Cordón et al.
defined according to the expert’s knowledge.
controller, and new driving instructions are outputted for AUV.
ocean currents.
Ocean Engineering 235 (2021) 109355
10
C. Cheng et al.
and Zhu
(
2011
) added an acceleration braking module in the fuzzy
reasoning process.
Fang et al.
(
2015
) obtained the corresponding hydro-
dynamic coefficients of AUV through the planar motion mechanism as
the important data input of the fuzzy control, and used the BK (
Bandler
and Kohout
,
1980
) triangular sub product of the fuzzy relations to
determine the turning angle of AUV when the obstacle exists. This self-
tuning fuzzy control system was tested to be able to avoid obstacles in
the horizontal plane. To solve the problem that the boundary design
of fuzzy logic in path planning relies heavily on expert experience,
Sun et al.
(
2018b
) used quantum-behaved particle swarm optimization
(
Sun et al.
,
2016
) to optimize the membership function value of fuzzy
logic. In addition, sonars are set in the horizontal plane and vertical
plane respectively, and the virtual acceleration and velocity of AUV
in 3D space can be obtained using the fuzzy system of acceleration or
deceleration module. Their results show that AUV with their proposed
method can automatically avoid dynamic obstacles. Because the speed
of moving obstacles in real environments is difficult to obtain,
Li
et al.
(
2019b
) take changes of the distance between AUV and obstacles
as additional input to the fuzzy membership function. Their results
show that when the obstacle moves rapidly, the proposed method is
obviously better than the traditional fuzzy logic algorithm.
The biggest advantage of the fuzzy logic algorithm is that it does
not need accurate mathematical models and the operation principle is
essentially similar to human cognition. Many researchers have achieved
good results in obstacle avoidance of AUV using the fuzzy logic al-
gorithm. However, the definition of fuzzy rules relies heavily on the
expert’s experience and cannot adapt to the environment. In complex
and uncertain underwater environments, the construction of fuzzy rule
base would be difficult or even impossible.
4.4. Neural network
Neural network is proposed for exploring the law of intelligent
activities by simulating human brains. The traditional neural network
for path planning takes the data collected by sensors as the input.
After training, the output drives the action for obstacle avoidance of
AUV.
Duan et al.
(
2001
) used a fuzzy neural network for real-time
AUV obstacle avoidance. In addition,
Li and Guo
(
2012
) designed a
dynamic neural network, which was tested to be able to bypass the
uneven obstacles and reach the target position in 3D spaces.
However, data for training a traditional neural network has to be
collected beforehand. Therefore,
Yang and Meng
(
2003
) proposed a
bio-inspired neural network which does not need any training process.
In bio-inspired neural network, the dynamic characteristics of neurons
are expressed by a shunt equation derived from the uniform diaphragm
model of the biological neural system (
Hodgkin and Huxley
,
1952
).
Fig. 10
shows the structure of a 2D bio-inspired neural network. In
Fig. 10
, each black circle represents a neuron. Each neuron interacts
with eight adjacent neurons to generate a real-time path according to
the dynamic activity diagram of the neural network. Because of the
high self-adaptability, bio-inspired neural network is more widely used
in the real-time path planning of AUV.
An improved bionic neural network based on
Yan and Zhu
(
2011
) is
used for AUV full coverage path planning. When encountering moving
obstacles, the activity of neural network will receive a great inhibition,
prompting AUV to adjust its direction.
Guerrero-González et al.
(
2011
)
used the self-organizing neural network to generate the transformation
between the spatial coordinates and the speed coordinates of AUV
propeller, and proposed a bio-inspired neural network based on animal
learning for obstacle avoidance of AUV. The proposed method can
drive AUV to rotate a certain angle according to the activation of
the neural network nodes. In unknown dynamic environment,
Zhu
et al.
(
2014
) applied the bio-inspired neural network and map planning
method to AUV path planning. Specifically, considering the uncertainty
of the ultrasonic sensor in underwater measurement, ultrasonic sensory
information is fused into a grid map to deal with dynamic obstacles
Do'stlaringiz bilan baham: