Page 35 - Read Online

Page 2 of 27 Wang et al. Intell Robot 2023;3(4):538-64 I http://dx.doi.org/10.20517/ir.2023.30

has become an important research direction [1,2] . In common search problems, target search can be divided
into stationary target search and moving target search based on the motion ability of targets. In stationary
target search, it is usually required to plan the UAVs’ search path in advance to achieve coverage of the mission
area [3–5] . However, since the target location may change at any time, UAVs need to update the search path in
real time based on the search results to achieve real-time and dynamic target search, leading to the problem of
moving target search.

Searchingformovingtargetstypicallyinvolvesconstructinganenvironmentalmodelwithoneorseveralsearch
[6]
informationmaps. Thesemapsusuallyincludeaprobabilitymap ,acertaintymap(alsoknownasuncertainty
[7]
[8]
map) , and a pheromone map . UAVs then update these search information maps based on the dynamic
search results, and the search path is planned based on the updated environmental information to improve the
search efficiency and probability of discovering targets. There are several concerns raised about this problem
in recent research.

Firstly, addressing the motion model of moving targets, Zhong et al. [9] used a Markov chain to represent a
hidden movement of targets and predict the position of targets. Hu et al. [10] assumed a type of moving target
thatcollaborateswithfixedsensorsprovidingUAV-sensingabilitydistributedinthemissionareaandproposed
a method to predict the distribution of targets. Yue et al. [11] proposed a specialized search map for the problem
of cooperative search in unknown sea areas, which achieves dynamic changes in target movement estimation
by assuming a fixed probability diffusion coefficient matrix for probability.

Regarding the target perception method of UAVs, Li et al. [12] studied the use of knowledge distillation (KD)
for target perception by presenting a comprehensive survey of KD-based object detection models developed
in recent years and offered valuable perspectives on incorporating object detection into the target search strat-
egy. Li et al. [13] provided a multi-modal perception method using the spatial-temporal graph obtained from
videos to promote latent space alignment in unsupervised multi-modal machine translation (UMMT), which
intersects with UAV perception capabilities.

On the topic of the encoding method of the search path, Shorakaei et al. [14] proposed a path planning method
that considers obstacles or threat areas using a novel encoding method by a matrix, which uniformly considers
all UAVs and the coordinates of their waypoint positions. Alanezi et al. [15] propose a motion-encoded genetic
algorithm with multiple parents, which realizes a unified motion-encoding on a series of UAVs. For optimiza-
tion, Luo et al. [16] used the fruit fly optimization algorithm to solve the search path, in which multiple fruit
fly swarms are used to enhance the global search ability. They adopted different search strategies through a
strategy switching method in the odor search and visual search stages, making the planning process more ef-
fective and stable. Similarly, other heuristic algorithms, such as differential evolution [17] and pigeon-inspired
optimization [18] , have been used in search path planning.

Toaddressthe resource allocation problem, Fang etal. [19] provideseveral policiesand optimization algorithms
to find a near-optimal solution associated both with high age of information as well as high power consump-
tion. Zhang et al. [20] investigated the reliable transmission scheme of downlink Non-Orthogonal Multiple Ac-
cess (NOMA) systems and provided valuable insights into realizing reliable transmission using NOMA with
randomly deployed receivers.

Whensearchingfortargets,theideaofmodelpredictivecontrol(MPC)iscommonlyusedforlong-termsearch.
Zhou et al. [21] applied distributed MPC (DMPC) combined with digital pheromone maps to realize the path
planning of regional cooperative search. The DMPC method based on Nash equilibrium can achieve global
optimization by locally optimizing the newly designed performance indicators. Yao et al. [22] considered the
communication network and information fusion between UAVs, designed a consensus algorithm with state