Ocean Engineering 235 (2021) 109355
7
C. Cheng et al.
Table 1
Comparison of main global path planning methods for known obstacle avoidance.
Algorithm
Advantages and disadvantages
Reference
Improvement
A* Algorithm
Direct search
Carroll et al.
(
1992
)
/
No preprocessing required
Szczerba et al.
(
2000
)
Sparse A* algorithm and optimization path
Large amount of calculation
Chen et al.
(
2012
)
/
Optimal solution not guaranteed
Yan et al.
(
2012a
)
The search process is more flexible
Yan et al.
(
2012b
)
Considering the anisotropy of current
Wang and Pang
(
2019
)
/
Li and Zhang
(
2020
)
Multi-direction A* algorithm is designed
Genetic Algorithm
Strong global searching ability
Alvarez and Caiti
(
2001
)
/
Slow convergence
Alvarez et al.
(
2004
)
Add iteration operator and random migration operator
Poor local optimization
Zhang
(
2006
)
/
Poor stability
Sun and Zhang
(
2012
)
Adjust the fitness function
Li et al.
(
2013
)
Add node deletion operator and smoothing operator
Cao et al.
(
2016
)
Change initial population and add tangential operator
Yan and Pan
(
2019
)
Improve diversity evaluation criteria and increase
mutation rate
Differential Evolution
Similar to genetic algorithm
Li et al.
(
2014a
)
/
Higher mutation probability
Zhang et al.
(
2014
)
Improved cost function
Mahmoud et al.
(
2018
)
Using current energy
Ant Colony
Strong global searching ability
Wang and Wei
(
2009
)
Add cutting operator and insertion point operator
Optimization
High efficiency
Zhang and Jia
(
2012
)
Add penalty factor
High convergence speed in later period
Yang et al.
(
2015
)
Pheromone elimination
Slow convergence speed in early stage
Dong and Xu
(
2017
)
Add the reinforcement idea
Yu et al.
(
2018
)
/
Ma et al.
(
2018
)
Use alarm pheromones to reduce invalid searches
Hu and Zhang
(
2019
)
Improved heuristic function
Che et al.
(
2020
)
Improved pheromone update rules
Ma et al.
(
2020
)
Improve the initial pheromone
Particle Swarm
Fast search time
Yang and Zhang
(
2009
)
Improved particle update strategy
Optimization
High convergence speed in early stage
Tang et al.
(
2010
)
Solving the critical point problem
Slow convergence speed in later period
Li et al.
(
2019c
)
Adaptive quantum particle swarm optimization
Easy to fall into local optimum
Wang et al.
(
2020
)
Improved quantum particle swarm optimization
Lim et al.
(
2020b
)
Combining with differential evolution algorithm
Lim et al.
(
2020b
)
learning
Sutton and Barto
(
1998
). Specifically, a positive reward will
be given to safe paths and a negative reward will be given to collision
paths during the training process.
Yu et al.
(
2018
) proposed
𝐴𝐶𝑂
-
𝐴
∗
by
combining the ant colony optimization (ACO) algorithm with
𝐴
∗
search
(
Phanthong
,
2015
). Based on fine-grained modeling,
𝐴𝐶𝑂
-
𝐴
∗
takes
the advantages of the
𝐴
∗
algorithm which uses admissible heuristic
functions for pairwise path planning and the ACO algorithm which
establishes the target’s marching order through coarse-grained model-
ing based on the representative estimation strategy. Their experimental
results show that AUV can successfully pass through the area with dense
obstacles. To reduce invalid search,
Ma et al.
(
2018
) introduced an
alarm pheromone in the ant colony optimization algorithm. When an
ant reaches the infeasible area, it releases alarm messages to alert the
ants behind. Compared with other methods, it can not only build a
collision-free path for AUV, but also can keep AUV away from obstacles
as far as possible.
To solve the limitation of single objective programming in the actual
marine environment,
Hu and Zhang
(
2019
) proposed an multi-objective
ant colony optimization strategy. In their method, the evaluation func-
tion for pheromone update is no longer determined by a single path
length, but also considers the impact of the current on energy con-
sumption and the safety of AUV. Their results in simulated experiments
show that the algorithm can make AUV reach the target point without
collision under the constraint of multi-target. With regard to energy
consumption which is an important factor for AUV path planning,
Ma
et al.
(
2020
) applied the firework-ant colony algorithm (
Zhang et al.
,
2018
) to the path planning of AUV. The initial pheromones of the ant
colony algorithm are derived from the initial reference path generated
by the improved fireworks algorithm, which can obtain the global
optimal solution in a shorter time. To allow AUV to find an optimal
path in complex seabeds,
Che et al.
(
2020
) proposed the ant colony
algorithm with particle swarm optimization. The particle swarm opti-
mization algorithm is introduced into the path length heuristic function
to improve the pheromone update rules. In addition, considering the
difficulty of AUV in three-dimensional space, the proposed method
limits the initialization range of individual population, greatly im-
proves the search efficiency and avoids a series of unnecessary pitching
and heave. Compared with the traditional ant colony algorithm, the
optimized path in the simulation is greatly shortened.
The ant colony optimization algorithm can easily find an optimal
path to the target point in environments with known static obstacles.
The searching efficiency can be greatly improved with distributed com-
puting and positive feedback mechanism, but the convergence might be
very slow. Therefore, the application of ant colony algorithm to AUV
path planning still needs further investigations.
3.6. Discussion
Table 1
shows advantages and disadvantages of global path plan-
ning methods with collision avoidance of known obstacles. Compared
with other algorithms, the
𝐴
∗
algorithm is the most direct heuristic
search algorithm without any preprocessing operation. However, if the
task is executed in a large-scale environment, a large number of nodes
need to be calculated, which results in a low searching efficiency.
According to the principle of survival of the fittest, the genetic algo-
rithm usually carries out a large number of evolutionary operations.
Therefore, it will occupy a large storage space and has more parameters
to be adjusted, resulting in a slow convergence speed. The advantage
of the genetic algorithm is that it has strong global search ability and
can overcome the sub-optimal solution problem of the
𝐴
∗
algorithm.
Similar to the genetic algorithm, a differential evolutionary algorithm
also has genetic operators such as selective cross mutation, which can
be defined with different methods. It is verified that the robustness
of the differential evolutionary algorithm is superior to the genetic
algorithm in a wide range of path planning applications.
The ant colony optimization algorithm and particle swarm opti-
mization algorithm are bionic methods inspired upon the behavior
mechanism of biological groups in nature. They can adapt to the
Ocean Engineering 235 (2021) 109355
8
C. Cheng et al.
Fig. 7.
The path planning mechanism of the basic RRT algorithm in the presence of
obstacles.
Source:
Modified from
Yu et al.
(
2017
).
environment in a shorter time with the cooperation between individual
organisms. The most significant feature of the ant colony optimization
algorithm is that it adopts a positive feedback mechanism, which can
promote the ants to move towards the direction with high pheromone
concentration. Because the distribution of the initial pheromones is
sparse, it takes more time to search at the beginning, which leads to
a slow convergence. In the later stage, with the increase of pheromone
concentration, the convergence rate increases obviously. Different from
the ant colony algorithm, the particle swarm algorithm makes use of
information shared among groups, and has faster convergence speed in
the initial stage. Meanwhile, the particle swarm optimization algorithm
has fewer parameters and certain memory functions, which can greatly
reduce the searching time. However, the lack of dynamic regulation of
particle velocity makes it prone to convergence at local optimum.
Do'stlaringiz bilan baham: |