Augmenting spacecraft maneuver strategy optimization for detection avoidance with competitive coevolution

Optimization in the spacecraft detection avoidance problem is computationally cost prohibitive given the size of the state and action spaces available to both players. Competitive coevolution can be used to augment strategy optimization by reinforcement learning in a manner that results in dynamic search spaces. An evading spacecraft and a pursuing sensor compete directly with each other and reciprocally drive one another to increasing levels of performance and complexity. This is accomplished by gradually increasing the size and complexity of the strategies. Using coevolution provides significant computational cost savings compared to traditional optimization methods while ensuring a globally optimal result. It is found using competitive coevolution that a spacecraft is able to successfully evade a tasked sensor nearly half of the observation window time with only one 19m/s maneuver every 5days.