You are on page 1of 25

Article 1

The path planning of Coal mine underground Autonomous 2

trackless rubber-tyred vehicle based on the fusion improved A* 3

and DWA algorithm 4

Chuanwei Zhang, Xinyue Yang*, Peilin Qin, Rui Zhou and Zhongyu Guo 5

College of Mechanical Engineering, Xi’an University of Science and Technology, Xi’an 710054, China; 6
zhangcw@xust.edu.cn(C.Z.); 20105016012@stu.xust.edu.cn(P.Q.); 22105016007@stu.xust.edu.cn(R.Z.); 7
21205016003@stu.xust.edu.cn(Z.G.) 8
Correspondence: 21205016036@stu.xust.edu.cn; Tel: +86-187-9255-8917 9

Abstract: To solve the issue of path planning of Autonomous trackless rubber-tyred vehicles under- 10
mines, a fusion path planning algorithm based on improved A* and DWA is proposed in this paper. 11
The optimized and improved A* algorithm can plan a shorter global path and have a faster effi- 12
ciency of path generation. Combined with the improved DWA local algorithm, global path planning 13
and local obstacle avoidance can be realized. The Experiments are carried out in actual scenarios, 14
and the 3D point cloud map is constructed by integrating LIDAR and IMU perception. Experiments 15
are conducted separately by setting no obstacles, static obstacles, and unknown dynamic obstacle 16
environments to verify the effectiveness and accuracy of the improved fusion algorithm. And com- 17
pare the proposed algorithm with existing similar algorithms, and the improved algorithm is better 18
adaptable, robust, and has low computational complexity. The results show that the path searched 19
by the improved fusion algorithm is smoother and more reasonable and can avoid unknown dy- 20
namic obstacles in time, it has higher navigation efficiency and stability. 21

Keywords: trackless rubber-tyred vehicle; coal mine, path planning; improved A*; improved DWA; 22
fusion algorithm 23
24

1. Introduction 25
Mine transportation is the internal transportation of coal mine enterprises and is an 26
important link in coal mine production. Underground mine transportation is mainly di- 27
Citation: To be added by editorial vided into main transportation and auxiliary transportation. In addition to the main trans- 28
staff during production. portation using the transport mechanism, the trackless rubber-tyred vehicle, as one of the 29

Academic Editor: Firstname Last-


important transportation equipment of the underground, auxiliary transportation system 30

name
of the coal mine, connects the mine face with the ground, making the coal mine produc- 31
tion orderly. Therefore, realizing the unmanned driving of underground vehicles in coal 32
Received: date
mines is an inevitable trend in the intelligent development of the coal industry [1]. 33
Revised: date
In the process of underground roadway tunneling, due to the uneven distribution of 34
Accepted: date
underground coal resources, rock hardness, structure, and the different functions of the 35
Published: date
roadway section itself, the underground roadway is complicated, and slope, turning, and 36
other phenomena often appear in the roadway [2]. There are many complicated under- 37
ground transportation tasks, the destinations are scattered, and the transportation routes 38
Copyright: © 2023 by the authors.
often change. Mine underground trackless rubber-tyred vehicles are prone to congestion 39
Submitted for possible open access
up and down a slope and across the roadway, which leads to an increased difficulty for 40
publication under the terms and con-
drivers to drive vehicles [3]. The environment under the mine is dark and dusty, and the 41
ditions of the Creative Commons At-
driver is easy distinguish the road conditions, which greatly reduces transportation effi- 42
tribution (CC BY) license (https://cre-
ativecommons.org/licenses/by/4.0/).
ciency and may lead to the occurrence of safety accidents. 43

Electronics 2023, 12, x. https://doi.org/10.3390/xxxxx www.mdpi.com/journal/electronics


Electronics 2023, 12, x FOR PEER REVIEW 2 of 25

In order to realize the path planning requirements of autonomous trackless rubber- 44


tyred vehicles under the mine, it is necessary to consider the road structure, environment 45
of the coal mine roadway, and the driving criteria that must be observed when driving. 46
The path planning of coal mine auxiliary transport vehicles is similar to the development 47
technology of intelligent vehicles on urban roads, the difference lies in the working envi- 48
ronment of mining vehicles; there are more blind spots caused by fork roads, more un- 49
known emergencies, low underground illumination, less positioning basis caused by sin- 50
gle scene, insufficient network transformation, In addition, the research and development 51
needs to consider the industry standards and safety production indicators of the mining 52
area, so the advancement and accuracy of path selection and obstacle avoidance work in 53
the work. 54
It is because of such special road structures, strict driving rules, and other reasons 55
the transportation problems of trackless rubber-tyred vehicles face many difficulties [4], 56
[5]. Ensuring the smooth transportation of trackless rubber-tyred vehicles is the primary 57
problem in improving coal mine production, which puts forward higher requirements for 58
underground unmanned trackless rubber-tyred vehicle technology [5,7]. However, the 59
research on unmanned driving technology in coal mines starts late, is difficult, and has 60
strict requirements. There are still bottlenecks in technology and application scenarios, 61
which still require in-depth exploration and research in all aspects. At present, scholars 62
have begun to study the technology of unmanned trackless rubber-tyred vehicles under 63
the mine, which has great research significance and broad application prospects for re- 64
ducing the occurrence of accidents under the mine, reducing property losses, and improv- 65
ing coal production efficiency. 66
The path planning of mining unmanned trackless rubber-tyred vehicles includes the 67
following steps: 1. It can successfully plan a smooth, complete path on the map. 2. The 68
vehicle can reach the destination smoothly along this path and complete the auxiliary 69
transportation task of the coal mine. 3. In the process of vehicle driving, unknown dy- 70
namic obstacles can be avoided in time, and a new path can be planned to continue to 71
complete the above two points. 72
Path planning can be divided into global path planning and local path planning [8] 73
according to the ultimate goal of the planning. Global path planning is the planning of 74
calculating the best route on the entire map of the environment, which is to guide and 75
constrain local planning [9]. Local path planning is to perceive the information of the sur- 76
rounding environment and obstacles by using sensors to constantly correct the planned 77
global path. At the same time, it is the exploration of the unknown environment and the 78
avoidance of dynamic obstacles which belong to real-time dynamic planning [10]. 79
Currently, more mature and faster-applied global path planning algorithms mainly 80
include the A* algorithm [11], the Dijkstra algorithm [12], and RRT [13] et al. The local 81
path planning algorithm includes an artificial potential field algorithm [14], dynamic win- 82
dow algorithm [15] and deep learning algorithm [16], and neural network algorithm [17] 83
et al. These methods have different characteristics, and choosing appropriate algorithms 84
in different situations can make the path planning results more optimal. Table 1 summa- 85
rizes the eight common path planning methods. 86

Table 1. Comparison of common general path planning algorithms. 87

88

89

90

Algorithm Characteristic Applicable environment


Electronics 2023, 12, x FOR PEER REVIEW 3 of 25

Path planning is directional, but it requires 91

high map requirements, which is only suitable Environmental information 92


A*
for path planning with a known end point and is known for static planning.
93
start point.
Optimal nodes are selected using a greedy 94
Environmental information
Dijkstra strategy, and internode distances are updated
is known for static planning. 95
for each iteration.
It belongs to the updated version of the A * al- Dynamic planning of envi- 96

