The invention discloses an
underwater glider rolling path
planning method based on meta
reinforcement learning, a total decision network comprises a first
convolution branch module, a second
convolution branch module, a feature splicing module and a full-connection neural network, the input of the total decision network is the state st = (T ', C,
delta x,
delta y) of the current position of an
underwater glider, and the output of the total decision network is the action at =
delta rt, delta rt is the course angle adjustment amount required by the
underwater glider to carry out the next section motion, T'is a
terrain height matrix corresponding to the current position of the
underwater glider, C is
ocean current speed data corresponding to the current position of the
underwater glider, delta x is the difference between the
longitude coordinate xT of the task end point and the
longitude coordinate x of the current position of the
underwater glider, and delta t is the difference between the
longitude coordinate xT and the longitude coordinate x of the current position of the underwater glider.
Delta y is the difference between the
latitude coordinate yT of the task end point and the
latitude coordinate y of the current position of the underwater glider. According to the total decision network, the navigation deviation caused by
ocean current in the navigation trajectory is reduced, the optimal path strategy generation efficiency is improved, and the
energy consumption of the underwater glider is reduced.