A Review On Carrier Aircraft Dispatch Path Planning and Control On Deck

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

CJA 1668 No.

of Pages 19
13 July 2020
Chinese Journal of Aeronautics, (2020), xxx(xx): xxx–xxx
1

Chinese Society of Aeronautics and Astronautics


& Beihang University
Chinese Journal of Aeronautics
cja@buaa.edu.cn
www.sciencedirect.com

3 A review on carrier aircraft dispatch path planning


4 and control on deck
5 Xinwei WANG a,b, Jie LIU c,*, Xichao SU d, Haijun PENG a, Xudong ZHAO b,
6 Chen LU e

a
7 Department of Engineering Mechanics, State Key Laboratory of Structural Analysis for Industrial Equipment, Dalian University
8 of Technology, Dalian 116024, China
b
9 Faculty of Electronic Information and Electrical Engineering, Dalian University of Technology, Dalian 116024, China
c
10 War Research Institute, Academy of Military Sciences, Beijing 100850, China
d
11 Department of Airborne Vehicle Engineering, Naval Aeronautical and Astronautical University, Yantai 264001, China
e
12 School of Reliability and Systems Engineering, Beihang University, Xueyuan Road, Haidian District, Beijing 100191, China

13 Received 7 April 2020; revised 28 May 2020; accepted 4 June 2020


14

16 KEYWORDS Abstract As an important part in sortie/recovery process, the dispatch of carrier aircraft not only
17
18 Carrier aircraft; affects the sortie/recovery efficiency and safety, but also has severe influence on the carrier’s combat
19 Dispatch; efficiency and the comprehensive support capability. Path planning is the key to improve the effi-
20 Kinematic model; ciency and safety during the dispatch process. The main purpose of this paper is to propose a com-
21 Path planning; prehensive investigation of techniques and research progress for the carrier aircraft’s dispatch path
22 Path tracking planning on the deck. Three different dispatch modes of carrier aircraft and the corresponding
modeling technologies are investigated, and the aircraft’s dispatch path planning techniques and
algorithms have been classified into different classes. Moreover, their assumptions and drawbacks
have been discussed for single aircraft and multiple aircraft. To make the research work more com-
prehensive, the corresponding tracking control methodologies are also discussed. Finally, due to the
similarity of path planning problem between the carrier aircraft’s dispatch and those in other fields,
this paper provides an exploratory prospect of the knowledge or method learned from other fields.
23 Ó 2020 Production and hosting by Elsevier Ltd. on behalf of Chinese Society of Aeronautics and
Astronautics. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).

1. Introduction 24

* Corresponding author.
As an important symbol of national power, the aircraft carrier 25
E-mail address: liuyexiaobao@163.com (J. LIU).
plays an irreplaceable role in maritime security and interests,1,2 26
Peer review under responsibility of Editorial Committee of CJA.
and carrier aircraft is the main combat weapon on it. In the 27
future, the number of aircraft carried by the carrier platform 28
will gradually increase. In addition, as the training/ combat 29

Production and hosting by Elsevier intensity increases, the activity of carrier aircraft will become 30

https://doi.org/10.1016/j.cja.2020.06.020
1000-9361 Ó 2020 Production and hosting by Elsevier Ltd. on behalf of Chinese Society of Aeronautics and Astronautics.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
2 X. WANG et al.

31 more and more frequent on the deck. The combat efficiency of other vehicles, in the process of dispatch, a collision-free taxi- 79
32 the entire aircraft carrier group largely depends on the sortie of ing path that connects the initial parking position and the des- 80
33 carrier aircraft, where dispatch is one of the key parts of it.3,4 tination is expected. And the generated path should be as 81
34 Compared with land-based aircraft operations, the space of smooth as possible while satisfying various constraints on state 82
35 the deck is much narrower.5 It is necessary to carry out many and control inputs. However, the dispatch path planning prob- 83
36 operations (as shown in Fig. 1) such as take-off, landing, and lem of carrier aircraft on the deck has distinct features. The 84
37 support, all with a dispatch space of 4.5 acres.6,7 However, the most notable one is that there is a high precision demand on 85
38 aircraft carrier deck is a completely open space, there is no the terminal posture to ensure success of the following catapult 86
39 fixed runway as land-based dispatch, and all idle parts of the and launch operation. As for the fleet launch process, the max- 87
40 deck can be used as a runway. Hence, the dispatch path plan- imum allowable number of carrier aircraft taxiing simultane- 88
41 ning on the deck is more complicated than land-based aircraft ously equals the number of catapult equipped on the flight 89
42 operations. There is a high risk in the transfer process of the deck. And there might be constraints on launch priority for 90
43 carrier aircraft on the deck. The flight deck incidents officially aircraft of different types due to the requirement of the certain 91
44 recorded by the U.S. navy include many serious injuries and mission. Last but not the least, considering that there is 92
45 deaths, as well as many instances of aircraft damage or loss. unavoidable base motion of the carrier under the effect of wind 93
46 According to accident reports from the Naval Safety Center, and wave loads, larger safety margins must be set in the path 94
47 there were 3228 related incidents from 1980 to 2008, and more planning problem. 95
48 than 30% of those happened in dispatch processes.8 Dispatch Having a good command of the dispatch process and dis- 96
49 accidents not only cause significant economic loss, more patch modes, as well as the operational environment on the 97
50 importantly, but also affect the carrier’s combat capability. flight deck and constraints encountered, is the prerequisite to 98
51 A collision between two aircraft on the flight deck may disrupt design high-quality dispatch paths. The dispatch path planning 99
52 maintenance and support plans for other aircraft, increasing can be generally divided into path planning and control for 100
53 the logistical pressure. And accident could lead to mission fail- single aircraft and coordinated path planning and control for 101
54 ure, whose impact may last for days or weeks. It can be con- multiple aircraft. There is few research of aircraft dispatch 102
55 cluded that a reasonable dispatch scheme is of great path planning problems or related model analysis. The existing 103
56 significance to ensure the operation safety of the carrier air- research generally focuses on taxiing path planning for single 104
57 craft on the deck. aircraft, and various constraints are not taken into considera- 105
58 The current methods to realize the dispatch for carrier air- tion. Hence, the obtained taxiing path is hardly directly 106
59 craft much depend on artificial expertise, which is feasible applied in practice. Besides, there is almost no relevant litera- 107
60 when the number of aircraft is relatively small.9 However, ture of the path tracking for the towed aircraft system with/ 108
61 when the number of dispatched aircraft is large, it is difficult without drawbar, and there is almost no concise and applica- 109
62 to plan a scientific and reasonable dispatch scheme based on ble kinematic or dynamic model. In the process of fleet dis- 110
63 manual experience. As a result, unreasonable dispatch plans patching, coordinated dispatch path planning is very 111
64 will be obtained in scenarios such as nights or other poor mete- common and important. However, studies on the path plan- 112
65 orological conditions.10 In addition, Unmanned Aerial Vehi- ning of multiple aircraft are rarely seen. Therefore, it’s mean- 113
66 cles (UAVs) have been gradually applied in the military field. ingful to establish the corresponding kinematic and dynamic 114
67 With the unique advantages, carrier UAVs will be used based model according to different dispatch modes, and generate 115
68 on the aircraft carrier platform in the future.3,11,12 Unlike optimal path and control command that satisfy all necessary 116
69 manned carrier aircraft, UAVs usually taxi by themselves, constraints so that the results can be directly applied to the 117
70 and it is true that the dispatch of UAVs is more automated practical dispatch processes. However, to the best of our 118
71 than that of manned carrier aircraft. Therefore, the research knowledge, there is no literature that summarizes the tech- 119
72 on carrier aircraft automatic transport planning and control niques for carrier aircraft dispatch path planning and control 120
73 technology can provide relevant technical reference for the car- on deck till now. Hence, this review will provide a comprehen- 121
74 rier UAVs. sive and broad view on every detail in the dispatch path plan- 122
75 There is rich review literature on path planning of vehicles ning and control on deck at length. 123
76 in other fields, such as mobile robots,13,14 UAVs,15 autono- The remainder of the paper is organized as follows. An 124
77 mous self-driving vehicles,16,17 and Autonomous Underwater overview of the dispatch of carrier aircraft on the deck is given 125
78 Vehicles (AUVs).18 Similar to the path planning problem of in Section 2. The kinematic model in three different dispatch 126

Fig. 1 An illustration of take-off and landing process on the deck.4

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
A review on carrier aircraft dispatch 3

127 modes are investigated and compared in Section 3. In Section 4, the research of path planning and control. It helps us to under- 163
128 the path planning techniques for single carrier aircraft are stand the dispatch environment and the path planning model 164
129 reviewed. In Section 5, the coordinated path planning tech- according to certain transport scenarios and tasks so that we 165
130 niques for multiple carrier aircraft are reviewed. In Section 6, can obtain accurate and practical results. 166
131 a review of trajectory tracking techniques used in the dispatch The dispatch of carrier aircraft on the deck is mainly 167
132 of carrier aircraft is given. In Section 7, several possible future divided into three modes: (A) autonomous taxiing (i.e., taxiing 168
133 research directions of this field are provided. Section 8 con- by the aircraft itself without a towing vehicle), (B) towed air- 169
134 cludes the paper. craft system without a drawbar, and (C) towed aircraft system 170
with a drawbar. Generally, the first and third types are mainly 171
135 2. An overview of the dispatch of carrier aircraft on the deck used for aircraft dispatch on the flight deck, and the second 172
type is mainly used for aircraft dispatch in the hangar. A brief 173

136 2.1. Dispatch process in sortie and recovery task for carrier comparison of the three dispatch modes is given in Table 1. As 174

137 aircraft the configurations of such three modes are different, their kine- 175
matics and control characteristics vary from each other. 176
Hence, it is meaningful to study the kinematics of three differ- 177
138 As illustrated in Fig. 2, the sortie process can be roughly
ent dispatch systems. 178
139 described as follows: Once the sortie task order is given, the
In towed modes, the tractor provides the power of the 179
140 corresponding aircraft will be dispatched from the hangar to
whole system. And it has three advantages over autonomous 180
141 the elevator with tractor. Next, the elevator will lift the aircraft
taxiing mode. First, it saves the fuel of the aircraft on the deck 181
142 from the hangar to flight deck, and the aircraft will be dis-
and thus improves the endurance capacity of the aircraft. Sec- 182
143 patched from the elevator to the corresponding gate position
ond, it avoids the engine of the aircraft switching frequently 183
144 with tractor. Then, the aircraft will taxi to the corresponding
between the on and the off state, improving lifetime of the 184
145 catapult from the gate position for ejection until the corre-
engine. Last but not the least, reversing motion of the aircraft 185
146 sponding supply and check work have been done. When the
can be implemented, improving the flexibility of dispatch. 186
147 combat mission is completed, the aircraft lands on the flight
148 deck and returns to the hangar, where the recovery process
2.3. Analogy of path planning between aircraft dispatch and 187
149 (the aircraft return to the hangar) is roughly the reverse of
unmanned ground systems 188
150 the sortie process. The dispatch efficiency and satisfaction of
151 collision-free conditions and other constraints are the core
152 issue in the corresponding path planning. Based on the above introduction, it’s found that the motion 189

153 It is seen that the aircraft dispatch takes large parts in the characteristics of three dispatch modes have much in common 190

154 sortie and recovery process. The transfer from the hangar to with those of some Unmanned Ground Systems (UGSs). For 191

155 the flight deck (or from the flight deck to the hangar) by eleva- aircraft taxiing by itself, front wheels provide the acceleration 192

156 tor is fixed, and the corresponding dispatch path is fixed too. and the steering operation is controlled by the nose wheel, 193

157 Hence, the dispatch path planning in this stage is not consid- which is alike to automobiles. The classical bicycle model 194

158 ered in this paper. widely used in trajectory planning for autonomous driving is 195
applicable to describe the motion of the carrier aircraft on 196

159 2.2. Three dispatch modes and corresponding application the deck. However, the autonomous taxiing aircraft cannot 197

160 scenarios realize backward motion by itself. As for towed aircraft sys- 198
tems (with/without a drawbar), their configurations are very 199
similar to tractor-trailer systems where the tractor provides 200
161 It is necessary for us to compare current dispatch modes and
the power of the whole system. Similar to tractor-trailer sys- 201
162 the corresponding application scenarios before conducting

Fig. 2 An illustration of dispatch process in sortie and recovery task for carrier aircraft.