D* gorithm, using the current environmental in- ronmental information in


97
formation to repair the planned route. known parts.
When planning, the next node is randomly 98

generated until the endpoint is reached, and


Environmental information 99
RRT the shortest path is inversely selected. The path
is known for static planning. 100
is not optimal, and the convergence rate is
slow. 101
Set repulsion for obstacles and gravity for the Dynamic planning of envi-
102
APF endpoint, which is suitable for obstacle avoid- ronmental information in
ance. known parts. 103
Considering the velocity and kinematic con- Dynamic planning of envi-
104
DWA straints of autonomous vehicles, there are lo- ronmental information in
cally optimal solutions for the resulting paths. known parts. 105

The algorithm has self-learning and self-deci- Dynamic planning of envi-


106
DL sion functions, strong data dependence, and ronmental information in
107 high demand for computational resources. known parts.
Distributed parallel information processing, Dynamic planning of envi- 108
NN strong robustness but poor generalization abil- ronmental information in 109
110
ity, more combined with other algorithms. known parts.
111
112
113
114
115
116
117
A*, Dijkstra, and RRT are all heuristic algorithms suitable for global path planning, 118
and the globally shortest path can be realized by selecting a directed and continuous path 119
set from the starting point to the endpoint. Among them, the A* algorithm is a heuristic 120
search algorithm that can solve the shortest circuit and the most effective direct search 121
method in the static road network. Compared with other global path planning algorithms, 122
the A* algorithm has the advantages of small operation amount, and high efficiency, but 123
when the environmental complexity increases and the obstacles are dense, the efficiency 124
of the A* algorithm is low, the computational amount increases sharply, and it is difficult 125
to search for the optimal path, and the planning path has many inflection points, which is 126
not conducive to vehicle tracking. 127
Dijkstra Uses breadth-first search by giving road weights, trying to choose the opti- 128
mal solution at each step, and finally obtains the shortest path, which can calculate the 129
Electronics 2023, 12, x FOR PEER REVIEW 4 of 25

shortest path from any node in the graph to all other nodes. Compared with other algo- 130
rithms, it has the advantages of simple principles and fast operation speed [18]. However, 131
this is only applicable to path planning in small environments, which is inefficient in large 132
environments and consumes computing resources. D* [19] algorithm is very effective in 133
pathfinding in dynamic environments, such as robot road finding. For distant shortest 134
path changes, it is less applicable. RRT is a sampling-based path planning algorithm, 135
which can explore unknown regions to the greatest extent and has probability complete- 136
ness, but the algorithm is blind in search, the convergence rate is slow, and the search 137
efficiency is low [20]. APF (Artificial potential field) method is a feedback control strategy 138
that is robust to real-time obstacle avoidance and error control. However, when multiple 139
obstacles appear in the configuration space, a zero potential energy point is easy to appear, 140
causing the vehicle to be unable to reach the target point [21]. DWA algorithm predicts 141
the motion trajectory of the vehicle in a certain period of time through multiple groups of 142
sampling speeds. It uses the evaluation function to score these prediction trajectories and 143
selects the speed corresponding to the optimal trajectory, so as to realize local obstacle 144
avoidance [22]. This algorithm is more efficient, uses less memory, and enables real-time 145
obstacle avoidance, but sometimes it is not a globally optimal path. Deep learning algo- 146
rithm, including reinforcement learning, can be expressed as the Markov decision prob- 147
lem [23] in solving the problem of path planning. This path planning algorithm generally 148
has a feedback mechanism, strong robustness, and good adaptability, but the convergence 149
rate is slow, high time complexity, easy to fall into local convergence, and large memory 150
consumption [24]. 151
Non-heuristic algorithms such as Dijkstra and D* are more inclined to random tra- 152
versal search in optimization, which runs counter to the path planning requirements of 153
autonomous vehicles in coal mines, So like A* classic heuristic global path planning algo- 154
rithm compared to other global path planning algorithm, has strong plasticity, and easy 155
to integrate with other algorithms. In the field of intelligent vehicles, the sweeping robot 156
has a relatively mature application; currently in the field of unmanned coal mines received 157
extensive research. 158
Hong J.Z. et al. [25] proposed a new improved A* algorithm, which seeks the optimal 159
search path by changing the weight of the enlightening function in real-time and applying 160
a different coefficient ratio u according to the characteristics of the scene. The improved 161
A* algorithm is more judgmental. The smoothness and robustness of the coal mine robot 162
are good, and the distance is significantly reduced, which meets the functional require- 163
ments of the system. 164
Meng J.L. et al. [26] considerd the factors affecting the roadway escape, chose the 165
reasonable evaluation function, put forward the use of a bidirectional A* algorithm, 166
made each step of the search according to the evaluation function in the direction of a 167
targeted search, effectively promote algorithm to accelerate convergence, the example 168
shows that bidirectional A* algorithm planning escape path node less, shorter length, can 169
meet the actual needs of 170
Yuan B.H. et al. [26] introduced the target gravity idea in the artificial potential field 171
method into the RRT algorithm, which corrected the growth direction of the extended tree 172
and avoided the generation of local minimum. At the same time, the idea of hierarchical 173
extension is added to the algorithm to make it has the characteristics of finding the target 174
by avoiding obstacles. Compared with the traditional methods, the path planning speed 175
and obstacle avoidance ability are improved. 176
Jiu S.B. et al. [28] proposed that based on improved A* and artificial potential field 177
joint algorithm, the methods of exponential function weighting and cubic spline interpo- 178
lation are used to improve the A* algorithm, respectively, join the correction factor and 179
escape to improve artificial potential field method, the two algorithm fusion improvement, 180
generated path more smooth and better continuity, can meet the trackless rubber wheels 181
in coal mine roadway path planning requirements. 182
Electronics 2023, 12, x FOR PEER REVIEW 5 of 25

Zi J.T. et al. [29] used the artificial potential field method under mine navigation, 183
introducing the relative speed field and relative acceleration field, putting forward the 184
concept of " a total situation field line" underground robot for local planning on the basis 185
of the global environment to complete the optimal path selection, the method is more au- 186
tonomy in navigation, and get the path is better. 187
Yuan X.M. et al. [30] used the traditional heuristic search A* algorithm for global path 188
planning. First, build a grid map and then plan the global desired path according to the 189
principle of the path of the shortest path; then add the local path planning DWA algorithm, 190
which selects the speed corresponding to the optimal trajectory to drive the robot move- 191
ment and realize the autonomous path planning of mining vehicles in the underground 192
coal mine. 193
The above examples are all explorations and studies of the safe operation of un- 194
manned autonomous path planning under the mine, but most of the algorithms are still 195
in the simulation test stage, and their adaptability to the environment of coal mine auxil- 196
iary roadway needs to be verified. In the study of driverless path planning methods, schol- 197
ars often optimize the shortcomings of the algorithm rules, enhance the adaptability of the 198
algorithm, improve the algorithm convergence speed, and reduce the amount of algo- 199
rithm calculation; by simplifying the algorithm model, accelerating the search speed, im- 200
prove the algorithm real-time [30]. For example, the variable step search strategy is added 201
to the RRT algorithm, and various external force constraints are added in the artificial 202
potential field method to establish the multi-constraint model. Or complement multiple 203
algorithms to overcome the shortcomings of the algorithm itself. For example, the A* al- 204
gorithm is combined with the artificial potential field method, particle swarm algorithm, 205
genetic algorithm fusion, etc. 206
In view of the complexity and particularity of the underground coal mine environ- 207
ment, it is difficult to achieve accurate, safe, and reliable path planning with a single algo- 208
rithm, and it is impossible for any algorithm to deal with all path planning problems. In 209
addition, it is difficult to develop new algorithms, and it does not necessarily perform 210
better than traditional algorithms. In contrast, it is more inclined to complement the ad- 211
vantages of various algorithms so as to overcome the defects of the algorithm itself. At the 212
same time, corresponding planning algorithms and control strategies are formulated for 213
specific mines and different types of auxiliary transport models to ensure the safe and 214
efficient operation of unmanned vehicles in underground coal mines. 215
Nowadays, unmanned vehicles at home and abroad are mainly aimed at the research 216
of urban structured road environments and unstructured road environments [32]. And 217
the research on driverless technology of mining trackless rubber-tyred vehicles has been 218
rarely mentioned. In this paper, a fusion path planning algorithm based on the traditional 219
heuristic A* and DWA is proposed for the special road environment of the underground 220
coal mine. The core idea is to add local path planning on the basis of global path planning. 221
According to the variable environmental characteristics of the underground coal mine, 222
the global path planning algorithm and the local path planning algorithm are improved, 223
respectively. Secondly, the two improved algorithms are designed and integrated. Finally, 224
the effectiveness and feasibility of the improved algorithm are verified through experi- 225
ments. The improved algorithm can not only enable the trackless rubber-tyred vehicle to 226
plan a complete and smooth shortest path in the mine roadway but also meet the driving 227
needs of mining vehicles to avoid static and dynamic obstacles in the driving process, 228
making the path more reasonable and with higher navigation efficiency and stability. 229
The innovation and main contributions of this article are as follows: 230
1. Improvement of the global path planning algorithm: Considering the influence 231
of complex environmental obstacles in mines, the environmental obstacle factor 232
is introduced to optimize the heuristic function to realize the adaptive adjust- 233
ment of the A* algorithm and enhance the adaptability of the A* algorithm to the 234
surrounding environment. At the same time, the global path adopts an optimi- 235
zation strategy to optimize the global path. 236
Electronics 2023, 12, x FOR PEER REVIEW 6 of 25

