Energy-Efficient Reinforcement Learning for Motion Planning of AUV

The accuracy of mapping results depends on the Autonomous Underwater Vehicles (AUVs) navigation errors. However, signals are attenuated drastically in water, which makes it difficult for AUVs to receive signals underwater. As a result, traditional navigation methods may become unreliable. In this paper, a terrain-aided navigation method that takes into account distance and energy consumption is proposed, which does not rely on a precise positioning system. Given the complexity of 3D terrain, this paper formulates the problem as a Markov decision process (MDP) and aims to minimise the energy cost function, where a motion planning method based on soft actor-critic (SAC) is formulated. Then a 3D energy consumption calculation method is developed for the AUV which is accurate for each time slot during the training process. Finally, experiments on the Gym platform were carried out to verify the effectiveness of the proposed method.