Q-learning-based space target selection for CNS/INS integrated navigation

To solve the problem that intermittent celestial observation will hamper the navigation accuracy of celestial and inertial integrated navigation system (CNS/INS), a Q-learning-based space target selection method is proposed. The visibility situation of space targets is analyzed under given flight mission, which lays the foundation for subsequent pointing optimization. Q-learning algorithm is employed to optimize the sequence of visible space target. The reward function is designed based on the trace of Fisher information matrix and the rotation angle of optical axis which could make a synthetic balance between navigation accuracy and rotation cost. By setting the optimized pointing sequence to CNS/INS integrated navigation system, unintermittent celestial observation can be achieved to guarantee filtering performance with relatively low rotation cost, which has a certain value of practical engineering application. Comparative simulation experiments demonstrate the effectiveness of the proposed method in terms of navigation accuracy and rotation cost.