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
4 X. WANG et al.

Table 1 Three dispatch modes and typical application scenarios.


Dispatch mode Typical application scenario Illustration

(1) In the process of sortie, the aircraft taxies from the gate position to the prepa-
ration spot or the catapult on the flight deck.
(2) In the process of recovery, the aircraft taxies for a while from the landing
Autonomous Taxiing
position on the flight deck.
(3) The aircraft accelerates from the catapult to takeoff (this process is fixed, and
there is no need to consider the path planning problem during this phase).

(1) In the process of sortie, the aircraft is dispatched from the gate position to the
Towed aircraft system elevator in hangar.
without drawbar (2) In the process of recovery, the aircraft is dispatched from the elevator to gate
position in hangar.

(1) In the process of sortie, the aircraft is dispatched from the elevator to the gate
position on the flight deck.
Towed aircraft system (2) In the process of recovery, the aircraft is dispatched to the gate position/ele-
with drawbar vator on the flight deck.
(3) In the process of maintenance, the aircraft will be dispatched between differ-
ent gate positions on the flight deck.

202 tems, the reversing motion is inherently unstable and the con- to avoid conflicts rather than only path planning. Besides, the 239
203 trol during this operation needs special care. starting time and arrival time of each aircraft can be different, 240
204 The research on path planning of UGSs is more extensive and each aircraft should strictly meet the corresponding termi- 241
205 than that of carrier aircraft, and some good ideas and methods nal constraints. This puts forward higher requirements for the 242
206 in these fields can be used as reference. Meanwhile, the path collaborative path planning algorithm. 243
207 planning problem of carrier aircraft on the deck has its unique
208 characteristics. Therefore, a brief comment is made on the dif- 3. Kinematic model of single carrier aircraft or traction system 244
209 ferences between carrier aircraft and these UGSs in path
210 planning. 3.1. A comparison between kinematic model and dynamic model 245
211 Generally speaking, the tractor-trailer systems or chained
212 robots usually focus on the whole system,19–22 and they seldom
The dispatch path planning of carrier aircraft is so delicate that 246
213 need to strictly satisfy a certain fixed terminal constraint (i.e.,
the aircraft should be modeled according to the dispatch char- 247
214 there is rarely such a scenario or requirement that they need to
acteristics, and it is hard to be simply reduced to a mass point. 248
215 be dispatched to a certain position accurately with a certain
Therefore, it is necessary to establish the corresponding dis- 249
216 orientation). So, they have a higher tolerance to satisfy the
patch dynamic or kinematic model that can describe the dis- 250
217 condition of terminal constraints. However, the purpose of
patch motion characteristics of aircraft. 251
218 towed aircraft systems (with/without drawbar) is to dispatch
Dispatch dynamic model of aircraft can be accurate to 252
219 the aircraft to the prescribed destination with a fixed terminal
show the behavior and motion characteristics of the dispatch 253
220 orientation, and the tractor will not get too much attention. As
system, the mechanical properties, and external force. Compli- 254
221 the deck space resources are so valuable and many supporting
cated factors, such as the deck surface turbulence and sag, and 255
222 facilities are fixed in the corresponding location, the aircraft
inertia force, can be accurately described.23,24 However, it will 256
223 need to strictly satisfy certain prescribed terminal constraints.
lead to a long time of numerical calculations. In addition, the 257
224 Therefore, compared to the tractor-trailer systems or chained
dynamic model involves many parameters, and it is usually 258
225 robots, the obstacles are denser.
necessary to identify the parameters of the dispatch system 259
226 Another difference between autonomous taxiing aircraft
and adopt multi-body modeling technology to obtain the accu- 260
227 and UGSs or robots mainly reflects in the collaborative path
rate dynamic model, which also brings inconvenience to the 261
228 planning problem. For the collaborative path planning of
application of dispatch dynamic model for carrier aircraft. 262
229 UGSs or robots, the initial or terminal time is identical for
The dispatch kinematic model of aircraft can accurately 263
230 all agents, and each agent has a different terminal position.
describe the motion characteristics of the system under the 264
231 However, the problem of coordinated taxi path planning for
condition of low-speed motion. Compared with the dynamic 265
232 carrier aircraft is very special. As the limited number of cata-
model, the main advantage of using the kinematic model is 266
233 pults, the path planning of carrier aircraft is coupled with
that the model structure is simple with fewer parameters to 267
234 the mission planning of sortie. If the number of aircraft
be identified. Additionally, system can also be solved accu- 268
235 involved in one sortie mission is bigger than the number of cat-
rately with less computational time. The disadvantage of the 269
236 apults, at least two aircraft would share one catapult, which
kinematic model is that it is based on some premises, for exam- 270
237 suggests that at least two aircraft would share the same termi-
ple, (A) the ground is completely flat; (B) the tire does not 271
238 nal position. As a result, mission planning must be considered
slide; (C) inertia and lateral forces are ignored. These premises 272

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
A review on carrier aircraft dispatch 5

273 are not applicable for the condition of high-speed motion. where aCA is the acceleration of the carrier aircraft, and xCA is 314
274 However, if the system is operating at a low speed, these the front wheel steering angular velocity. The vector 315
 
a1 T
assumptions do not significantly affect the accuracy of the ua1 ¼ ua1 ¼ ½xCA ; aCA T constitutes the control variable
1 ; u2
275 316
276 results. list of model (1). This model is widely used in the motion plan- 317
277 According to Karkee’s research on tractor-trailer systems,25 ning of Automated Ground Vehicles (AGVs).26,27 However, 318
278 the kinematic model can describe the motion characteristics of different from general AGVs, the carrier aircraft cannot realize 319
279 the system with almost the same accuracy as a dynamic model backward motion by itself. In addition, the taxiing velocity of 320
280 when the speed of the system is small (approximately less than the carrier aircraft should be restrained within a safety range in 321
281 4.5 m /s). As the dispatch speed of carrier aircraft on the deck case of emergency brake. Hence, the following constraints on 322
282 is slow, the kinematics model is widely used to study the path taxiing velocity should be imposed: 323
283 planning of carrier aircraft, and the dynamic model is hardly 324

284 used in the path planning problem for carrier aircraft. 0 6 vCA 6 vCA;max ð2Þ 326

285 Based on the assumption made above, the kinematics and In addition, several other constraints should also be taken 327
286 corresponding mechanical constraints of three dispatch modes into consideration: 328
287 are given in the rest of this section. It is noted that when the 329

288 aircraft is towed by a tractor, no matter for the tractor with jbCA j 6 bCA;max ð3Þ 331

289 or without a drawbar, the configuration can be categorized 332

290 into on-axle or off-axle hitching according to whether the hitch aCA;min 6 aCA 6 aCA;max ð4Þ 334
335
291 between the tractor and the drawbar/carrier aircraft lies on the
jxCA j 6 xCA;max ð5Þ 337
292 rear axle of the tractor. Actually, the off-axle hitching is the
293 general form of on-axle hitching. Hence, we will take the off- Considering that the aircraft can tune its front wheel steer- 338
294 axle hitching configuration for example when we model the ing angle without obviously changing its position, Liu et al.28 339
295 traction system in Section 3.3 and Section 3.4 reduced the state space as xa2 ¼ ½x1 ; y1 ; hCA ; vCA T , and mean- 340
 
a2 T
while taking ua2 ¼ ua2
1 ; u2 ¼ ½tanbCA ; aCA T as the new con- 341
296 3.2. Carrier aircraft autonomous taxiing
trol variable, the model given in Eq. (1) can be further 342
simplified as a model with four state variables as follows: 343
344
297 When the carrier aircraft taxi on the flight deck by itself, as 2 3 2 3
298 illustrated in Fig. 3, the positive acceleration and the negative x1 vCA coshCA
299 acceleration are provided by the engine and the brake, respec- dxa2 d6 7 6
6 y1 7 6 vCA sinhCA 7
7
¼ 6 7¼6 7 ð6Þ
300 tively. And the steering operation is controlled by the nose dt dt 4 hCA 5 4 vCA ua2 1 =LCA
5
301 wheel. vCA ua2 346
2
302 Point ðx1; y1 Þ, which is the middle point of the rear axle, is
303 used to represent the position of the carrier aircraft. It is under In model (6), the constraint in Eq. (5) is eliminated. And the 347

304 the condition of pure translation and the corresponding veloc- constraint in Eq. (3) needs to be rewritten into an equivalent 348

305 ity is denoted as vCA . Parameter LCA is the longitudinal dis- constraint on the control variable ua21 as follows: 349
 a2  350
306 tance between the nose wheel and the rear wheel axis. The u  6 tanbCA;max ð7Þ 352
1
307 variable hCA and bCA represent the orientation of the carrier
308 aircraft and the front wheel steering angle, respectively. When
309 selecting the state space as xa1 ¼ ½x1 ; y1 ; hCA ; vCA ; bCA T , the 3.3. Traction system without drawbar 353
310 kinematic equations can be written as
311
2 3 2 3 In this dispatch mode, the tractor provides the power of the 354
x1 vCA coshCA
6 7 6 7 whole traction system instead of the carrier aircraft itself, 355
6 y1 7 6 vCA sinhCA 7
dxa1 d6 7 6 7 and the steering operation is controlled by the front wheel 356
¼ 6 hCA 7 6 7
7 ¼ 6 vCA tanbCA =LCA 7 ð1Þ
dt dt 6
6 7 6 7
steering angle of the tractor. An illustration of the traction sys- 357
4 vCA 5 4 aCA 5 tem without drawbar is shown in Fig. 4. Similar to the mod- 358

313 bCA xCA elling process in Section 3.2, Point (x1,y1) and hCA are used 359
to represent the position and the orientation of the carrier air- 360
craft, respectively. The variable hT is the orientation of the 361
tractor. The variable bCA and bT represent the nose wheel steer- 362
ing angle of the carrier aircraft and the front wheel steering 363
angle of the tractor, respectively, and we have 364
bCA ¼ hT  hCA . The parameters LT is the longitudinal dis- 365
tance between the nose wheel and the rear wheel axis of the 366
carrier aircraft, while the parameter LT is the wheelbase of 367
the tractor. The parameter M1 is the distance between the hitch 368
and the rear wheel axle of the tractor. By velocity decomposi- 369
tion, one can draw the conclusion that the translational veloc- 370
ities of the carrier aircraft and the tractor, which are denoted 371
as vCA and vT , have the following relationship: 372
373
vCA ¼ vT ðLT cosbCA þ M1 tanbT sinbCA Þ=LT ð8Þ 375
Fig. 3 An illustration of carrier aircraft autonomous taxiing.

