Learning-Based Path Planning Algorithm in Ocean Currents for Multi-Glider

Algorithm 1

Learning-based path planning algorithm.

(i)	Training phase (offline)
(1)	for epoch 1 to T do
(2)	for step 1 to M do
(3)	Randomly select one batch of training data. To restrict the movement based on equation (9)–(11).
(4)	Calculate the loss and its gradient based on equation (8).
(5)	Update Doc-CNN .
(6)	end for
(7)	end for
Planning phase (online)
(8)	Receive the initial state ( constant)
change over time and location,
(9)	while do
(10)	Input into Doc-CNN and output .
(11)	Using action based on , equation (7), (9)–(11).
(12)	Update state to .
(13)	end while
(14)	Send path planning results to glider formation.