“…From this comparison, the neural network with 256 number of hidden layer nodes and ReLU will be compared with other methods in Sections V-B2 and V-B3. From Section V-B1, the proposed method with the best network configuration is compared with other algorithms: i) offline path planning without the replanning, i.e., Section III-A, ii) method APG+LQG, consisting of APG [13] on the vertical plane (ED-plane) and LQG [2] on the horizontal plane (EN-plane), and iii) method VFG [14] that generates a reference path in a virtually designed vector space and form a vector field to make a vehicle follow the generated path in the vector space. Because most algorithms for terminal speed control are developed based on the vertical plane [13], [24], [36], the composition of guidance laws, APG+LQG, is selected as a compared method: APG [13] on the vertical plane takes charge of terminal speed control and vertical impact angle, and LQG [2] on the horizontal plane is designed for horizontal impact angle control.…”