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
6 X. WANG et al.

addition, the model in Ref. 30 is constructed under the condi- 420


tion M1 ¼ 0. 421

3.4. Traction system with drawbar 422

The traction system with drawbar can be seen as three-body 423


articulated system, as illustrated in Fig. 5. The variables h1, 424
h2, and h3 represent the orientation of the carrier aircraft, 425
the drawbar, and the tractor, respectively. Points (x1,y1) and 426
(x3,y3) are used to represent the position of the carrier aircraft 427
and the tractor. The variables bCA and bT denote the steering 428
angles of the carrier aircraft and the tractor, where 429
Fig. 4 An illustration of traction system without drawbar. bCA ¼ hT  hCA and bB ¼ hT  hB . The parameters LCA, LB, 430
and LT are the longitudinal distance between the nose wheel 431

376 If one selects the state space as and the rear wheel axis of the carrier aircraft, the length of 432
the drawbar, and the wheelbase of the tractor, respectively. 433
377 x ¼ ½x1 ; y1 ; hCA ; hT ; vT ; bT T , the motion of the traction system
b1
And the parameter M2 represents the off-axle distance. 434
378 can be described by the following kinematics:
379 According to the velocity decomposition, the translation veloc- 435
23 2 3
x1 vT ðLT cosbCA þ M1 tanbT sinbCA ÞcoshCA =LT ity of the carrier velocity and the tractor, which are denoted as 436
6 y 7 6 v ðL cosb þ M tanb sinb Þsinh =L 7
6 1 7 6 T T CA 1 T CA CA T 7 vCA and vT, meet the following relationship: 437
6 7 6 7 438
dxb1 d6 6
hCA 7 6 vT ðLT sinbCA  M1 tanbT cosbCA ÞsinhCA =LCA LT 7
7 6 7
¼ 6 7 ¼6 7 vCA ¼ vT cosbCA cosbB ðLT þ M2 tanbT tanbB Þ=LT ð16Þ 440
dt dt 6 hT 7 6 vT tanbT =LT 7
6 7 6 7
4 vT 5 4 aT 5 On the basis of kinematic representation of any n-body 441
bT xT articulated vehicle proposed by Park et al.,31 Johnston et al. 442
381 ð9Þ developed a kinematic model for the traction system with 443
drawbar by combining the single-body kinematics of each 444
382 where aT is the acceleration of the tractor, and xT is the front
component.32 However, this model selects 14 state variables 445
383 wheel steering angular velocity of the tractor. The vector
 
b1 T
where many of them are abundant for the purpose of imposing 446
384 ub1 ¼ ub1
1 ; u2 ¼ ½xT ; aT T represents the control variable list constraints in the DIDO software.33 It results in huge compu- 447
385 of model (9). In addition, the following constraints on state tational burden in reality. In addition, this model is con- 448
386
387
and control variables should be considered: structed under the on-axle hitching condition, i.e., M2 ¼ 0. 449

389 jvT j 6 vT;max ð10Þ Liu et al. selected the state space as 450
T
390 x ¼ ½x1 ; y1 ; hCA ; bCA ; bB  and proposed a simple kinematic
c
451
392 jbCA j 6 bCA;max ð11Þ model as34 452
2 3 453
393
2 3 uc2 coshCA
395 jbT j 6 bT;max ð12Þ x1 6 7
6 7 6 uc2 sinhCA 7
396
6 y 7 6 7
aT;min 6 aT 6 aT;max ð13Þ c 6 1
7 6 u c
tanb =L 7
398 dx d6 6  CA CA
 7
¼ 6 hCA 7
2
7 ¼6 7
7 6  7
tanb tanb
399 dt dt 6 uc B CA

jxT j 6 xT;max ð14Þ 4 CA 5 6 


b 6 2 LB cosbCA LCA 7
401
7
4 u c 5
402 Similar to the simplification of carrier aircraft autonomous bB uc2 cosb cosb L1 þM uc tanb  LB cosbBCA tanb
CA Bð T 2 1 BÞ
403 taxiing in Section 3.2, by taking tanbT as a new control vari-
404 able, Liu et al. simplified the model into the following form29: ð17Þ 455
405 2 3 2 3
x1 vT ðLT cosbCA þ M1 tanbT sinbCA ÞcoshCA =LT
6 7 6 7
6 y1 7 6 vT ðLT cosbCA þ M1 tanbT sinbCA ÞsinhCA =LT 7
dx d6
b2 7 6 7
¼ 6 h 7 ¼ 6 vT ðLT sinbCA  M1 tanbT cosbCA ÞsinhCA =LCA LT 7 ð15Þ
dt 6 7 6 7
CA
dt 6 7 6 7
4 hT 5 4 vT u1 =LT
b2
5
407 vT ub2
2

 
b2 T
408 where ub2 ¼ ub2 1 ; u2 ¼ ½tanbT ; aT T is the control variable
409 list. The treatments of constraints are similar to what is done
410 to model (6), i.e., the constraint on the front wheel steering
411 angular velocity as described in Eq. (14) is eliminated, and
412 the constraints on the front wheel steering angle are replaced
413 by an equivalent constraint on the control variable ub2 1 .
414 In addition to modelling by Ordinary Differential Equa-
415 tions (ODEs), Shi et al. modelled the traction system without
416 drawbar using Differential-Algebraic Equations (DAEs).30
417 Modelling by DAEs is institutive and convenient, but it intro-
418 duces extra state variables and differential equations when
419 compared with the kinematic equations in other work.29 In Fig. 5 An illustration of traction system with drawbar.

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
A review on carrier aircraft dispatch 7
 T
456 where uc ¼ uc1 ; uc2 ¼ ½tanbT ; vCA T is the control variable list. whether it is a geometric algorithm or a trajectory planning 511

457 Model (17) is under the assumption that all steering angles are method depends on its formulation. 512

458 no more than 55° and M 6 L3 so that denominators involved Various constraints can be considered simultaneously in 513

459 in Eq. (17) are non-zero. This assumption is generally true for trajectory planning methods. However, methods in this cate- 514

460 the actual traction system with drawbar on aircraft carriers. To gory could be extremely time-consuming since a complicated 515

461 the best of our knowledge, model (17) is so far the most concise optimization problem, which is commonly non-convex due 516

462 kinematic model of traction system with drawbar that consid- to the highly nonlinearity in system equations and collision- 517

463 ers all necessary states of the carrier aircraft. However, a tiny free conditions, is required to be solved. Compared to trajec- 518

464 shortcoming of this model is that the control variables selected tory planning methods, geometric methods are much computa- 519

465 are not the actual control input for the traction system. For tionally efficient. However, it should be noted that path 520

466 this model, constraints on the three steering angles and the obtained by this category of methods cannot avoid the possi- 521

467 velocity of the carrier aircraft in the following form should bility of kinematical infeasibility since the kinematical con- 522

468 be imposed: straints are not accounted. Hence, we cannot simply 523
469 determine which kind of methods is better. The five kinds of 524
471 vCA;min 6 uc2 6 vCA;max ð18Þ methods and related work are reviewed and their pros and 525
472  c cons are summarized in the rest of this section. 526
474 u  6 tanbT;max ð19Þ
1
475 4.1. Dijkstra’s algorithm based method 527
477 jbCA j 6 bCA;max ð20Þ
478
In this kind of method, the active carrier aircraft is outlined by 528
480 jbB j 6 bB;max
a circle that contains itself with the radius as minimal as pos- 529
481 Although this model can better describe the kinematics of sible, and the circle is called the feature circle of the carrier air- 530
482 the system, it simplifies the functions of the derivative of h2 . craft. The obstacles in the environment are modelled by convex 531
483 If the effect of each part on the system is fully considered, it hulls according to their feature shapes. The threatened zone is 532
484 will be in the following form, which is more complex than obtained by expanding these convex hulls according to the 533
485 the model listed in Eq. (17). radius of the active carrier aircraft. Then one has to find all 534
486
2 3 way-points within the search zone and compute the cost 535
uc2 coshCA between each pair of way-points to generate the feasible 536
2 3
6 x1 uc2 sinhCA 7 weighted graph. It should be noted that modelling the obstacle
6 7 537
6 7 6 7 by different feature shapes would lead to distinct manners to
6 y 7 6 u c
tanb =L 7 538
d6 7 6  7
1 CA
dxc 2 CA
 define way-points.
¼ 6 7¼6 7 539
6 hCA 7 6 LT tanbB M2 u1 c
tanbCA 7
dt dt 6 7 6 u2 LB cosbCA ðLT þM2 uc tanbB Þ  LCA 7
c Zhang et al.35 studied the path planning problem on the 540
4 CA 5 6
b 7
6  7 flight deck. The dumb carrier aircraft are simply modelled as
1
 541
bB 4 LB uc1 LT sinbB þM2 uc1 cosbB 5 circles, and the island and a team of dumb carrier aircraft 542
c
u2 L cosb cosb L þM uc tanb
B CA B ð T 2 1 B Þ parked in order closely are seen as rectangles. To fulfill the 543
488 ð22Þ requirement on the terminal heading angle, a predefined 544
way-point is introduced and taken as the new destination. 545
489 However, it remains to be verified that whether it can be Two tangent circle threatened regions whose radii are the min- 546
490 directly applied to the trajectory planning problem by using imum turning radius of the active carrier aircraft are placed at 547
491 optimal control methods due to the extremely high either side of initial position, and their tangent line is in the 548
492 nonlinearity. same direction with initial heading angle. Similar threatened 549
zones are added around the destination. Together with the 550
493 4. Path planning techniques for single carrier aircraft threatened zones generated by the obstacles, the search space 551
is obtained. It is seen that all candidate path segments consist 552
494 There are typically five kinds of path planning techniques for of arcs and common tangent lines of threatened zones. Then 553
495 single carrier aircraft or single traction systems, which are one has to traverse every pair of midpoints of these common 554
496 Dijkstra’s algorithm, A* algorithm and its variants, behavior tangent lines set corresponding distance cost. Under the same 555
497 dynamics-based method, intelligent search method, and framework, they further study the path planning problem of 556
498 dynamic optimization method. These techniques can be gener- the traction system.36 557
499 ally classified into two categories, i.e., the geometric algorithms Compared with the flight deck, the hangar is a much nar- 558
500 and the trajectory planning algorithms, according to whether rower space with carrier aircraft and other staffs closely stored. 559
501 the kinematic model (or the dynamic model) is considered. Thus, it would be impossible to generate a feasible path for the 560
502 The main difference between these two categories is that there tracking system if we still model dumb aircraft by their feature 561
503 is only a series of way-points in the result obtained by geomet- circles., Si et al.37 studied the path planning problem of the 562
504 ric algorithms. However, information on time corresponding traction system in the hangar, where the dumb aircraft is mod- 563
505 to the state is also included in the solutions obtained by trajec- elled by the minimum pentagon instead. And all way-points, 564
506 tory planning algorithms. Under this way of classification, which consist of the vertexes of threatened zones, are searched 565
507 Dijkstra’s algorithm, A* algorithm and behavior dynamics- by an iterative manner. This method has two drawbacks. First, 566
508 based method can fall into geometric algorithms, while the final obtained path is composed of several line segments, 567
509 dynamic optimization method is categorized as a trajectory which suggests that extra smoothening techniques must be 568
510 planning algorithm. And as for an intelligent search method, incorporated. Secondly, it omits the orientation of the carrier 569

