decentralized control, they are without a single point of failure and, thus, robust to complete crashes occurring when one part of the system breaks [4, 17, 24, 28, 29, 30]. Unfortunately, the benefits of using multiple agents to solve problems do not come free. Researchers focused on multi-robot systems have found that robot-robot interference can cause task completion performance to decrease as more robots are added to the team [9, 14]. Along with the physical limitations, computational problems also arise in multi-agent and multi-robot systems. When searching through the multi-agent literature, four main problems are found. Two of the problems, role allocation and task allocation, are nearly exclusive to multi-agent systems, while the other two, task decomposition and action selection, are related to both multi and single-agent systems. Task decomposition is a divideand-conquer approach to problem solving, as it involves breaking up large tasks into smaller, manageable subtasks. Task allocation is the problem of optimally assigning these subtasks to agents. Action selection involves determining which low level actions to take in order to complete an assigned task. Related to task allocation is the higher level problem of role allocation. Typically, roles define which tasks an agent should complete, and tasks influence the actions that an agent takes. Because action selection is a problem that concerned researchers focusing on single-robot systems, it will not be discussed here. An overview of action selection research can be found in [26, 11]. Similarly, task decomposition is a problem for both single-agent systems [] and multi-agent systems