2. Improvement of the local path planning algorithm: To prevent the optimal local 237
results of the traditional DWA algorithm in local planning, the global path infor- 238
mation is introduced as a new evaluation subfunction. Secondly, in order to im- 239
prove the flexibility of the algorithm, the original evaluation function is designed. 240
3. Design the fusion algorithm: The key point of global path planning is added to 241
the local path planning as the local target point, the A* algorithm is improved to 242
provide the globally optimal path, and the global path key point as the local tar- 243
get point of improved DWA guides the vehicle to avoid local obstacles, and the 244
fusion algorithm is designed. Finally, this paper compares and analyzes the im- 245
proved algorithm with the existing path planning algorithm similar to the A*- 246
artificial potential field joint algorithm. 247

2. Global Path Planning Based on Improved A* Algorithm 248

2.1. A* Algorithm 249


A*, as a classical heuristic algorithm, can solve the most short-circuit and the most 250
effective direct search method in the static road network. By calculating the cost function, 251
the child node with the smallest generation value around the parent node is obtained, and 252
as the new node, the search is continuously promoted until the target point is found and 253
the global path search is completed. The cost function is as follows in Equation (1): 254

F (n) = G(n) + H (n) (1)

where n indicates the current node and F (n) Indicates total generation value, G(n) repre- 255
sents the actual generation value of the starting point to the current node n, H (n) is the 256
heuristic function for the estimated generation value of the current node n to the target 257
point. The closer the estimate of the heuristic function is to the true value, the faster the 258
speed and accuracy of the algorithm will be; however, in practical application, the envi- 259
ronment often becomes complex and unknown due to the existence of obstacles and the 260
choice of H (n) in different environments can often determine the efficiency and accuracy 261
of the search path. 262

2.2. Improved A* Algorithm 263


2.2.1. Quantification of the Obstacle Information 264
In order to improve the search efficiency of the algorithm, considering the influence 265
of obstacles in the complex environment, the heuristic function needs to be optimized to 266
realize the adaptive adjustment of the algorithm by quantifying the environmental obsta- 267
cle information. The Euclidean distance, as shown in Equations (2). Improves the heuristic 268
function with the optimization of the proportion of obstacles in the occupancy grid map. 269
Assuming that the vehicle has a starting point of (X i ,Yi ) an ending point of, (X g , Yg ) and 270
the number of obstacle grids in the region of the rectangular grid composed of the starting 271
point to the ending point of N. The proportion of the obstacle grid number w is shown 272
in Equation (3): 273

H (n) = (x g − xi ) + (y
2
g − yi )
2
(2)

N
w=
(1 + x − x ) (1 + y − y )
i g i g (3)

2.2.2. Optimization Heuristic Function H(n) 274


Electronics 2023, 12, x FOR PEER REVIEW 7 of 25

The quantized environmental information is introduced into the exponential func- 275
tion weighting to optimize the heuristic function. The proportion of the number of obsta- 276
cle grids w indicates the density of obstacles within a certain range. The improved A* al- 277
gorithm is shown in Equation (4): 278

H ( n)
F (n) = G (n) + ln H ( n) (4)
w
where when the number of obstacles is small or far away from the ending point, the 279
weight coefficient of the heuristic function will increase, which accelerates the search effi- 280
ciency and improves the agility of the algorithm. And when the number of obstacles in- 281
creases, or the ending point is close, the weight coefficient of the heuristic function will 282
decrease, expanding the search range and preventing the occurrence of locally optimal 283
solutions. The optimized A* algorithm can adaptively adjust the weight of the heuristic 284
function according to the number of obstacles in the environment and the distance be- 285
tween the current nodes and the ending points so as to improve the search efficiency and 286
sensitivity of the algorithm in the environment. 287
2.2.3. Global Path Optimization Strategy 288
Due to the global path planned by A* algorithm, with multiple nodes and multiple 289
turns, the vehicle needs to change its course many times in the driving process, which is 290
not conducive to the path tracking of the vehicle [33]. Therefore, this paper proposed a 291
strategy to delete the redundant points in the global path of A* and retain the key points 292
to reduce the turning point of the line. The strategy is shown in Figure 1, where the path 293
6point set is V , n1, n2 , n3 , n4 , n5 , n6 , n7 , n8 , n9 ,U , key point selection of the global path is per- 294
formed, with the following steps:Determine whether the point V , n1, n2 is collinear; if the 295
three points are collinear, it means that n1 is a redundant point and needs to be deleted, 296
then the local path becomes V ,n2  . 297
298

299

300

301

302

303

304

(a) (b) 305

Figure 1. Global path smoothing strategy: (a) is the global path points of the traditional A *, 306
and (b) is the global path points of the A * after the Key point selection strategy is adopted. 307

the three points are not collinear, then it is necessary to judge whether there are ob- 308
stacles between V ,n2  ; if there are obstacles between them, then n1 is a key point, and 309
the path needs to be retained, then the path becomes V , n1, n2 ; if there is no obstacle be- 310
tween the point connections, then it means that n1 is a redundant point and needs to be 311
eliminated, then the path becomes V ,n2  . 312
(1)Determine whether the point is collinear; if the three points are collinear, it 313
means that is a redundant point and needs to be deleted, then the local path becomes . If 314
the three points are not collinear, then it is necessary to judge whether there are obstacles 315
between ; if there are obstacles between them, then is a key point, and the path needs to 316
Electronics 2023, 12, x FOR PEER REVIEW 8 of 25

be retained, then the path becomes ; if there is no obstacle between the point connections, 317
then it means that is a redundant point and needs to be eliminated, then the path becomes. 318
(2)And so on, finally backtracking through all critical path points, and the opti- 319
mized global path point set is , as shown in Figure 1(b). 320
After taking the above optimization measures for the A* algorithm, simulation ex- 321
periments were conducted using Matlab to verify its feasibility. The simulation results of 322
the improved A* algorithm are shown in Figure 2, where green is the starting point, and 323
yellow is the ending point. 324