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
8 X. WANG et al.

570 aircraft, limiting its application to scenarios where high dynamic weight heuristic function and developed a multi- 631
571 demand for terminal orientation exists. step optimization scheme to solve the problem in presence of 632
572 The key in Dijkstra’s algorithm based method is how to obstacles. In order to achieve efficient collision avoidance, they 633
573 generate a feasible weighted graph according to the informa- extended the heuristic function designed in Ref. 38 to 634
635
574 tion on obstacles and the boundary conditions. By using uðPk Þ ¼ w1  lðPk Þ þ w2  k  hðPk Þ  w3  bj ðPk Þ ð25Þ 637
575 proper strategy to construct the threatened zone, the con-
576 straint on terminal orientation and the mechanical constraint where wi ði ¼ 1; 2; 3Þ are positive dynamic weight that satisfy 638
577 on minimum turning radius can be easily taken into consider- w1 þ w2 þ w1 ¼ 1; bðPk Þ is a function that describes the sum 639
578 ation. Once the feasible graph is formulated, Dijkstra’s algo- of the distances between the current sparse node and obstacles. 640
579 rithm can efficiently solve the shortest path. However, since Using the idea of MPC for reference, the multi-step optimiza- 641
580 the threatened zone and the search space directly depend on tion scheme is made up of two modules, i.e., the prediction 642
581 the position of obstacles, this kind of method, to the best of module and the optimization module. The prediction module 643
582 our knowledge, has not been applied to environments with determines the direction of the next path node and outputs it 644
583 dynamic obstacles. In addition, the velocity in this kind of to the optimization module, and then the position of the next 645
584 method is constant. Hence, it cannot precisely describe the path node is determined by optimization according to the cost 646
585 accelerating motion and the decelerating motion around the function of each possible sparse node. As seen in Fig. 6, the 647
586 initial position and the terminal position, respectively. multi-step optimization scheme can extensively decrease the 648
probability that none of the sparse path nodes is available 649
587 4.2. Improved A* algorithm based method due to the unforeseen obstacle in single-step method. This 650
paper presents two strategies to cope with dynamic obstacle, 651

588 The A* algorithm expands nodes according to the cost func- i.e., the waiting strategy and the detour strategy. 652

589 tion in the following equation: In both Ref. 38 and Ref. 39, they pointed out that the 653
590 searching procedure may fail due to the possible condition 654
592 fðPk Þ ¼ gðPk Þ þ uðPk Þ ð23Þ where constraint on terminal orientation cannot be satisfied 655

593 where gðpk Þ is the true cost from the initial node to the current in the last step. Hence, as illustrated in Fig. 7, it would be a 656

594 sparse node and uðpk Þ is the heuristic function that denotes the good idea to start the searching procedure from the destina- 657

595 estimated cost from the current sparse node to the destination tion. However, this trick leads to another issue that the con- 658

596 node. The sparse node that minimizes fðpk Þ will be taken as the straint on initial parking orientation cannot be satisfied. 659

597 actual path node in the current step. The selection of the Inspired by the skill of introducing an extra way-point around 660

598 heuristic function, which actually requires many skills and the destination in Ref. 38 and Ref. 39, we think that it would 661

599 experience, directly determines the quality of the generated be wise to select a straight-forward point to the initial orienta- 662

600 path and the search efficiency. tion near the initial parking position and take it as the virtual 663

601 The constraint on terminal orientation cannot be easily sat- initial position when applying the A* algorithms. 664

602 isfied in the classic A* algorithm. Hence, various improved A*


603 algorithms are developed, where the key is to design a reason- 4.3. Behavior dynamics-based method 665

604 able heuristic function to meet the precision of the terminal


605 orientation and collision-free conditions. And various con- Behavior dynamics-based method is originally developed for 666
606 straints (e.g., minimum length of path segment, max turning path planning problems for mobile robots40,41 and recently 667
607 angle, maximum total path length, direction of reaching desti- applied to those of carrier aircraft. It is considered that moving 668
608 nation) are integrated simultaneously when expanding path towards the object and obstacle avoidance constitute the path 669
609 nodes to narrow the search space and correspondingly improve planning behavior modes of carrier aircraft. The velocity and 670
610 the search efficiency. the heading angle are commonly selected as the behavior vari- 671
611 To solve the taxiing path planning problem from the eleva- ables, and the behavior modes are realized by two independent 672
612 tor to the catapult in obstacle-absent situations, Wu and Xu38 differential equations of these two behavior variables. Actu- 673
613 proposed an improved A* algorithm with dynamic weighted
614 heuristic function. The heuristic function therein consists of
615 two items, which can be given in the following equation:
616
618 uðPk Þ ¼ lðPk Þ þ a  k  hðPk Þ ð24Þ
619 where lðPk Þ represents the distance between the current sparse
620 node and the destination; a is a constant weight factor; k rep-
621 resents the serial number of current sparse node; hðPk Þ is the
622 reaching orientation. When all path nodes are expanded, a
623 path that consists of line segments is obtained, and a path
624 smoothening method is provided to generate the feasible path.
625 They also pointed out that it is a good choice to start the
626 search from the destination with prescribed orientation. This
627 could decrease the failure rate in case of possible conditions
628 where prescribed terminal orientation cannot be satisfied.
629 By incorporating the idea of Model Predictive Control
630 (MPC) with the A* algorithm, Wu and Xu39 designed a new
Fig. 6 Multi-step optimization scheme using the idea of MPC.39

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
A review on carrier aircraft dispatch 9

674 ally, this kind of method is, to some extent, similar to the arti- rithm was as the core solver. The method in Ref. 46 is actually 719
675 ficial potential field method since the design of fields is the a trajectory planning method since the time variable is consid- 720
676 major concern (especially for complicated environment).42 ered. A strategy of segmented fitness function is developed, 721
677 When the fields are properly designed, the method shows high numerical results therein demonstrate that the terminal con- 722
678 efficiency. straints on position and orientation can be restricted within 723
679 Zhang et al.43 proposed an improved strategy for obstacle prescribed ranges. In addition, dynamic and static obstacles 724
680 avoidance based on the generalized symbolic threshold func- can be treated under the uniform framework in this method. 725
681 tion to solve the taxiing planning problem. Recently, the above However, trajectories obtained by this method are not smooth 726
682 method is extended to solve the dispatch path of traction where many abrupt steering operations exist. 727
683 systems.44 Su et al.47 developed a taxiing path planning method that 728
684 It should be noted that constraints on terminal orientation combines artificial experience in selecting intermediate nodes 729
685 are hard to be considered in the behavior dynamics-based and the Modified Artificial Bee Colony (MABC) algorithm 730
686 method and the design of the field requires high skill. for local optimization. When optimized intermediate path 731
nodes are obtained, the Dubins method or the Reeds-Shepp 732
687 4.4. Intelligent search method method is used to generate a feasible trajectory. Hence, the 733
constraints on the minimum turning radius and terminal orien- 734

688 In this kind of method, the path is first discretized at several tation are strictly satisfied. A similar idea is used in Ref. 48, but 735

689 points and the state variables or the control variables at dis- the genetic algorithm instead is used as the optimization prob- 736

690 cretized nodes are taken as the unknown variables. Then intel- lem solver, where a penalty term and a gene repairing strategy 737

691 ligent search methods are adopted to solve the unknown are introduced to accelerate the genetic process. The main 738

692 variables and then the complete path is finally obtained. Car- drawback of these two methods is that human intervention is 739

693 rier aircraft are generally modelled by their convex hulls and necessary since such two algorithms cannot determine the 740

694 collision detection algorithm based on convex hulls must be proper amount of intermediate node automatically. 741

695 incorporated. A flowchart of general intelligent search algo- In fact, a proper discretization and unknown variable selec- 742

696 rithm is illustrated in Fig. 8. From Fig. 8, it is seen that there tion strategy, instead of what intelligent search method is 743

697 are three key processes when designing an intelligent method, adopted, is the key factor determining the quality of the gener- 744

698 i.e., the encoding formulation of path/trajectory, the evolu- ated path. Generally speaking, this method owns strong com- 745

699 tionary strategy of population and the selection of fitness putational robustness and good calculation efficiency. 746

700 function. Moreover, it can theoretically achieve global optimality under 747

701 Han et al. modelled carrier aircraft by convex hulls and the given discretization scheme. However, due to the limit of 748

702 proposed a collision detection algorithm based on convex computational time in practice, it cannot avoid the possibility 749

703 hulls.45 The flight deck is divided into segments along the x- of falling into local optimum. 750

704 axis direction, and Clustering Particle Swarm Optimization


705 (CPSO) algorithm is used to optimize the corresponding y- 4.5. Dynamic optimization method 751

