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

Table 2

Accuracy and success rate of all architectures on two datasets.

Dataset	Metrics	Doc-CNN	DB-CNN	VIN
Grid maps	Acc1	96.1%	87.6%	85.8%
	Acc2	90.8%	85.4%	77.4%
	SR1	90.6%	87.6%	69.0%
	SR2	90.0%	83.0%	64.4%
	UEP	0.15	0.12	0.12
Ocean maps	Acc1	97.5%	88.5%	83.0%
	Acc2	92.2%	84.9%	75.6%
	SR1	94.3%	81.0%	48.7%
	SR2	86.5%	72.4%	46.2%
	UEP	0.16	0.12%	0.12%

Acc1 : path planning accuracy on training data. Acc2 : path planning accuracy on test data. SR1 : path planning success rate on training data. SR2 : path planning success rate on test data. UEP : unit energy forward path length for the glider. The bold values are the experiment results of the Doc-CNN algorithm proposed in this paper. The bold values of this group data are to strengthen its comparison with other data of algorithms, and it is to show the superiority of the Doc-CNN algorithm.