Decentralized Control for an Autonomous Team

This paper addresses the development of decentralized algorithms for the dynamic planning and assignment of multiple coupled tasks in cooperative decision and control of an autonomous UAV team. Generic vehicle-target scenarios are considered with task precedence, task timing, communication constraints, and uncertainty. Various distribution philosophies are discussed as a function of performance, information, and control. A trade space between task coupling, message cost, and decentralization is postulated. This structure is analyzed with respect to optimality, task coupling, communications throughput, delays, decentralization, and state estimation. Algorithms discussed include: mixed integer linear programming, auction, distributed constraint satisfaction, network sow, and hybrid iterative variations. Preliminary analysis has revealed that, for strongly coupled tasks, a decentralized solution may not be better than a centralized solution, unless communication costs are low. Also, team performance with restricted communications is driven by the capability of the onboard team state estimators.