706 axis coordinates. Constraints on minimum segment length,


707 total path length, and maximum steering angle are also consid- In dynamic optimization method, a nonlinear dynamic opti- 752
708 ered when formulating the optimization problem. By adopting mization problem is formulated by considering various con- 753
709 a multi-population strategy, several spare paths can be gener- straints of the system, such as system kinematics, constraints 754
710 ated as well. When the segments of the path are obtained, the on taxiing velocity, avoidance of input saturation, collision- 755
711 extra smoothening procedure is required. This method has two free conditions, etc. A general form of constrained optimal 756
712 drawbacks: (A) the orientation of the carrier aircraft is not control problem is given as 757
713 taken into consideration; (B) due to the selected discretization 8 Rt 758
>
> minJ ¼ uðxðtf Þ; tf Þ þ tsf /ðxðtÞ; uðtÞ; tÞdt
714 strategy, it inherently lacks the capability to describe the u- >
>
>
> s:t:
715 turn behavior. <
716 In previous work,46 Wu et al. took the front wheel angular system equations x_ ¼ fðxðtÞ; uðtÞ; tÞ ð26Þ
>
>
717 velocity at Chebyshev collocation points as variables to be >
> boundary constraints Uðxðts Þ; ts ; xðtf Þ; tf Þ ¼ 0
>
>
718 optimized, and the Chicken Swarm Optimization (CSO) algo- :
path constraints hðxðtÞ; uðtÞ; tÞ 6 0 760

where J is the cost functional to be optimized and can be set as 761


minimum time, minimum energy, or their linear combination. 762
It should be noted that, for the purpose of numerical stabil- 763
ity, optimal control-based methods usually require that the 764
functions involved are at least C1 -continuous (i.e., the function 765
itself and its first-order derivative are both continuous). Hence, 766
in optimal control-based methods, the active carrier aircraft or 767
traction system is usually seen as a circle, and the obstacles are 768
usually described by the quasi-rectangle, leading to the 769
collision-free condition in the following form49: 770
771
  2p   2p
xx ðtÞ ðtÞ
aþrCA þrsafe
þ bþryy
CA þrsafe
61 ð27Þ 773
Fig. 7 Strategy of starting searching process from the destina-
tion in A* algorithms in Ref. 38 and Ref. 39.

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
10 X. WANG et al.

Fig. 8 An illustration of dispatch process in sortie and recovery task for carrier aircraft.

 

774 where (x,y) and x; y are the centers of the carrier aircraft and precision but smaller convergence radius. Wang et al. cre- 816

775 the obstacle, respectively; rCA is the radius of feature circle of atively incorporated indirect methods with pseudospectral dis- 817

776 the carrier aircraft; parameters a, b and p 2 Zþ are used to cretization and constructed a series of Symplectic 818

777 define the shape of the quasi-rectangle. We should point out Pseudospectral Methods (SPMs).53–55 The SPMs were then 819
 
applied to path planning of various autonomous systems, 820
778 that the parameters x; y can be time-variant, and thus opti-
including carrier aircraft on the deck. Compared with direct 821
779 mal control-based methods can treat static obstacle and
pseudospectral methods, constraints are satisfied more strictly 822
780 dynamic obstacles under the identical framework.
and the computational efficiency is higher, which makes SPM 823
781 Liu et al.29 and Shi et al.30 studied the path planning prob-
an appealing method for solving path planning of carrier air- 824
782 lem of the traction system without drawbar using an ODE
craft on the deck. 825
783 model and a DAE model, respectively. In both works, the path
Though optimal control based methods own incomparable 826
784 planning problem is formulated as a minimum energy problem
advantages over other kinds of path planning techniques, two 827
785 within a fixed domain. The numerical methods adopted in
shortcomings limit their wide application in reality. On the one 828
786 these two studies are both symplectic, leading to sound compu-
hand, it is well-known that the solution of a nonlinear opti- 829
787 tational efficiency.
mization problem severely depends on the quality of initial 830
788 Based on the classic bicycle model, Li et al.50 used the
guesses. Hence, for the complicated flight deck environment, 831
789 Legendre Gauss Pseudospectral (LGP) method to solve taxiing
the solving procedure could be extremely time-consuming or 832
790 path planning. When formulating the optimal control prob-
even failed without proper initial guesses on state and control 833
791 lem, the terminal time is not specified therein. In Ref. 51, they
variables. And such effect of initial guesses is more significant 834
792 used genetic algorithms to optimize the catapult allocation
for indirect methods. On the other hand, once carrier aircraft 835
793 scheduling for a team of aircraft launching under a given pri-
are closely parked, e.g., in the hangar, it is impossible to gen- 836
794 ority. By relaxing the lower bound of the taxiing velocity to
erate a path because the feature circle wastes too much feasible 837
795 negative, the towing path and time for each carrier aircraft
space. 838
796 therein are calculated under the framework of Ref. 50.
At the end of this section, a brief comparison of the above 839
797 Johnston and Swenson32 focused on the path planning
five kinds of dispatch path planning method is provided in 840
798 problem of the traction system with drawbar. They took the
Table 2. 841
799 front-wheel steering angle of the tractor and the velocity of
800 the tractor as control variables and the optimal control was
801 solved by the software package DIDO. However, due to the 5. Coordinated path planning techniques for multiple carrier 842

802 limit of DIDO, they adopted a kinematic model (it is made aircraft 843

803 up of 14 equations) for the simplicity of imposing constraints.


804 The computational efficiency is extremely low and cannot be In the previous section, we reviewed the existing research on 844
805 applied in real time. path planning techniques for single carrier aircraft. However, 845
806 Numerical methods for optimal control problems are for launch mission in reality, there are usually more than 846
807 divided into direct and indirect methods. Direct methods one carrier aircraft to be dispatched. On the one hand, if deci- 847
808 transform optimal control problems into NonLinear Program- sions are made manually, the complicated situation is hard for 848
809 ming (NLP) which is then solved by NLP solvers. The pseu- staff. On the other hand, if we still adopt the path planning 849
810 dospectral methods,52 as the most popular direct method, technique for single carrier aircraft, only one carrier aircraft 850
811 have been widely used in solving optimal control problems is allowed to taxi on the flight deck. If so, much sparse space 851
812 resulted from path planning of carrier aircraft on the on the flight deck will be underutilized, leading to extremely 852
813 deck.32,50,51 In contrast, indirect methods transform optimal low dispatch efficiency.51 Hence, coordinated path planning 853
814 control problems into two-point boundary value problem. techniques for multiple carrier aircraft are of more practical 854
815 Compared with direct methods, indirect methods own higher value than those for single carrier aircraft. 855

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
A review on carrier aircraft dispatch
10.1016/j.cja.2020.06.020
Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/

13 July 2020
CJA 1668
Table 2 Performance comparison among five kinds of dispatch trajectory planning methods.
Features Dijkstra’s algorithm Improved A* algorithm Behavior dynamics Intelligent search Dynamics optimization
Critical step (which is also core difficulty) Construction of search Design of heuristic Proper selection of ODE to Reasonable design of Robust and efficient solving of complex
in the method space and threatened function to meet describe behavior dynamics discretization scheme nonlinear optimal control problems
zones constraints
Satisfaction level of Motion 1 1 1 3 5
various kinds of equations
constraints
Minimum 5 4 4 3 5
turning radius
Obstacle 5 4 4 3 5
avoidance
Velocity 1 1 2 3 5
range
Terminal 5 2 2 3 5
position and
pose
Control 1 1 2 3 5
saturation
Extra trajectory smoothing? No need Need No need Normally need No need
Efficiency High Relatively high Relatively high Depends on Relatively low
discretization scheme
Robustness High Medium Medium Normally high Relatively low
Note: The value for satisfaction level of constraints represents the following meaning: 1-not considered; 2-poor; 3-depends on discretization scheme; 4-relatively good; 5-strictly satisfied.

No. of Pages 19
11
CJA 1668 No. of Pages 19
13 July 2020
12 X. WANG et al.

856 According to the core mission, research related to coordi- However, RVO-like methods have several inherent drawbacks. 880
857 nated path planning for multiple carrier aircraft can be put On the one hand, when an agent approaches the destination, it 881
858 into two categories. In the first category,56 they only focus wanders around the destination rather than comes to a com- 882
859 on how to generate a feasible path from the initial parking plete stop. On the other hand, the orientations of agents are 883
860 position to the destination for multiple carrier aircraft taxiing not considered. To overcome these two shortcomings, as 884
861 on the flight deck simultaneously. In the second category,57–59 shown in Fig. 9(a), the whole taxiing trajectory of each carrier 885
862 they focus on modelling and optimizing the sortie task schedul- aircraft was decomposed into two stages. In the first stage, air- 886
863 ing, where the path planning for multiple carrier aircraft is seen craft taxies from the initial parking position to a position near 887
864 as an independent module that integrates to the upper-level the destination, i.e., approximately the distance of two times of 888
865 mission planner. The path planning module generates crucial the minimum turning radius away. And in the second stage, 889
866 information (i.e., the taxiing distance and the taxiing time) the aircraft completes the remaining path. In the first stage, 890
867 for evaluating a certain sortie scheduling and outputs it to the improved Dubins-RVO method as illustrated in Fig. 9(b) 891
868 the top-level optimization solver. is applied to realize efficient obstacle avoidance. As for the sec- 892
869 Liu et al.56 proposed a compound path planning technique ond stage, it is assumed to be an obstacle-free condition (and 893
870 for multiple carrier aircraft. The Reciprocal Velocity Obstacle the assumption is reasonable according to the real flight deck 894
871 (RVO) method60 is an efficient obstacle avoidance and path environment), and the One-Sided Symplectic Pseudospectral 895
872 planning method for a multi-agent system. Based on the con- (OSSP) method is adopted to generate the remaining path that 896
873 cepts of the RVO method and the Dubins curve method, an strictly satisfies the prescribed terminal orientation. 897
874 improved Dubins-RVO method is constructed. In the Li et al.57 studied the path planning for aircraft fleet launch- 898
875 Dubins-RVO method, various kinds of constraints (i.e., the ing on the flight deck of carriers with limited space and 899
876 kinematic model, maximum front wheel turning radius, and resources, according to the launching plan received. An 900
877 maximum taxiing velocity) are considered. Thus, compared improved A* algorithm with a segmented dynamic weighted 901
878 to the classic RVO method, the improved Dubins-RVO heuristic function is adopted. A top-level mathematical model 902
879 method can effectively avoid the ‘‘dead-lock” phenomenon. that determines the total consumption of aircraft fleet launch- 903

Fig. 9 Main idea of compound path planning method in Ref.56.

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
A review on carrier aircraft dispatch 13

Fig. 10 Hierarchical distributed mission planning architecture in Ref. 58.

