Deep Reinforcement Learning for Gravity Matching: A Siamese-CNN and LSTM Enhanced Particle Filtering Framework

The matching algorithm is one of the key technologies in gravity-aided inertial navigation systems. Usually, the filtering range of particle mass filtering(PMF) is determined according to the system noise characteristics, which limits the accuracy of particle filtering to a certain extent. This paper proposes a method for dynamically adjusting the filtering range based on deep reinforcement learning. The agent learns through interaction with the environment and can perceive the current system state in real-time. It then adapts the filtering range accordingly, optimizing the particle weight distribution during the filtering process and improving the filtering performance. Experimental results show that, compared to traditional PMF algorithm, the proposed method significantly improves positioning accuracy.