Geomagnetic sensing navigation method based on deep reinforcement learning and simulated annealing

In the unknown environment without prior conditions, the path planning and navigation of an underwater autonomous underwater vehicle (AUV) is a big challenge. This study presents a perceptive navigation approach without a prior geomagnetic map. It achieves efficient path planning and geomagnetic map creation by combining deep reinforcement learning (DRL) with a simulated annealing (SA) algorithm. A deep Q network (DQN) is built to explore the environment of the carrier, collect local geomagnetic data, and then use the collected data to train a regression model to predict the global geomagnetic map. At the same time, the simulated annealing algorithm is used to optimize the path of an underwater AUV to avoid the local minimum problem of carrier space search. The success of the suggested approach is confirmed through a number of simulated tests in terms of path length, exploration efficiency, and geomagnetic map correctness. The results show that this method can significantly improve the navigation performance of underwater AUV in an unknown environment, and provide a new way for the construction of geomagnetic maps.