904 ing is established at first. This model contains the constraints multiple times in each step of the path search to ensure the 921
905 of maneuverability, the work mode of catapults, ground safety of each path. It should be noted that the aircraft must 922
906 motion, and collision detection of aircraft taxiing on the flight track the planned path in a chronological order. 923
907 deck. As stated previously, since the path obtained by A* algo- Focusing on the same issue with Ref. 57, Wu et al.58 pro- 924
908 rithms consists of several path segments, extra smoothening posed a novel distributed mission planning architecture. The 925
909 techniques must be incorporated to get the feasible path. Thus, architecture is hierarchical and is comprised of four levels as 926
910 it will be a two-step process to get the feasible path, and cannot illustrated in Fig. 10, namely, the input level, the coordination 927
911 guarantee that the path meets the constraints of maneuverabil- level, the path planning level, and the execution level. Realistic 928
912 ity. Therefore, the path tracking is combined with path plan- constraints in each level of the distributed architecture, such as 929
913 ning method developed in Ref. 39. When generating the taxi the vortex flow effect, the crowd effect, and the motion of air- 930
914 path for a period of time in future, the rolling optimization craft are considered in this model. A distributed path planning 931
915 method is used to optimize the formulated performance index algorithm is developed, and an asynchronous planning strat- 932
916 of path tracking, and the planning domain and the control egy based on token passing that can avoid the dead lock and 933
917 domain are defined to execute the optimal control instructions Buckets effect is adopted. 934
918 (the nose wheel steering angle is selected therein) only over a Liu et al.59 described the sortie scheduling as a Flexible 935
919 short period of time. Not only the constraints of a single air- Flowshop Scheduling Problem (FFSP) model, and the collab- 936
920 craft are considered, but also collision detection is executed orative trajectories planning of taxiing for multiple aircraft is 937
the core stage therein. In Ref. 59, the pre-determined timing 938
scheme of the sortie task is used as the input of collaborative 939
trajectories planning of the taxiing of multiple aircraft, and 940
optimal scheme is finally obtained by the dynamical adjust- 941
ment of the pre-determined timing scheme according to collab- 942
orative trajectories planning. To solve the collaborative 943
trajectories planning, the OSSP method is adopted to calculate 944
the offline feasible trajectories from each gate position to all 945
catapults/preparation spots, and the obtained paths are stored 946
in the trajectory database. Then the timing sequence of each 947
aircraft is optimized according to the priority principle. This 948
method is robust since it inherently avoids the ‘‘dead-lock” 949
phenomenon. However, as illustrated in Fig. 11, the collision 950
avoidance strategy in this method is not rich, where aircraft 951
with lower priority can only wait for the aircraft with higher 952
Fig. 11 Strategy to determine the waiting time adopted in Ref. priority until the collision does not happen. 953
59.

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
14 X. WANG et al.

954 6. Trajectory tracking inputs, respectively; h represents the constraints considered, 995
which may contain constraints on control variables, 996

955 As discussed in Section 4 and Section 5, control inputs cannot collision-free conditions, maximum allowable taxiing velocity, 997

956 be obtained simultaneously with the path except for optimal and other necessary mechanical constraints. 998

957 control based methods. Actually, even for optimal control- An illustration of the MPC controller is given in Fig. 12. It 999

958 based methods, the obtained control input still cannot be is seen that when the open-loop optimal control problem is 1000

959 directly applied to the carrier aircraft or the traction system solved, the first step following in the control sequence is taken 1001

960 due to the existence of various disturbance. There are generally as the control input for the current control window. And tak- 1002

961 two kinds of disturbances considered in the existing research, ing the autonomous taxiing as an example, an overall proce- 1003

962 including the disturbance in the initial parking position and dure of the MPC trajectory tracking controller is provided in 1004

963 the continuous external disturbance induced by the wind and Fig. 13. 1005

964 sway of the aircraft carrier.


965 Hence, it is necessary to adopt a closed-loop controller, 6.2. Related work 1006

966 where the MPC control is frequently used, to track the refer-
967 ence trajectory provided by the motion planner. In this section, Zhang et al.35–36 studied the path planning problem for carrier 1007
968 the basic mathematical formulation of the MPC trajectory aircraft and traction systems without drawbar, respectively. 1008
969 tracking controller is given first. Then, related researches on When the feasible trajectory is obtained, the MPC controller 1009
970 the trajectory tracking of carrier aircraft are reviewed. Skills is applied to track the reference trajectory. Numerical results 1010
971 to design a better MPC controller are then discussed. Finally, therein demonstrate that the terminal errors, including both 1011
972 the analogy of trajectory tracking between carrier aircraft and position error and orientation error, are within an allowable 1012
973 other vehicles is provided. range. However, the purpose of applying the MPC tracking 1013
controller is only to verify the feasibility of the reference trajec- 1014
974 6.1. Mathematical formulation of MPC trajectory tracking tory and no perturbation is considered therein. Besides, the 1015
975 controller control variable (e.g., the steering angle of the tractor in Ref. 1016
36) may vary sharply since the actual control saturation is 1017

976 The MPC controller is designed based on the idea of receding not considered. 1018

977 horizon optimization. There are typically two forms of MPC, Liu et al.29 focused on the path planning and tracking con- 1019

978 i.e., the continuous and the discrete form. In this section, the trol of the traction system without drawbar. Similar to the off- 1020

979 continuous form is considered due to its wider application in line path planning module, the kinematic model given in Eq. 1021

980 the trajectory tracking problem of the carrier aircraft. (15) is used for the online path tracking module. Numerical 1022

981 In an MPC controller, the sampling period is set as dT and results therein suggest that the initial error can be quickly elim- 1023

982 the length of the prediction window is fixed as T ¼ MdT. The inated by the MPC controller, and the MPC controller can 1024
1 efficiently track the reference trajectory even under continuous 1025
983 sequence of sampling instant is denoted as t0Dk k¼1 such that
external disturbances. By using the symplectic pseudospectral 1026
984 t0Dk  t0Dk1 ¼ dT. At the kth sampling instant, it is assumed MPC controller developed therein, the average computational 1027
985 that all state variables of the carrier aircraft (or the traction time for each sampling instant can be restrained within 10 ms, 1028
986 system) are observable and denoted as xk , and then an open- satisfying the real-time calculation demand. 1029
987 loop optimal control defined over the current prediction win- Liu et al.56 studied the cooperative taxiing path planning of 1030
988 dow is required to be solved as follows: multiple carrier aircraft. When the feasible trajectory of each 1031
989 8 R t0 þT  
>
> min t0Dk k e k2P þ k u k2R dt
carrier aircraft is obtained, the MPC controller is applied to 1032
>
>
>
> Dk track the reference trajectory separately. However, as for the 1033
>
< s:t: kinematic equations used when designing the MPC controller, 1034
Pk : Kinematic equations ; k ¼ 1; 2; . . . ð28Þ they further removed the taxiing velocity from the state vari-
>
> 0
0

1035
>
> able form the model in Eq. (6) and meanwhile took the taxiing
>
> x tDk ¼ xk ; x tDk þ T is free 1036
>
: velocity and the front wheel steering angle as the control 1037
991 hðx; u; tÞ 6 0
inputs. Numerical results therein show that taxiing velocity 1038
992 where the variable e is used to represent the error between jumps between the maximum allowable value and zero fast, 1039
993 the actual trajectory and the reference trajectory; the matrices which means that the carrier aircraft frequently comes to a full 1040
994 P and R are weighted matrices on tracking error and control stop during the taxiing. On the one hand, it is not possible for 1041
the engine to switch that fast. On the other hand, even though 1042
the fast on–off switch of the engine is allowable, the frequent 1043
acceleration-deceleration could make the pilot extremely 1044
uncomfortable. 1045

6.3. Some skills in designing the MPC controller 1046

6.3.1. Selection of kinematic model 1047

It is seen that the kinematic model used in the MPC controller 1048
is identical to or much simpler than that used in trajectory 1049
planning in most studies. However, this may cause that some 1050

Fig. 12 An illustration of model predictive controller. variables vary extremely fast within the designed allowable 1051

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
A review on carrier aircraft dispatch 15

Fig. 13 An overall procedure of MPC tracking controller for carrier aircraft autonomous taxiing.

1052 range. Hence, it is necessary to use a relatively precise model in ing performance. Since we mainly focus on the tracking perfor- 1090
1053 the MPC controller. For example, one can use the kinematic mance of carrier aircraft itself, the elements in the matrix P 1091
1054 model (6) in trajectory planning while using the kinematic corresponding to the state of carrier aircraft should be 1092
1055 model (1) in trajectory tracking. Thus, by taking the front assigned with bigger values. 1093
1056 wheel steering angular velocity instead of the front wheel steer-
1057 ing angle as a control variable, the profile of the front wheel 6.3.5. Design of initial guesses 1094
1058 steering angle obtained by the MPC tracking controller would For the MPC controller, an open-loop optimal control prob- 1095
1059 be smooth. lem is required to be solved at each sampling instant, and 1096
the quality of initial guesses may severely determine the com- 1097
1060 6.3.2. Setting of allowable ranges of state and control variables putational efficiency. If the actual trajectory does not deviate 1098
1061 Considering the existence of external disturbances, the bounds much from the reference trajectory, one can directly take the 1099
1062 set in the tracking module should be slightly larger than those interpolation of reference trajectory as initial guesses. An alter- 1100
1063 in the planning module. Otherwise, the tracking error cannot native way is that, the solutions obtained at the current sam- 1101
1064 be effectively restrained or even diverge in the presence of large pling instant are interpolated and then form the initial 1102
1065 external disturbance. A reasonable strategy is that: (1) first, guesses for the next sampling instant. 1103
1066 when conducting path planning, the necessary bounds imposed
1067 should be much tighter than corresponding allowable ones; (2) 6.3.6. Selection of the length of prediction window 1104
1068 then, when conducting trajectory tracking, the bounds The appropriate length of the prediction window is of great 1105
1069 imposed can use the maximum allowable ones. significance to ensure the tracking accuracy and efficiency. If 1106
the window is too short, it is difficult for the system to elimi- 1107
1070 6.3.3. Setting of collision-free conditions nate large errors in a very short period of time under the 1108
1071 In fact, safety distance is usually prescribed in the trajectory control-constrained condition, which may lead to divergence. 1109
1072 planning module. Hence, when sufficient safety distance is If it is too long, the calculation accuracy and calculation effi- 1110
1073 given, it is unnecessary to consider the collision-free conditions ciency in a single window are usually difficult to be taken into 1111
1074 when designing the MPC controller if external perturbations account. 1112
1075 are not large. Eliminating the collision-free conditions in the
1076 MPC controller could extremely improve the online computa- 6.4. Analogy of trajectory tracking between carrier aircraft and 1113
1077 tional efficiency.29 other vehicles 1114
1078 However, for the sake of safety, one may introduce a con-
1079 straint that restrains the deviation between the actual and the Research on the dispatch trajectory tracking of carrier aircraft 1115
1080 reference trajectories within a relatively small range.56 Such a is rarely seen. However, a lot of research results have been 1116
1081 constraint can be expressed as obtained in the field of UAVs and UGSs. The widely used 1117
1082
1084
2 2
ðxactual  xreference Þ þ ðyactual  yreference Þ 6 r 2
ð29Þ tracking control methods can be divided into geometric and 1118
kinematic tracking control methods,61 classical tracking con- 1119
trol methods,62 methods based on dynamic state feedback,63,64 1120
1085 6.3.4. Design of cost function instantaneous optimal control,65 methods based on neural net- 1121
1086 When designing the MPC controller, one can simply set the work,66,67 fuzzy logic,68,69 etc. 1122
1087 control related term as uT u, i.e., the weighted matrix on control The general idea of trajectory tracking for dispatch on deck 1123
1088 input R is set to be an identity matrix. Thus, the design of is similar to tracking problems for other vehicles. However, the 1124
1089 tracking error-related terms could severely influence the track- motion of carrier aircraft is much susceptible to various kinds 1125

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
16 X. WANG et al.