(a) (b) (c)


Figure 2. Matlab simulation results of the improved A * algorithm: (a) is the path of the traditional 325
A*, in Figure. 2 (b), the blue line is the global path after adopting the above path optimization strat- 326
egy, and in (c), the green line is the global path of the improved A * algorithm. 327

The results are shown in Table 2 compared with the traditional A* algorithm, the 328
running time of the improved A* algorithm was shortened by 23.38%, and the path length 329
was reduced by 27.27%. After adopting the key point selection strategy, the number of 330
points of global path planning by the improved A* algorithm was also reduced by 57.78%. 331
From the simulation experiment analysis, on the basis of ensuring efficient and accurate 332
global path planning, the improved A* algorithm greatly saves time and shortens the 333
length used in planning the path. Therefore, this paper considers adopting the improved 334
A* algorithm as a global planning algorithm. 335

Table 2. Improve before and after A* algorithm matlab simulation results 336

Path Length Number of


Algorithm Time(s)
(cm) points

Tradition A* 3.649 44 45

Improved A* 2.796 32 19

2.3. Experimental Validation of Improved A* Algorithm 337


To verify the applicability and universality of the improved A* algorithm in each 338
scenario, this section sets up several navigation scenarios to verify the improved A* algo- 339
rithm, and at the same time compares the improved A* and bidirectional A*, to prove the 340
general applicability of the improved A* algorithm. 341
In this section, Matlab is used to conduct experiments, constructing multiple complex 342
navigation scenarios and conducting comparative experiments. The experimental data are 343
strictly evaluated and compared from the above evaluation indicators of "planning time, 344
number of nodes, path length, and success rate", so as to prove the superior performance 345
of improved A* algorithm. The results of the path planning experiment in the scene are 346
shown in Figure 3, Figure 4 and Figure 5. 347
Electronics 2023, 12, x FOR PEER REVIEW 9 of 25

This section designs three complex scenarios for path planning experiments, de- 348
scribed as follows:The map size is 20×20, the black grid represents the obstacle, the white 349
grid represents the available regions, the coordinates of the start point S are (0,0), and the 350
coordinates of end point G are (19,19). 351
(1) the number of obstacles generated is 100. 352
353

354

355

356

357

358

359

(a) (b) 360

Figure 3. It represents the path planning experiment of two A* algorithms, where (a) represents the 361
path planning results of the bidirectional A* algorithm. (b) represents the result of path planning 362
for improving the A* algorithm. 363

It indicates the results of the path planning experiment in a navigation map with a 364
randomly generated obstacle number of 100. 365
(2) the number of obstacles generated is 150. 366

367

368

369

370

371

372

(a) (b) 373

Figure 4. It represents the path planning experiment of two A* algorithms, where (a) represents the 374
path planning results of the bidirectional A* algorithm. (b) represents the result of path planning 375
for improving the A* algorithm. 376

It shows the results of the path planning experiment in a navigation map with a ran- 377
domly generated obstacle number of 150. 378
(3) the number of obstacles generated is 180. 379
Electronics 2023, 12, x FOR PEER REVIEW 10 of 25

(a) (b)
Figure 5. It represents the path planning experiment of two A* algorithms, where (a) represents the 380
path planning results of the bidirectional A* algorithm. (b) represents the result of path planning 381
for improving the A* algorithm. 382

It indicates the results of the path planning experiment in a navigation map with a 383
randomly generated obstacle number of 180. 384
According to the experimental data obtained from the above experiments, the results 385
are shown below in Figure 6, Figure 7 and Figure 8. 386

Figure 6. The planning time of two global path planning algorithms for global path planning in 387
three different complexity environments. 388

In a comparative analysis of the running time of the path planning algorithm, it can 389
be clearly seen in Figure 6 that the planning time of the improved A* algorithm over the 390
bidirectional A* algorithm is reduced by 14.55%, 10.78% and 10.01%, respectively. It 391
shows that the improved A* algorithm proposed in the paper also has obvious advantages 392
in planning time in complex navigation environments. 393
Secondly, the experimental results of the number of nodes of the two A* algorithms 394
in the above three complex navigation environments are compared and analyzed, as 395
shown in Figure 7. 396
Electronics 2023, 12, x FOR PEER REVIEW 11 of 25

Figure 7. The number of nodes searched by the two global path planning algorithms in three differ- 397
ent complexity environments. 398

In the navigation scenario with the number of obstacles of 100, the improved A* al- 399
gorithm decreased by 20.83% over the bidirectional A* algorithm. With the number of 400
obstacles of 150, the improved A* algorithm decreased by 26.67% over the bidirectional 401
A* algorithm. In the navigation scenario with the number of 180, the improved A* algo- 402
rithm decreased by 19.44% compared with the bidirectional A* algorithm. 403
Then, a comparison of the global path length of the two algorithms is shown in Figure 404
8. 405

Figure 8. The path length of the two global path planning algorithms in three different complexity 406
environment. 407

According to the searched global path length in Figure 8, in the three different com- 408
plexity navigation scenarios, the path length of the improved A* algorithm over the bidi- 409
rectional A* algorithm is reduced by 8.57%, 8.82% and 8.70%, respectively. It shows that 410
the improved A* algorithm proposed in complex navigation environment is significantly 411
reduced in the path length. 412
Electronics 2023, 12, x FOR PEER REVIEW 12 of 25

Figure 9. The success rate of two global path planning algorithms in three different complexity en- 413
vironments. 414

Finally, multiple global path planning experiments were conducted in three different 415
complexity navigation scenarios, recording the success times of experiments. The experi- 416
mental data are shown in Figure 9. The success rate of improved the A* algorithm is gen- 417
erally higher than that of the bidirectional A* algorithm, indicating that the success rate 418
of improved the A* algorithm in multiple complex navigation environments is higher. 419
According to the above experimental data analysis, the improved A* algorithm pro- 420
posed in this paper is better adapted to global search in different complex environments, 421
with higher search efficiency, higher success rate, better robustness, lower computational 422
capacity, and the global path is smoother. 423

3. Local Path Planning Based on Improved DWA Algorithm 424

3.1. DWA Algorithm 425


DWA is a local path planning algorithm that predicts the motion trajectory of the 426
vehicle in a certain period of time through multiple groups of sampling speeds. The eval- 427
uation function is used to score these prediction trajectories, and the speed corresponding 428
to the optimal trajectory is selected to drive the vehicle forward so as to realize local dy- 429
namic obstacle avoidance [34]. 430
3.1.1. Kinematic Model of Vehicle 431
DWA is to determine the direction and speed of the next moment according to the 432
predicted trajectory, so it is very significant to determine the kinematic model of the vehi- 433
cle [35]. In this paper, the unmanned vehicle with four-wheel independent drive chassis 434
is selected as the research object. 435
The incomplete kinematic model of the vehicle is shown in Figure 10, where assum- 436
ing that regardless considering the movement of the vehicle in the Z axis direction, only 437
the movement of the XY horizontal plane is considered. When the unmanned vehicle is 438
transformed into a two-dimensional plane space, the body and suspension system are re- 439
garded as rigid because the working route of the trackless rubber-tyred vehicle is fixed, 440
and the vehicle is driving in the roadway at low speed and uniform speed, Therefore, it is 441
considered that the vehicle will not occur lateral sliding. Consider only forward and rota- 442
tion, not longitudinal movement [36]. The kinematic model of the vehicle is as follows in 443
Equation (5): 444
Electronics 2023, 12, x FOR PEER REVIEW 13 of 25

