Page 50 - Read Online

Wang et al. Intell Robot 2023;3(4):538-64 I http://dx.doi.org/10.20517/ir.2023.30 Page 17 of 27

Table 1. Procedures of the priority-encoded IGAFA
Name: Priority-encoded IGAFA
Goal: Obtain the optimal set of control input sequence
1: Input: probability map Ω ( ), certainty map Ω ( ), detection response map Ω ( ), multi-UAV system state ( ), max-

imum iteration step , population size , probability of sexual crossover 1, probability of asexual crossover 2,
probability of mutation ;
2: Output: the optimal set of control input sequence U ( | );
∗
3: Generate random chromosomes as current population , let parent population = , child population = ∅, minimum
fitness = inf, weight matrix W = 0;
4: for =1:
5: From , generate 2 · ⌈ · 1 /2⌉ children by sexual crossover and move them into ;
6: From , generate ⌈ · 2 ⌉ children by asexual crossover and move them into ;
7: From , generate ⌈ · ⌉ children by mutation and move them into ;
8: ← + ; //Let all parents and children be the current population
9: for =1:size( )
10: if the fitness of i-th chromosome is not calculated
11: Decode i-th chromosome to a set of control input sequence U( | );
12: Calculate the fitness J( ( ), U( | )) by Equation (25);
13: if J( ( ), U( | )) <
14: ← J( ( ), U( | )); //Note the minimum fitness
15: U ( | ) ← U( | ); //Note the corresponding control input sequence
∗
16: end
17: end
18: end
19: Update weight matrix W by Equation (30);
20: ← ∅, ← ∅; //Clear the parent and child population
21: From , select chromosomes by ”binary tournament selection” and move them into ;
22: end
23: return U ( | );
∗
Table 2. Procedures of a complete search process using CSMTPE

Name: Complete search process using CSMTPE
Goal: Find all moving targets

1: Initialize multi-UAV system state (0) and probability map Ω (0), set search step = 0, certainty map Ω (0) = 0 and

detection response map Ω (0) = 0;
2: while not all targets are found
3: Get the optimal set of control input sequence U ( | ) by IGAFA with the procedures in Table 1;
∗
4: From U ( | ) get ( | ) as ( ), then move one step and do search;
∗
∗

5: Update Ω ( ) by Equations (11-13);

6: Update Ω ( ) by Equations (14) and (15);

7: Update Ω ( ) by Equations (16-18);
8: for =1:

9: if any ( )≥ for ∈ Ω ( ) //The existence probability in exceeds the threshold
ℎ
10: Turn i-th into tracking state and no longer participate in subsequent search missions;

11: Clear the element of the discovered target in Ω and Ω to block the influence on the subsequent search;
12: end
13: end
14: ← + 1;
15: end

detection configuration for UAVs to be: = 5, = 0.95, = 0.2, = 100 , = 400 , and

= 0.75.
ℎ
6.1. Motion prediction of moving targets
Assuming that there are two targets to be searched in the mission area, the initial position of the target
is subject to the two-dimensional normal distribution, where the means of position are [2500 1500] and

[2500 2600] ,andthevariancesofpositionare [160000 0; 0 160000] and [160000 −128000; −128000 160000].
To test the update process of the probability map, three experimental scenarios were used: (I) target prediction
without the search interference of UAVs, (II) under the search interference of static UAVs, and (III) moving
UAVs. Figure 8 shows the updating process of the probability map in different scenarios at simulation steps
k = 0, k = 10 and k = 20. In scenario I, as the targets are not affected by the UAV search, the target follows