Distributed algorithms for dynamic reassignment

In this paper, we consider the problem of task partitioning among members of a team of cooperating agents in response to discovery of new tasks or potential failures of some agents. We assume that information about new targets or agent failures is received by individual team members, and communicated asynchronously with delays to the rest of the team. These delays create potential differences in information across the team of agents. We describe an asynchronous approach to coordinating the team response, where individual agents compute modifications to assignments based on local information. We show that the asynchronous algorithms converge to the same optimal assignments in the presence of arbitrary finite communication delays as a centralized information approach. We extend the asynchronous protocol to the solution of a class of stochastic dynamic resource assignment problems, and show asynchronous convergence of the resulting algorithms. Simulations illustrate the delays in computing an optimal assignment of tasks in response to dynamic events.