•
X = v cos
•
Y = v sin  (5)
 • v tan f
 =
 L
where v is the speed of the vehicle, ( x, y ) is the position coordinate information of the 445
vehicle, L is the wheelbase of the vehicle, and  is the current heading angle of the vehi- 446
cle. 447

Figure 10. A kinematic model of the vehicle. 448

3.1.2. Speed Sampling 449


• Vehicle speed constraints 450
The sampling speed of the vehicle shall be controlled within the optimal angular 451
speed and linear speed range of the trackless rubber-tyred vehicle [37]. The constraint is 452
shown in Equation (6): 453

vt = (v,  ) vmin , vmax ,   min , max  (6)

• Vehicle motor increase and decrease speed constraint 454


The sampling speed of the vehicle shall take into account its own motor performance; 455
the speed is constrained by the maximum addition and deceleration speed under the cur- 456
rent speed [38]. The variable range of the speed is shown in Equation (7): 457

 v  vc − a ' max t , vc + amax t , 


vm = (v,  )
    c − a ' max t ,  + a
c  max t 

(7)

where vc is the linear speed of the vehicle, c is the angular speed of the vehicle, and 458
am ax 、 a'max are maximum addition and deceleration of the vehicle, a max 、 a' max are 459
maximum acceleration of the vehicle, t is time predicted by the simulation. 460
• Vehicle safe distance constraints 461
In order to ensure the safety of the vehicle, the speed and angular speed of the vehicle 462
reaching the obstacle front are reduced to zero under the constraint of maximum deceler- 463
ation [39]. The formula is shown in Equation (8): 464


 v  2dist(v,  )a'max  
vs = (v,  ) 
( )
(8)

   2dist v,  a ' max 

where dist(v, )represents the shortest distance between the trajectory and the obstacle at 465
this speed. 466
Electronics 2023, 12, x FOR PEER REVIEW 14 of 25

The speed constraint of the DWA is the intersection of the above three constraints, 467
and v is dynamic window speed, which satisfies the following Equation (9): 468

vr  vt  vm  vs (9)

• Track Simulation 469


Based on the above kinematic equation of the vehicle, assuming that the moving ve- 470
hicle moves at a constant speed between adjacent moments and the time interval t is 471
very short, the approximate motion of the vehicle from t time to t + t time can be con- 472
sidered as a straight line [40], so that the track calculation formula of the vehicle can be 473
obtained as shown in Equation (10): 474

 xt + t = xt + vt cost

 yt + t = yt + vt sint (10)
t + t = t + t

where v and  are the speed and angular speed of the vehicle, t is the heading angle 475
of the vehicle at time t . The sampling trajectory of the DWA algorithm in the search space 476
can be calculated, as shown in Figure 11. 477

Figure 11. Sampling trajectory simulation. 478

3.1.3. Evaluation Function of the Traditional DWA 479


After the speed sampling, multiple sets of feasible vehicle simulation trajectories will 480
be obtained, so it is necessary to use the evaluation function to find the optimal trajectory 481
of the current vehicle, The evaluation function of the traditional DWA algorithm is shown 482
in Equation (11) [41]: 483

G(v x , v y ,  ) =  head(v,  ) + dist(v,  )


+ vel(v,  )  (11)

where head(v,  ) represents the azimuth deviation of the vehicle attitude at the end of the 484
simulated trajectory and the ending point, dist (v,  ) indicates the distance between the 485
end of the simulated trajectory and the nearest obstacle, vel (v,  ) expresses the linear 486
speed corresponding to the simulated trajectory, and the faster the vehicle is, the larger 487
the value of the evaluation function. ,, are the weighting coefficient of each sub- 488
function of the angle, distance, and velocity.  is the normalized parameter. 489

3.2. Evaluation Function Optimization of DWA 490


First, in order to solve the problem of vehicles rapidly avoiding unknown dynamic 491
obstacles in complex environments, it is necessary to optimize the subfunction dist (v,  ) 492
Electronics 2023, 12, x FOR PEER REVIEW 15 of 25

representing the distance between the end of the simulated trajectory and the nearest ob- 493
stacle in the traditional DWA evaluation function. A threshold is set for it, which repre- 494
sents the distance between the end of the simulated trajectory and the obstacle within the 495
"nearest range". If there is an obstacle in the "nearest range", it indicates that the simulated 496
trajectory is relatively dangerous; otherwise, if there is no obstacle in the "nearest range", 497
it is relatively safe. Meanwhile, to prevent the weight of this function from being too large, 498
a maximum threshold D is set to show that if there is no obstacle in the "nearest range" or 499
the nearest distance dist (v,  ) is greater than the set value D, then the weight value is set 500
to constant D. 501
Secondly, since the traditional DWA algorithm only has the guidance of one target 502
point, the target may be inaccessible or even trapped in the local optimal. In order to pre- 503
vent the occurrence of the above phenomena, a new evaluation subfunction is introduced. 504
It is expressed as the distance between the end of the simulated trajectory corresponding 505
to each sampling speed and the next local target point, guiding the vehicle in the direction 506
of the local target point. For the above two improvements, the evaluation function of the 507
optimized DWA algorithm is shown in Equation (12): 508

G(v x , v y ,  ) =  head(v,  ) + dist(v,  ) (12)


+ guid(v,  ) + vel(v,  ) 
where dist (v,  ) is the distance between the end of the simulated trajectory and the obsta- 509
cle in the "nearest range", guid (v,  ) represents the distance of the end of the simulated 510
trajectory from the next local target point, ,,, are the weighting coefficient of 511
each subfunction of the angle, distance, and velocity.  is the normalized parameter. 512

3.3. Fusion Algorithm Based on Improved A* and DWA 513


The improved A* global path planning algorithm can realize the adaptive adjustment 514
of the algorithm with the change of obstacles in the environment so as to improve the 515
sensitivity of the A* algorithm and improve the efficiency of global path search. At the 516
same time, in view of the problem of multiple nodes and multiple turns in the global path 517
planned by the A* algorithm, it is considered to take to eliminate the redundant points on 518
the global path, and only retain the path smoothing strategy of the key points to optimize 519
the global path. In this paper, the global path information is added to the evaluation func- 520
tion to locally guide the vehicle. By introducing the key points in the optimized A* algo- 521
rithm into the DWA algorithm, the key point of the global path planning is taken as the 522
local target point of the DWA algorithm, the dynamic setting of the target point is realized, 523
and multiple sets of feasible simulated vehicle trajectories are obtained by speed sampling 524
through the established vehicle kinematics model, and then the optimal speed trajectory 525
of the vehicle is found by using the optimized evaluation function to guide the vehicle 526
towards the target point. Thus, the fusion of the two improved algorithms is realized. The 527
flowchart of the fusion design of the algorithm is shown in Figure 12. 528
Electronics 2023, 12, x FOR PEER REVIEW 16 of 25

Figure 12. Flowchart of the fusion algorithm. 529

4. Results 530

4.1. Experimental Environment and Equipment 531


This paper experiment using the corridor as mine roadway simulation experiment, 532
as shown in Figure 13, as a long straight corridor and roadway road structure similar, 533
conform to the roadway structure of single and less characteristic conditions, so can be 534
used to do test site simulation mine roadway again based on ROS intelligent vehicle plat- 535
form to carry out the path planning experiment, to verify the reliability of the method 536
proposed in this paper. 537

(a) (b)

Figure 13. It shows the long straight corridor as the experimental scene. (a) is the first half of the 538
long straight corridor,and (b) is the second half of the corridor at the corner. 539

