Collaborative Goal Distribution in Distributed Multiagent Systems

Distributed multiagent systems consist of multiple agents which perform related tasks. In this kind of system, the tasks are distributed amongst the agents by an operator based on shared information. The information used to assign tasks includes not only agent's capability, but also agent's state, the goal's state, and conditions from the surrounding environments. Distributed multi agent systems are usually constrained by uncertain information about nearby agents, and by limited network availability to transfer information to the operator. Given these constraints of using an operator, a better designed system might allow agents to distribute tasks on their own. This paper proposes a goal distribution strategy for collaborative distributed multi agent systems where agents distribute tasks amongst themselves. In this strategy, a goal model is shared amongst all participating agents, enabling them to synchronize in order to achieve complex goals that require sequential executions. Agents in this system are capable of transferring information over the network where all others belong to. The approach was tested and verified using StarCraft II APIs, introduced by Blizzard and Google Deepmind.