Information Space Sensor Tasking for Space Situational Awareness

In this paper, we apply a receding horizon control approach to the sensor tasking aspect of a simplified version of the Space Situational Awareness (SSA) problem: "Given a small number of sensors and a large number of satellites, how should the sensors be used to maximize the information gained about the states of the satellites" Finding the globally optimal solution to this partially observed Markov decision process is computationally intractable. However, by using a stochastic gradient ascent algorithm proposed in previous work to improve an open-loop control policy over a shortened horizon, large performance improvements can be made over a baseline myopic tasking policy in a computationally tractable manner. The structure of this approach also allows for a distributed implementation in which each sensor acts as an agent that is semi-independent from the others.