In order to avoid the problem of scene degradation in the construction and localiza- 540
tion of the vehicle, leading to the failure of the construction and localization, and affect 541
Electronics 2023, 12, x FOR PEER REVIEW 17 of 25

the subsequent path planning effect. In this paper, the fusion localization method of LI- 542
DAR and IMU is used to realize the localization of the vehicle, and the GeLO-SLAM algo- 543
rithm is used to build a 3D point cloud map on the corridor, as shown in Figure 14. 544

Figure 14. Map of the corridor simulated roadway. 545

The intelligent vehicle experimental platform used in this paper, which includes 546
three parts: chassis system, sensor system, and information processing system, is as shown 547
in Figure 15. includes three parts: chassis system, sensor system, and information pro- 548
cessing system. 549
550

551

552

553

554

555

556

557

558

Figure 15. Autonomous vehicle platform. 559

The chassis system has four-wheel steering independent drive, which can achieve a 560
maximum speed of 40 km/h, and can meet the experimental requirements of autonomous 561
localization and navigation system. The IPC is equipped with the ROS operating system 562
that integrates the lidar data obtained by the sensor system with the odometer information 563
to build a 3D point cloud map [42]. At the same time, the information processing system 564
is operated by an embedded industrial control machine, which processes the data from 565
the sensor system, carries out autonomous localization, map construction, and autono- 566
mous navigation algorithm, and communicates with the chassis of the vehicle, and then 567
the chassis issues corresponding instructions to realize the autonomous localization and 568
navigation of the vehicle. 569
Electronics 2023, 12, x FOR PEER REVIEW 18 of 25

The obstacles are set up in the experimental scene to simulate the real roadway scene 570
and make the final preparation before the experiment. As shown in Figure 16. Static ob- 571
stacles are set in the corridor as known static obstacles in the path planning experiment, 572
and the ones set at the corner represent unknown dynamic obstacles. The relevant param- 573
eters of the experimental unmanned vehicles are shown in Table 3. 574
575

576

577

578

579

580

581

(a) (b) 582

Figure 16. It shows the preparation of setting up obstacles in the corridor for the path planning 583
experiment. Among them, (a) is the setting of a static-known obstacle,and (b) is unknown dynamic 584
obstacles set at the corner position. 585

Table 3. Vehicle body structure parameters. 586

Name Parameters
Shape 2490×1550×616mm
Bearing spacing 1900mm
Wheel spacing 1355mm
Maximum speed 40km/h
Steering type Four-wheel steering
Braking type Four-wheel disc brake
And the experimental hardware equipment information is shown in Table 4. 587

Table 4. Information about experimental equipment. 588

Equipment Model
CPU i7-9700
Computer Graphics card
RTX3060
LIDAR velodyne VLP-16
IMU LPMS-IG1

4.2. Global Path Planning Experiment Based on Improved A* Algorithm 589


The corridor is similar to the road structure of the mine roadway and was selected 590
for the path planning simulation experiment to verify the effectiveness and accuracy of 591
the improved fusion algorithm. First, the global path planning experiment was conducted 592
on the constructed environmental map. The point cloud area in the figure represents the 593
planning area, the red box indicates the obstacle, the starting point is A, and the ending 594
point is B. This experiment set up two sets of experiments: (1) path planning in the envi- 595
ronment without obstacles. (2) path planning in the static obstacle environment. The re- 596
sults are shown in Figure 17 and Figure 18. 597
Electronics 2023, 12, x FOR PEER REVIEW 19 of 25

(a) (b)
Figure 17. It shows the path planning experiments in an obstacle-free environment. Among them, 598
(a) is the experimental result of the traditional A* algorithm, and (b) is the experimental result of the 599
path planning of the improved A* algorithm. 600

(a) (b)
Figure 18. It shows a path planning experiment in an environment with static obstacles. Where (a) 601
is the experimental result of the traditional A* algorithm, and (b) is the experimental result of the 602
path planning of the improved A* algorithm. 603

Table 5. Improved A* algorithm for path planning experimental data 604

Path length Vehicle driving Average velocity


Algorithm Time (s)
(m) time (s) (m/s)
Tradition
2.375 48.678 111 0.408
No A*
obstacles Improved
2.179 37.547 92 0.410
A*
Tradition
2.486 48.654 115 0.443
Static A*
obstacles Improved
2.249 37.819 102 0.394
A*
And the experimental data are shown in Table 5. From the analysis of the experi- 605
mental results, during the path planning in the same obstacle-free environmental map, 606
the improved A* algorithm reduced the path planning time by 8.25%, the path length by 607
22.87%, and the vehicle driving time by 17.17%. In the static obstacle environment, the 608
time of path planning was reduced by 9.53%, the length of the path was shortened by 609
22.27%, and the vehicle driving time was reduced by 11.30%. Compared to the traditional 610
A* algorithm, the path planning of improved A* was reduced in both search time and 611
Electronics 2023, 12, x FOR PEER REVIEW 20 of 25

length in two different environmental maps. As can be seen from the figure, the path plan- 612
ning by the improved A* algorithm was less than the path turning point planned by the 613
traditional A* algorithm, with a smaller vehicle angle and smoother path, which is con- 614
ducive to the tracking of the vehicle and reduced the driving time of the vehicle. The ex- 615
periments proved that the improved A* algorithm has higher flexibility and faster search 616
efficiency. 617

4.3. Fusion Algorithm Experiment 618


In this experiment, we set unknown dynamic obstacles in the actual scene and used 619
the improved A* algorithm and the optimized fusion algorithm in the environmental map 620
so as to verify the effectiveness of the improved fusion algorithm. The experimental re- 621
sults are shown in Figure 19 and Figure 20. 622

(a) (b)
Figure 19. The improved A* algorithm with unknown dynamic obstacles. 623

(a) (b)
Figure 20. The improved fusion algorithm for unknown dynamic obstacles. 624

Table 6. improved A*algorithm and improved fusion algorithm path planning experimental data. 625

Vehicle
Time Path length Average ve-
driving
(s) (m) locity (m/s)
time (s)
Improved A* 2.323 53.747 110 0.409
Improved fusion
2.198 40.189 95 0.439
algorithm
Electronics 2023, 12, x FOR PEER REVIEW 21 of 25

The experimental data is shown in Table 6. Compared with the single improved A* 626
algorithm, the improved fusion algorithm reduced the path planning time by 5.38%, re- 627
duced the length of the path by 25.23%, and the vehicle driving time by 13.64%. It can 628
be seen that the path length planned by the improved fusion algorithm is shortened, and 629
the time of the algorithm planning is reduced, indicating that the efficiency of the algo- 630
rithm has been improved, and the ability of the algorithm to avoid obstacles in the un- 631
known dynamic environment has been improved. 632
The vehicle data in the experiment are shown in Figure 21 and Figure 22. 633

(a) (b)
Figure 21. (a), (b) are the heading angle and heading angle velocity of the vehicle while driving 634
under the A * algorithm. 635

(a) (b)
636

Figure 22. (a), (b) are the heading angle and heading angle velocity of the vehicle during driving 637
under the optimized and improved fusion algorithm. 638

It can be seen from the driving state of the vehicle in the Figure 19 and Figure 20 that 639
compared with the single improved A* algorithm, the heading angle steering of the plan- 640
ning path vehicle tracking is stable, and the heading angular velocity of the vehicle 641
changes little, which proves that the planned path conforms to the steering constraint of 642
the vehicle, so the vehicle has good stability when tracking the path. 643

