Distributed Satellite Resource Management for Space Superiority

Dynamic tasking of constellations of satellites is subject to bottlenecks caused by uncertainties in the outcome of tasking and in the time required to complete tasks. The situation is exacerbated in situations where an adversarial force may be intentionally threatening the satellites, attacking them or otherwise denying access. By dynamically predicting the outcome of the task and completion time, it is possible to allocate and schedule sensing and processing resources to improve mission effectiveness in executing against multiple tasks. Parietal Systems, Inc. (PSI) proposes to develop fast algorithms for task scheduling in a threatening environment using Bayesian learning and stochastic scheduling methods to optimize a user-specified measure of effectiveness. The algorithm will exploit an innovative approach based on generalized multi-armed bandits to develop an efficient online algorithm with minimal overhead. Because this approach works with various user-defined optimization criteria, it is able to optimize predictions with respect to criteria which match mission objectives. Because this approach accounts explicitly for alternative future outcomes, it is able to generate far-sighted (i.e., non-myopic) predictions that achieve near-optimal performance with respect to the selected optimization criterion. This results of this research will be a proof-of-concept demonstration and performance evaluation of the multi-armed bandit approach for task scheduling, along with detailed documentation of the algorithm and an analysis of its computational complexity. BENEFITS: The research described offers the potential of a significant improvement over existing techniques for performing dynamic scheduling in a wide set of military and commercial applications, including mission planning, work flow management, manufacturing process control, and computer-communications network control.