1126 of external disturbances whose magnitudes may vary accord- neous configurations running on the flight deck simultane- 1181
1127 ing to the position of aircraft on the deck. Since the security ously. A common scenario is that during a sortie task, some 1182
1128 is the ‘‘lifeline” to carrier aircraft, it puts a higher requirement of the carrier aircraft taxi autonomously while others are 1183
1129 on the control performance of the tracking method used. towed by tractors. A more special scenario is that when an 1184
1130 Tractor-aircraft systems can be seen as tractor-trailer sys- autonomous taxiing carrier aircraft breaks down, an extra 1185
1131 tems. However, most studies on the tractor-trailer system tractor is required to quickly tow it away to make clear space 1186
1132 merely focus on the tracking of tractor, and most tracking for following dispatch. In the above two scenarios, dispatch 1187
1133 algorithms can only solve the problem of reversing or forward systems of more than one single configuration are involved. 1188
1134 driving of the system. It is mainly because the kinematics or It leads to the coordinated trajectory planning problem of 1189
1135 dynamics of the system is relatively complex, and the tracking heterogeneous configurations.70,71 Actually, the coordinated 1190
1136 process of switching between reversing and forward driving is path planning methods mentioned in Section 5 are still capable 1191
1137 easy to diverge. The existing methods are difficult to effectively of dealing with the heterogeneity in configurations. However, 1192
1138 solve the trajectory tracking control problem of this process. to the best knowledge of authors, related research is rarely 1193
1139 Therefore, how to make a method robust is a key research seen. On the other hand, the functions of aircraft in a fleet 1194
1140 direction in this field. launching process are generally heterogeneous, and aircraft 1195
of different types should collaborate to execute a task.72 1196
1141 7. Outlook Hence, there might be precedence constraints on the launching 1197
order for aircraft in the fleet. Coordinated path planning tech- 1198

1142 Considering the limits and constraints of the existing trajectory niques that consider precedence constraints, such as in Ref. 59 1199

1143 planning methods of carrier aircraft, future research directions and Ref. 72, can effectively deal with the issue. 1200

1144 are discussed in this section. They are concerned with the com-
1145 pound of different algorithms, path planning of heterogeneous 7.3. Selection between distributed and centralized path planning 1201

1146 dispatch systems, the selection between distributed and cen- methods 1202

1147 tralized path planning methods, and integration with machine


1148 learning techniques. The cooperative path planning methods can be generally 1203
divided into distributed planning methods and centralized 1204
1149 7.1. Compound of different algorithms planning methods.73,74 1205
In centralized planning methods, the state of all individuals 1206

1150 As discussed in Section 4, each path planning algorithm has its must be considered simultaneously. It does not need much 1207

1151 advantages and disadvantages. An appealing idea is to utilize communication between different individuals.75,76 However, 1208

1152 two or more of them to construct a compound path planning it leads to a large-scale planning problem and the correspond- 1209

1153 method where the advantages of them can be explored. There ing solving procedure would be extremely time-consuming.77 1210

1154 are typically two manners to combine two isolated path plan- And the path tracking cannot be implemented until the whole 1211

1155 ning algorithms. solving procedure ends. If extra obstacles appear suddenly in 1212

1156 In the first manner, the whole path is divided into two parts the environment, the previous obtained feasible path fails, 1213

1157 by the prescribed criterion. And two different path planning and the carrier aircraft should slow down and wait for the 1214

1158 algorithms are applied to two parts of the path respectively. result for a new plan. 1215

1159 The algorithm should show its unique advantages in solving Different from centralized cooperative path planning meth- 1216

1160 the path planning problem with characteristics in which part ods, paths of each carrier aircraft are gradually generated in a 1217

1161 it is applied. Liu et al.56 just adopted this manner to combine chronological order in distributed methods.78,79 Thus, when a 1218

1162 the improved Dubins-RVO and the idea of optimal control to small part of the feasible path is generated, the path tracking 1219

1163 develop a compound taxiing path planning method for multi- can be carried out immediately. This leads to the ‘‘simultane- 1220

1164 ple carrier aircraft. ous path planning and tracking”.58 Another advantage of dis- 1221

1165 In the second manner, the results obtained by one algo- tributed methods over centralized methods is that the planning 1222

1166 rithm are taken as the initialization of another algorithm. This can go on when other obstacles suddenly appear. However, the 1223

1167 manner is not used in the path planning of carrier aircraft yet, distributed path planning methods require all individual to 1224

1168 but has been widely used in path planning of AGVs. For exam- communicate with each other frequently, resulting in a high 1225

1169 ple, to solve the path planning of AGV efficiently, Li et al.27 communication burden in practice. 1226

1170 proposed a first-search-then-optimize approach. In Ref. 27, a Since the on-board computing devices nowadays generally 1227

1171 rough path is first generated by the A* algorithm and then cannot meet the requirement of the real-time computational 1228

1172 taken as the initial guesses in optimal control based method, efficiency for centralized planning methods, distributed coop- 1229

1173 thus accelerating the solving procedure of the optimal control erative path planning methods of multiple carrier aircraft draw 1230

1174 problem. much attention than centralized ones. However, the authors do 1231

1175 By introducing a proper compound strategy, it is possible believe that more researchers, in the future, will focus on cen- 1232

1176 to construct a fast planning method which could provide paths tralized planning methods when high-performance computing 1233

1177 of high quality. devices are equipped on the aircraft carriers. 1234

1178 7.2. Path planning of heterogeneous systems on the deck 7.4. Integration with machine learning techniques 1235

1179 The heterogeneous properties can fall into two categories. On Machine learning techniques have been successfully applied to 1236

1180 the one hand, there generally would be vehicles of heteroge- path planning problems in other fields, such as asteroid land- 1237

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
A review on carrier aircraft dispatch 17

1238 ing,80 space debris removal,81 interplanetary trajectory plan- 2. Wu Y, Sun L, Qu X. A sequencing model for a team of aircraft 1293
1239 ning,82 mobile robot,83 etc. However, machine learning tech- landing on the carrier. Aerosp Sci Technol 2016;54:72–87. 1294

1240 niques have not been applied to the path planning of carrier 3. Ryan JC, Cummings ML. A systems analysis of the introduction 1295

1241 aircraft. of unmanned aircraft into aircraft carrier operations. IEEE Trans 1296
Hum-Mach Syst 2016;46(2):209–20. 1297
1242 Sample-based methods, such as SVM,84 deep learning,85
4. Wang N, Meng X, Liu Q, et al. High level architecture based 1298
1243 etc., rely on high-quality and large-scale samples. It is neces- simulation for aircraft carrier deck operations. IEEE advanced 1299
1244 sary to design the appropriate input and output indicators information management, communicates, electronic & automation 1300
1245 for the path planning problem. Actually, the dispatch path control conference; 2016 October 3-5; Xi’an, China. Piscat- 1301
1246 of carrier aircraft and the corresponding operational com- away: IEEE Press; 2017. 1302
1247 mands of the pilot are precious data and can be utilized to con- 5. Michini B, How JP. A human-interactive course of action planner 1303
1248 struct sample-based machine learning based path planning for aircraft carrier deck operations. AIAA Infotech; 2011 March 1304
1249 techniques. However, the record of such data has not drawn 29-31; St. Louis, Missouri. Reston: AIAA; 2011. 1305

1250 much attention. 6. Thate TJ, Michels AS. Requirements for digitized aircraft spotting 1306

1251 Reinforcement learning, as the most popular machine (Ouija) board for use on U.S. Navy aircraft carriers [disserta- 1307
tion]. Monterey: Naval Postgraduate School; 2002. 1308
1252 learning techniques since the last decade, has been widely stud-
7. Johnston JS. A feasibility study of a persistent monitoring system 1309
1253 ied in path planning.86 It can directly interact with the environ- for the flight deck of U.S. Navy Aircraft Carriers [disserta- 1310
1254 ment without establishing an environment model, having any tion]. Maxwell: Air Force Air University; 2009. 1311
1255 prior knowledge, or training with expected samples. Hence, 8. Johnston JS, Swenson E. A persistent monitoring system to reduce 1312
1256 many researchers have applied it in the field of path planning navy aircraft carrier flight deck mishaps. AIAA guidance, naviga- 1313
1257 and achieved a lot of results.87 However, the training process tion, and control conference; 2009 August 10-13; Chicago, Illi- 1314
1258 in reinforcement learning could be extremely time-consuming. nois. Reston: AIAA; 2009. 1315
1259 One of the most concerning issues in machine learning tech- 9. Ryan JC, Banerjee AG, Cummings ML, et al. Comparing the 1316

1260 niques is the generalization capability. Unlike the environment performance of expert user heuristics and an integer linear 1317

1261 for self-driving urban vehicles, the operational environment program in aircraft carrier deck operations. IEEE Trans Cybernet 1318
2014;44(6):761–73. 1319
1262 for dispatch on deck is relatively fixed.17 Thus, a reliable path
10. Zhang Z, Lin S, Dong R, et al. Designing a human-computer 1320
1263 or control commands can be generated quickly once the neural cooperation decision planning system for aircraft carrier deck 1321
1264 network is well trained. One thing to be noted is that various scheduling. AIAA infotech; 2015 January 5-9; Kissimmee, 1322
1265 constraints may not be strictly satisfied in the results obtained Florida. Reston: AIAA; 2015. 1323
1266 by machine learning techniques. Hence, it seems appealing to 11. Zhen Z, Tao G, Jiang J, et al. An adaptive control scheme for 1324
1267 take the obtained results to initialize other path planning carrier landing of UAV. 37th Chinese control conference; 2018 July 1325
1268 methods, as discussed in Section 7.1. 27; Wuhan, China. Piscataway: IEEE Press; 2018. 1326
12. Gajjar BI, Zalewski J. On-ship landing and takeoff of Unmanned 1327
Aerial Vehicles (UAV’S). IFAC Proc Vol 2004;37(20):42–6. 1328
1269 8. Conclusion
13. Kunchev V, Jain L, Ivancevic V, et al. Path planning and obstacle 1329
avoidance for autonomous mobile robots: A review. Lect Notes 1330
1270 The path planning of carrier aircraft on the deck is a key tech- Comput Sci 2006;4252:537–44. 1331
1271 nology for the sortie and recovery task. Based on the present 14. Vinjarapu ASHH, Gawre SK. A survey of autonomous mobile 1332
1272 research and development of carrier aircraft’s dispatch path robot path planning approachesProceeding international confer- 1333
1273 planning technology and the related works, this paper reviews ence on recent innovations is signal processing and embedded 1334