4.4. Comparison of Experimental Results Between Proposed Algorithm and Similar Algorithm 644
The content described in this section is to compare the proposed algorithm with the 645
existing A*+APF similar algorithm, and strictly compare and evaluate the performance of 646
the fusion algorithm, to prove that the proposed improved fusion algorithm in the 647
static/dynamic environment in the path planning ability has been significantly improved. 648
This section sets up two groups of path planning experiments. The first group: the 649
experiment in the barrier-free navigation scenario, and the second group: the experiment 650
Electronics 2023, 12, x FOR PEER REVIEW 22 of 25

in the static/dynamic unknown obstacle navigation scenario. Experiments on the two al- 651
gorithms mentioned above, and the results are shown in Figure 23, where the yellow line 652
represents the A*+APF similar algorithm and the green line indicates the A*+DWA im- 653
proved fusion algorithm proposed in this paper. 654

(a) (b)
Figure 23. It shows the results of the proposed improved algorithm nd the existing similar algorithm 655
under two environmental conditions. Among them, (a) shows the results of separate path planning 656
for the two algorithms in a simple environment, and (b) shows the results of path planning of two 657
algorithms in complex environments. 658

The experimental data from the experimental results are shown in Table 7. 659

Table 7. Experimental results of the proposed algorithm with a similar algorithm. 660

A*+APF A*+DWA
Average Average Average Average
Experimental Path Path
Suc- vehicle algorithm Success vehicle algorithm
scene length length
cess rate driving running rate driving running
(m) (m)
time (s) time (s) time (s) time (s)
Simple envi-
98.4% 90 2.714 38.819 98.7% 96 2.201 35.645
ronment

Complex en-
86.7% 120 3.512 50.357 97.1% 107 2.461 43.189
vironment
It can be seen that in the simple general environment, the proposed path planning 661
method and the existing similar algorithms are not much different. However, with the 662
complexity of the environment, the success rate of the present algorithm is significantly 663
higher than the compared similar algorithm. The improved algorithm under complex con- 664
ditions has a 10.71% higher success rate than the existing similar algorithms. In the simple 665
navigation scenario, the A*+DWA algorithm increases the average driving time of the ve- 666
hicle by 6s over the A*+APF algorithm planning, but the driving time in the complex en- 667
vironment is reduced by 13s, indicating that in the more 668
complex environment, the improved fusion algorithm is better. Compared with 669
A*+APF, the average running time of A*+DWA algorithm is reduced by 0.513s in simple 670
environment and 1.051s in complex environment, that is, from 18.9% in simple environ- 671
ment to 29.93% in complex environment. It shows that the algorithm has improved the 672
computational efficiency in complex environments. The path length of the above two al- 673
gorithms also varies from 3.174m in simple environment to 7.168m in complex environ- 674
ment. 675
Electronics 2023, 12, x FOR PEER REVIEW 23 of 25

In conclusion, as the environment complexity changes, the proposed improved algo- 676
rithm has better planning efficiency, more adaptable to the environment, more robust, 677
and lower computational complexity. As the environmental complexity increases, alt- 678
hough, the improved algorithm performance decreases, the overall decrease is smaller 679
than the existing similar algorithms. 680

5. Discussion 681
In this paper, in order to solve the issue of path planning of Autonomous trackless 682
rubber-tyred vehicles under mines, we proposed a fusion path planning algorithm based 683
on improved A* and DWA algorithms. We optimized the heuristic function of the A* al- 684
gorithm and adopted the global path optimization strategy, and then modified and opti- 685
mized the evaluation function of the DWA algorithm, and finally integrated the two im- 686
proved global and local path planning algorithms. To validate the improved fusion algo- 687
rithm, the experiments were performed in two real-world environments. The results show 688
that the improved fusion algorithm reduces the time of path planning by 5.38%, reduces 689
the path 690
length by 25.23%, and the vehicle run by 13.64%. It is proved that the improved fu- 691
sion algorithm has a much smoother path, more efficient path search, and more rapid 692
avoidance of dynamic obstacles. The paper also compares experiments with similar algo- 693
rithms, and the results show that the proposed improved algorithm has better adaptabil- 694
ity in the more complex environment, the path is more accurate and stable, and the navi- 695
gation efficiency has also been greatly improved. 696
The selection of the experimental site and the complex environment under the mine 697
is not fit, and the complexity of obstacles is too low, at the same time, without considering 698
the influence of dust, gas, light, and road friction; all these may have some impact on the 699
construction of maps, the positioning of vehicles and the results of path planning algo- 700
rithm, which cannot guarantee the universality of experimental data. Therefore, this 701
method also needs to be verified in the real mine roadway. 702

Author Contributions: Conceptualization, C.Z. and X.Y.; methodology, C.Z. and X.Y.; software, 703
R.Z.; validation, X.Y, Z.G. and P.Q.; formal analysis, X.Y.; investigation, C.Z.; resources, C.Z.; data 704
curation, R.Z.; writing—original draft preparation, X.Y.; writing—review and editing, X.Y. and R.Z.; 705
visualization, C.Z.; supervision, C.Z.; project administration, C.Z.; funding acquisition, C.Z. All au- 706
thors have read and agreed to the published version of the manuscript. 707

Funding: This paper was funded by the Shaanxi Innovation Talent Promotion Plan - Science and technology 708
innovation team (2021TD-27) and the 2022 Youth Innovation Team Construction Scientific Research Program of 709
Shaanxi Provincial Education Department (22JP045) 710

Data Availability Statement: Not applicable. 711

Conflicts of Interest: The authors declare no conflict of interest. 712

References 713
1. Hao, X.Y.; Shuai, L. The Function Design for the Communication-Based Train Control (CBTC) System: How to Solve the Prob- 714
lems in the Underground Mine Rail Transportation. Applied System Innovation, 2021, 4,2, 31-31. [CrossRef] 715
2. Xi, L.M.; Rui, Q.M. Path planning for coal mine robot to avoid obstacle in gas distribution area. International Journal of Advanced 716
Robotic Systems, 15. [CrossRef] 717
3. Ahmet,Gunes, Y., and Celal, K. Shortest path optimization of haul road design in underground mines using an evolutionary 718
algorithm. Applied Soft Computing Journal, 2019, 83, 105668-105668. [CrossRef] 719
4. Miao, C. Application of mine explosion-proof trackless rubber wheel truck in mine auxiliary transportation. Equipment manage- 720
ment and maintenance, 2018, 111-112. [CrossRef] 721
5. Yu, X.W. et al, Reform and development of coal mine safety in China: An analysis from government supervision, technical 722
equipment, and miner education. Resources Policy, 2022, 77. [CrossRef] 723
6. Mao, S.W.; Jiu, S.B. et al, Research Status and Development Trend of Unmanned Driving Technology in Coal Mine Transporta- 724
tion. Energies, 2022, 15, 9133-9133. [CrossRef] 725
7. Ming, Y.; Yangqiang, L. et al, A Survey of Path Planning Algorithms for Autonomous Vehicles. SAE International Journal of 726
Commercial Vehicles, 2021, 14, 97-109. [CrossRef] 727
Electronics 2023, 12, x FOR PEER REVIEW 24 of 25

