Research on geomagnetic sensing navigation method based on deep reinforcement learning + simulated annealing

Path planning and navigation of autonomous under- water vehicles in unknown environments without a prior condi- tions is a major challenge. In this paper, we propose a perceptual navigation method without a prior geomagnetic map, which combines deep reinforcement learning and simulated annealing algorithm to realize effective path planning and geomagnetic map construction. The method constructs a deep Q-network to explore the carrier environment, collects local geomagnetic data, and then uses the collected data to train a regression model to predict the global geomagnetic map. Meanwhile, a simulated annealing algorithm is used to optimize the path of the autonomous underwater vehicle to avoid the local minimization problem in the carrier space search. Through a series of simulation experiments, it is demonstrated that the method in this paper significantly outperforms the traditional method in terms of performance, exhibiting higher path smoothing and better path planning capability, while reducing the path variance and standard deviation. In addition, the method has faster convergence speed and higher loss convergence stability. The results show that the method can significantly improve the navigation performance of autonomous underwater vehicles in unknown environments and provides a new method for the construction of geomagnetic maps.