The role of information assumptions in decentralized task allocation: A tutorial

A single robotic agent can perform cost-effective work in many domains, especially those that are dirty, dangerous, or dull for humans. Forming multiagent teams of autonomous robots can even further improve their capabilities. A major reason for this increase in performance is that teams of agents can take measurements and act in many places at once, and thus can react much faster than a single agent to a dynamic environment. Additionally, multiple individual agents with overlapping capabilities can provide mission completion robustness in the event of an individual agent failure. However, the benefits of multiagent teams come at the expense of added complexity and resource consumption required to coordinate team behavior. For the purposes of this tutorial, there are two main paradigms for solving this multiagent coordination problem: (1) centralized coordination, where a single control center receives all relevant information necessary to define the motion and actions of every agent in the team, and (2) decentralized coordination, where individual agents make their own decisions about motion and action with the use of local communication and a priori information about the planning environment.