Page 36 - Read Online

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

predictor based on the minimum spanning tree structure to realize the fusion of predicted target probability
map, and proposed a future-dependent MPC framework to realize the cooperative trajectory optimization
and obtain the optimal control input. In addition, deep reinforcement learning (DRL) is also a promising tool
for solving such problems. Wei et al. [23] proposed a joint design of the unmanned aerial/surface/underwater
vehicle (UAV-USV-UUV) network for cooperative underwater target hunting and conceived a novel deep Q-
learning (DQN) algorithm to solve the target hunting problem. These research results provide several new
perspectives to solve the target search problem.

Based on the studies mentioned above, this paper considers a multi-UAV regional cooperative search problem
for targets with the ability to perceive and evade. The targets are equipped with a UAV-sensing device and
move randomly in the mission area in a Markovian fashion. However, they can perceive the UAVs and take
corresponding evasive actions by increasing the distance from the UAVs to achieve the ability of autonomous
evasion, which increases the difficulty of target search.

This paper proposes a novel Cooperative Search Method for Targets with the ability to Perceive and Evade
(CSMTPE). Firstly, we define the motion model of such targets in detail and design various search information
maps together with their update methods based on the prediction of moving targets and the search results of
UAVs. Secondly, we establish a multi-UAV search path planning optimization model based on MPC and
design objective functions of search benefits and costs. Finally, we propose an improved genetic algorithm
with a fine-adjustment mechanism (IGAFA) to solve this optimization model. The simulation results confirm
the effectiveness of the proposed method.

Several contributions are made in this paper to the regional cooperative search problem for targets with the
ability to perceive and evade:

(i) A new motion model of moving targets and a detection model for UAVs are proposed, which are more
consistent with real-world search scenarios compared to the traditional methods.
(ii) The updated formula for the traditional probability map used to predict target probability is improved,
enhancing UAV search efficiency when dealing with the evasive maneuvers of moving targets.
(iii) A search information map that reflects the detection response of moving targets is introduced, providing
real-time feedback to UAVs on their search results.
(iv) A multi-UAV search path planning optimization model is established, and a series of objective functions
are designed for this model.
(v) A priority-encoding method for multi-UAV search paths and a priority-encoded IGAFA algorithm are
proposed, effectively solving the optimization problem of multi-UAV search path planning effectively.

Overall, the effectiveness and efficiency of cooperative search for moving targets with the ability to perceive
and evade are improved by these contributions, which could have practical applications in fields such as search,
rescue, surveillance, and military operations.

The rest of this paper is structured as follows: Section 2 describes the search problem and presents a model
for it. Section 3 defines the search information maps and their update methods. Section 4 introduced a multi-
UAVcooperativesearchmethodandoptimizationmodel. Section5presentstheencodingmethod, theIGAFA
algorithm, and the complete search method. In Section 6, simulation experiments confirm the effectiveness of
the proposed method. Finally, in Section 7, we present the conclusion.

2. PROBLEM FORMULATION
Assuming that there are UAVs equipped with detection sensors searching for moving targets with the
ability to perceive and evade within a mission area , where the width and height of are and , and the