DQN-Based Adaptive Filtering Scheme for Multisource Celestial Integrated Navigation

Celestial navigation provides several promising solutions for autonomous navigation, and the Kalman filter (KF) is widely utilized to address multi-source integrated navigation problems. The covariance information related to environmental changes and sensor noise is crucial for the performance of KF, and it is a key parameter in fusing information from different sensors. However, the precise description of the measurement models faces significant challenges. The complex dynamic environment in space and the variable observation conditions make it difficult for the system to obtain accurate covariance information in real-time. A multi-source celestial navigation scheme based on Deep Q-Network (DQN) adaptive filtering is proposed in this paper, in which DQN is utilized to select proper action to adjust the measurement noise variance matrix. The agent evaluates the quality of actions based on diversified reward functions. Monte Carlo simulation experiments show that the DQN algorithm can rapidly adjust the measurement noise covariance in real-time and improve the accuracy and reliability of the celestial integrated navigation system. Compared to the Sage-Husa adaptive filtering scheme, the method proposed in this paper provides faster convergence and higher precision, demonstrating greater stability in multi-source integrated navigation systems.