Semi-autonomous management of multiple ad-hoc teams of UAVs

A Distributed Battle Manager (DBM) system based on Concurrent MACro Partially Observable Markov Decision Processes (CMAC-POMDP) is proposed to manage multiple competing missions. Implementation details are presented for a semi-autonomous management of multiple ad-hoc teams of UAVs operating under extreme uncertainty in a contested environment. The proposed solution is semi-autonomous since command and control teams perform target confirmation, weapon selection, weapon release, and bomb damage assessment. Our proposed approach decomposes the resource management requirements into specialized mission timelines. There would be one adaptive mission timeline for each system objective, and each timeline would be represented by a POMDP. At any time, the DBM would be controlling multiple concurrent timelines that are competing for system resources. Each timeline is solved independently to obtain the optimal MDP solution parameters. These solution parameters are then combined using linear programming with constraints to obtain a suboptimal system solution. The resulting suboptimal solutions provide satisfactory real-time system performance. Under extreme uncertainty, it is preferable to maximize robustness utilizing an adequately performing solution as opposed to finding an optimal but brittle solution. Performance results of the core CMAC-POMDP algorithm are presented for a system implemented to manage multiple heterogeneous sensors addressing multiple simultaneous threats. This simpler resource management system utilizes the same timeline for all of the threats. Simulations demonstrate better than real-time performance even for large numbers of threats. Furthermore, computational requirements are shown to grow linearly with the number of threats. Extension to multiple simultaneous timelines for the DBM system is straightforward.