8. Xiao, B.Z.; Xiang, Y. et al, Trajectory Planning and Tracking Strategy Applied to an Unmanned Ground Vehicle in the Presence 728
of Obstacles. IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, 2021, 18, 1575-1589. [CrossRef] 729
9. Thi Thao,M.; Cosmin, C. et al, "Heuristic approaches in robot path planning: A survey," Robotics and Autonomous Systems, 2016, 730
86, 13-28. [CrossRef] 731
10. Lun, Q.; Lun, X.H. et al, Survey of UAV motion planning. IET Cyber‐Systems and Robotics, 2020, 2, 14-21. [CrossRef] 732
11. Zen, L.W.; et al, Improved A* algorithm and model predictive control- based path planning and tracking framework for hexa- 733
pod robots. The Industrial Robot, 2023, 50, 135-144. [CrossRef] 734
12. Ade, C.; Mohammad, A.B.; and Hartanto, K. Dijkstra's and A-Star in Finding the Shortest Path: a Tutorial. 2020 International 735
Conference on Data Science, Artificial Intelligence, and Business Analytics (DATABIA), 2020, 28-32. [CrossRef] 736
13. SM, L. Rapidly-exploring random trees: A new tool for path planning. 1998. URL http://citeseerx. ist. psu. edu/viewdoc/sum- 737
mary, [CrossRef] 738
14. Farid, B.; Denis, G.; Herve, P.; Dominque, G. Modified artificial potential field method for online path planning applications," 739
2017 IEEE Intelligent Vehicles Symposium (IV), 2017, 180-185. [CrossRef] 740
15. Chen, X.G.; Shuai, Y.X. Dynamic Path Planning Based on Improved A* and DWA Algorithms. 2022 4th International Conference 741
on Control and Robotics (ICCR), 2022. [CrossRef] 742
16. Zhi, W.X.; Xin, X.; et al. A federated learning system with enhanced feature extraction for human activity recognition. Knowledge- 743
Based Systems, 2021, 229, 107338. [CrossRef] 744
17. Huan, L.X.; Zhi, W.X.; et al. An efficient federated distillation learning system for multitask time series classification. IEEE 745
Transactions on Instrumentation and Measurement, Aut. 2022, 71, 1-12. [CrossRef] 746
18. Dian, R., Lysander, Gustin. Analysis of Dijkstra’s algorithm and A* algorithm in shortest path problem. Journal of Physics: Con- 747
ference Series, 2020, 1566, 26-27. [CrossRef] 748
19. Hasan, A.H.; Sadiq, A. Robot Path Planning Based on Hybrid Improved D* with Particle Swarm Optimization Algorithms in 749
Dynamic Environment. Journal of Computational and Theoretical Nanoscience, 2019, 16, 1062-1073. [CrossRef] 750
20. Xin, G.M.; Xi, Q.M. Robot dynamic path programming incorporating improved RRT and Dijkstra algorithms. Modular Machine 751
Tool & Automatic Manufacturing Technique, 2023, 64, 5-9. [CrossRef] 752
21. Li, S.L.; Bin, W.; Hui, X. Research on Path-Planning Algorithm Integrating Optimization A-Star Algorithm and Artificial Poten- 753
tial Field Method. Electronics, 2022, 11, 3660-3660. [CrossRef] 754
22. Tian, Y.L.; Rui, X.Y.; Guang, R.W.; Lei, S. Local Path Planning Algorithm for Blind-guiding Robot Based on Improved DWA 755
Algorithm. Chinese Control And Decision Conference, 2019, 31, 871-875. [CrossRef] 756
23. Fu, H.S.; Hua, L.X.; Xin, H.W.; et al. Evolutionary Multi-Objective Reinforcement Learning Based Trajectory Control and Task 757
Offloading in UAV-Assisted Mobile Edge Computing. IEEE Transactions on Mobile Computing, 2022, vol. 71, pp. 1-18. [CrossRef] 758
24. Yardimci, A.G.; Karpuz, C. Shortest path optimization of haul road design in underground mines using an evolutionary algo- 759
rithm. Applied Soft Computing Journal, 2019, 83. [CrossRef] 760
25. Hong, J.Z. Research on the path planning of coal mine scene inspection robot. Coal Mine Machinery, 2022, 43, 45-47. [CrossRef] 761
26. Meng, J.L.; Xi, A.Z.; Zhan, G.W.; De,M.L. Research on the application of mine flood escape path based on bidirectional A * 762
algorithm. Coal Engineering, 2019, 51, 42-47. [CrossRef] 763
27. Yuan, B.H.; Li, B.H. Research on the autonomous obstacle avoidance method of coal mine rescue robot. Coal Science and Tech- 764
nology, 2011, 39, 90-92+21. [CrossRef] 765
28. Jiu, S.B.; Mu, M.Z.; Shi, R.G. Based on the improved A * and artificial potential field algorithm Path planning. Journal of China 766
Coal Society, 2022, 47, 1347-1360. [CrossRef] 767
29. Zi, J.T.; Xue, H.G. Path planning of mine navigation device based on improved artificial potential field. Journal of China Coal 768
Society, 2016,vol. 41, 589-597. [CrossRef] 769
30. Yuan, X.M.; , Ming, R.H. Research on the key technology of mine-assisted transportation robot. Industry and Mine Automation, 770
2020, 46, 8-14. [CrossRef] 771
31. Feng, S.; Xin, G. Research on the unmanned target path planning method for underground mining vehicles. Mining and Metal- 772
lurgy, 2018, 27, 87-91. [CrossRef] 773
32. Michael, C.L.; Xiangpeng, W.; et al. Leveraging Intelligent Transportation Systems and Smart Vehicles Using Crowdsourcing: 774
An Overview. Smart Cities, 2020, 3, 341-361. [CrossRef] 775
33. Rachma, W.D.; and Lysander, G. Analysis of Dijkstra’s Algorithm and A* Algorithm in Shortest Path Problem. Journal of Physics 776
Conference Series, 2020, 1566. [CrossRef] 777
34. Hao, J.W.; Xiang, H.M. Obstacle avoidance path planning of mobile robot based on improved DWA. Journal of Physics: Conference 778
Series, 2022, 2383. [CrossRef] 779
35. Xun, Y.Z.; Jun, T.; Huosheng, H.; and Xiafu, P. Hybrid Path Planning Based on Safe A* Algorithm and Adaptive Window 780
Approach for Mobile Robot in Large-Scale Dynamic Environment. Journal of Intelligent & Robotic Systems, 2020, 99, 1-13. [Cross- 781
Ref] 782
36. Hermosillo, J.; Pradalier, C.; & Sekhavat, S. Modeling odometry and uncertainty propagation for a bi-steerable car. Intelligent 783
Vehicle Symposium, 2002, 2, 401-406. [CrossRef] 784
37. Lanyong, Z.; Yu, H.; and Bo, J. Research on Path Planning Method of Unmanned Boat Based on Improved DWA Algorithm. 785
Journal of Sensors, 2022, 2022. [CrossRef] 786
Electronics 2023, 12, x FOR PEER REVIEW 25 of 25

38. Bin, W.; Xiaonan, C.; Congcong, Z. Dynamic Path Planning for Forklift AGV Based on Smoothing A* and Improved DWA 787
Hybrid Algorithm. Sensors, 2022, 22, 7079-7079. [CrossRef] 788
39. Zhen, Y.X.; and Wen, Y. Mobile robot path planning based on fusion of improved A* algorithm and adaptive DWA algorithm. 789
Journal of Physics: Conference Series, 2022, 2330. [CrossRef] 790
40. Xiong, B.; Haikun, J.; Junjie, C.; et al. UAV Path Planning Based on Improved A and DWA Algorithms. International Journal of 791
Aerospace Engineering, 2021. [CrossRef] 792
41. Jun, D.; Dongfang, L.; Junwei, Z.; Yangqin, L. Autonomous Navigation of Robots Based on the Improved Informed-RRT∗ Al- 793
gorithm and DWA. Journal of Robotics, 2022. [CrossRef] 794
42. Shaojiang, Z.; Yanning, G.; Qiang, Z.; Zhiyuan, L. Lidar-IMU and wheel odometer based autonomous vehicle localization sys- 795
tem. 2019 Chinese control and decision conference (CCDC), 2019, 4950-4955. [CrossRef] 796

You might also like