Path planning for underwater gliders in time-varying ocean current using deep reinforcement learning

The objective of this paper is to solve the application research of underwater glider (UG) and UGs formation, it is aiming to solve the path planning of gliders in ocean current environment by deep deterministic policy gradient (DDPG). Gliders can be deployed individually or collectively to execute ocean missions. Using the existing glider model and the interactions between gliders and environment, models close to the practical application of UGs are established. The deep reinforcement learning (DRL) based planning algorithm by integrating artificial intelligence, and solution to planning problem of UGs is provided. For a single UG planning, the designed RL algorithm can solve the compliance of UG motion constraints. The algorithm can calculate the appropriate path for the UGs formation, and change the shape of formation as necessary, which is useful for navigation in the environment of dense obstacles. With the same reward function, the improved DDPG outperforms the deep Q-network (DQN). Based on Tokyo Bay geography and unacquainted ocean, the developed algorithm is tested in ocean current environments.