1274 the latest developments and current status of carrier aircraft’s systems. p. 624–8. 1335
15. Aggarwal S, Kumar N. Path planning techniques for unmanned 1336
1275 dispatch path planning on the deck. Furthermore, the analogy
aerial vehicles: A review, solutions, and challenges. Comput 1337
1276 of path planning between aircraft dispatch and other fields
Commun 2020;149:270–99. 1338
1277 such as UGSs is discussed. Finally, the future research direc- 16. Katrakazas C, Quddus M, Chen W, et al. Real-time motion 1339
1278 tions are discussed. By synthetically adopting the path plan- planning methods for autonomous on-road driving: State-of-the- 1340
1279 ning and control techniques, it is promising to realize the art and future research directions. Transport Res C-Emerg 1341
1280 autonomous dispatch for carrier aircraft on deck in the future. 2015;60:416–42. 1342
17. Paden B, Čáp M, Yong SZ, et al. A survey of motion planning and 1343
1281 Acknowledgements control techniques for self-driving urban vehicles. IEEE Trans 1344
Intell Veh 2016;1(1):33–55. 1345
18. Zeng Z, Lian L, Sammut K, et al. A survey on path planning for 1346
1282 The authors are grateful for the financial support of the
persistent autonomy of autonomous underwater vehicles. Ocean 1347
1283 National Key Research and Development Plan (No.
Eng 2015;110(A):303–13. 1348
1284 2017YFB1301103), the National Natural Science Foundation 19. Keymasi KA, Moosavian SAA. Robust adaptive controller for a 1349
1285 of China (Nos. 11922203, 11772074 and 11761131005), and tractor–trailer mobile robot. IEEE-ASME Trans Mech 2014;19 1350
1286 the Fundamental Research Funds for the Central Universities (3):943–53. 1351
1287 (Nos. DUT19TD17 and DUT19TB18). 20. Li B, Shao Z. An incremental strategy for tractor-trailer 1352
vehicle global trajectory optimization in the presence of 1353

1288 References obstaclesIEEE Conference on robotics and biomimetics; 2015 1354


December 6–9; Zhuhai, China. Piscataway: IEEE Press; 2015. 1355
p. 2015. 1356
1289 1. Ryan JC, Cummings ML, Roy N, et al. Designing an interactive
21. Leng Z, Minor MA. Curvature-based ground vehicle control of 1357
1290 local and global decision support system for aircraft carrier deck
trailer path following considering sideslip and limited steering 1358
1291 scheduling. AIAA Infotech; 2011 March 29–31; St. Louis,
actuation. IEEE Trans Intell Transp 2017;18(2):332–48. 1359
1292 Missouri; Reston: AIAA; 2011.

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
18 X. WANG et al.

1360 22. Yuan J, Sun F, Huang Y. Trajectory generation and tracking 43. Zhang Z, Lin S, Xia G, et al. Collision avoidance path planning 1427
1361 control for double-steering tractor–trailer mobile robots with on- for an aircraft in scheduling process on deck. J Harbin Eng Univ 1428
1362 axle hitching. IEEE Trans Ind Electron 2015;62(12):7665–77. 2014;1:9–15 [Chinese]. 1429
1363 23. Wang N, Liu H, Zhou L. Moving characteristics analysis of 44. Zhang Z, Lin S, Qiu B, et al. Collision avoidance path planning of 1430
1364 carrier-based aircraft traction system on deck. J Cent South Univ carrier aircraft traction system in dispatching on deck. Syst Eng 1431
1365 2013;44(6):2304–10. Electron 2014;36(8):1551–7 [Chinese]. 1432
1366 24. Wang N, Liu H, Yang W. Path-tracking control of a tractor- 45. Han W, Si W, Ding D, et al. Multi-routes dynamic planning on 1433
1367 aircraft system. J Mar Sci Appl 2012;11(4):512–7. deck of carrier plane based on clustering PSO. J Beijing Univ 1434
1368 25. Karkee M, Steward BL. Study of the open and closed loop Aeronaut Astronaut 2013;39(5):610–4 [Chinese]. 1435
1369 characteristics of a tractor and a single axle towed implement 46. Wu Y, Hu N, Qu X. A general trajectory optimization method for 1436
1370 system. J Terramech 2010;47(6):379–93. aircraft taxiing on flight deck of carrier. Proc Inst Mech Eng G-J 1437
1371 26. Li B, Wang K, Shao Z. Time-optimal maneuver planning in Aer 2018;233(4):1340–53. 1438
1372 automatic parallel parking using a simultaneous dynamic opti- 47. Su X, Li Z, Song J, et al. A path planning method for carrier 1439
1373 mization approach. IEEE T Intell Transp 2016;17(11):3263–74. aircraft on deck combining artificial experience and intelligent 1440
1374 27. Fujimura K, Samet H. A hierarchical strategy for path planning search. IOP Conf. Series: Mater Sci Eng 2018;381(1) 012194. 1441
1375 among moving obstacles. IEEE Trans Robot Autom 1989;5 48. Zhang Z, Lin S, Zhu Q, et al. Genetic collision avoidance planning 1442
1376 (1):61–9. algorithm for irregular shaped object with kinematics constraint. 1443
1377 28. Liu J, Han W, Liu C, et al. A new method for the optimal control Acta Aeronaut Astronaut Sin 2015;36(4):1348–58 [Chinese]. 1444
1378 problem of path planning for unmanned ground systems. IEEE 49. Liu J, Han W, Zhang Y, et al. Design of an online nonlinear 1445
1379 Access 2018;6:32251–60. optimal tracking control method for unmanned ground systems. 1446
1380 29. Liu J, Han W, Peng H, et al. Trajectory planning and tracking IEEE Access 2018;6:65429–38. 1447
1381 control for towed carrier aircraft system. Aerosp Sci Technol 50. Li Y, Zhu Y, Li Q. Analysis of aircraft path planning optimal on 1448
1382 2018;84:830–8. carrier flight deck. Adv Mater Res 2013;664:1122–7. 1449
1383 30. Shi B, Peng H, Wang X, et al. Optimal motion planning of four- 51. Li Y, Li Z, Zhu Y. Genetic algorithm based aircraft spotting 1450
1384 wheeled trailer systemProceedings of the 38th Chinese control allocation optimal scheduling approach on carrier flight deck. 1451
1385 conference; 2019 July 27-30; Guangzhou, China. Piscataway: IEEE Appl Mech Mater 2013;373–375:1196–201. 1452
1386 Press; 2019. p. 1856–62. 52. Ross IM, Karpenko M. A review of pseudospectral optimal 1453
1387 31. Park M, Chung W, Kim M, et al. Control of a mobile robot with control: From theory to flight. Annu Rev Control 2012;36 1454
1388 passive multiple trailersProceedings of 2004 IEEE international (2):182–97. 1455
1389 conference on robotics and automation (ICRA ’04); 2004 26 April-1 53. Peng H, Wang X, Li M, et al. An hp symplectic pseudospectral 1456
1390 May; New Orleans, LA. Piscataway: IEEE Press; 2004. p. 4369–74. method for nonlinear optimal control. Commun Nonlinear Sci 1457
1391 32. Johnston JS, Swenson ED. Feasibility study of global-positioning- 2017;42:623–44. 1458
1392 system-based aircraft-carrier flight-deck persistent monitoring 54. Wang X, Peng H, Zhang S, et al. A symplectic pseudospectral 1459
1393 system. J Aircraft 2010;47(5):1624–35. method for nonlinear optimal control problem with inequality 1460
1394 33. Ross IM. A Beginner’s Guide to DIDO (Ver. 7.3), A MATLAB constraints. ISA Trans 2017;68:335–52. 1461
1395 application package for solving optimal control problems. Elissar, 55. Peng H, Wang X, Shi B, et al. Stabilizing constrained chaotic 1462
1396 LLC; 2007. system using a symplectic psuedospectral method. Commun 1463
1397 34. Liu J, Han W, Xu W, et al. Optimal path tracking control of Nonlinear Sci 2018;56:77–92. 1464
1398 carrier-based aircraft on the deck based on RHC. Acta Aeronaut 56. Liu J, Han W, Wang X, et al. Research on cooperative trajectory 1465
1399 Astronaut Sin 2019;40(8) [Chinese]. planning and tracking problem for multiple carrier aircraft on the 1466
1400 35. Zhang J, Yu J, Qu X, et al. Path planning for carrier aircraft based deck. IEEE Syst J 2020;14(2):3027–38. 1467
1401 on geometry and Dijkstra’s algorithm. Third IEEE international 57. Li Y, Wu Y, Su X, et al. Path planning for aircraft fleet launching 1468
1402 conference on control science and systems engineering; 2017 August on the flight deck of carriers. Mathematics-Basel 2018;6(10):175. 1469
1403 17-19; Beijing, China. Piscataway: IEEE Press; 2017. 58. Wu Y, Wang Y, Qu X, et al. Exploring mission planning method 1470
1404 36. Zhang J, Wu Y, Qu X. Path planning method for traction system for a team of carrier aircraft launching. Chin J Aeronaut 2019;32 1471
1405 on carrier aircraft. J Beijing Univ Aeronaut Astronaut 2018;44 (5):203–14. 1472
1406 (10):2125–33 [Chinese]. 59. Liu J, Han W, Li J, et al. Integration design of sortie scheduling 1473
1407 37. Si W, Qi Y, Han W. Carrier plane transportation in hangar based for carrier aircrafts based on hybrid flexible Flowshop. IEEE Syst 1474
1408 on convex hull algorithm combined with Dijkstra. Syst Eng J 2020;14(1):1503–11. 1475
1409 Electron 2015;37(3):583–8 [Chinese]. 60. Berg JVD, Lin M, Manocha D. Reciprocal velocity obstacles for 1476
1410 38. Wu Y, Qu X. Path planning for taxi of carrier aircraft launching. real-time multi-agent navigation. IEEE international conference on 1477
1411 Sci China Technol Sc 2013;56(6):1561–70. robotics and automation; 2008 May 19-23; Pasadena. Piscat- 1478
1412 39. Wu Y, Qu X. Obstacle avoidance and path planning for carrier away: IEEE Press; 2008. 1479
1413 aircraft launching. Chin J Aeronaut 2015;28(3):695–703. 61. Mike M, Kenny L, Pascal S, et al. Stanley: The robot that won the 1480
1414 40. Panov AI, Yakovlev K. Behavior and path planning for the DARPA grand challenge. J Field Robot 2006;23(9):661–92. 1481
1415 coalition of cognitive robots in smart relocation tasks. 4th 62. Pan Y, Li X, Yu H. Efficient PID tracking control of robotic 1482
1416 International conference on robot intelligence technology and manipulators driven by compliant actuators. IEEE Trans Contr 1483
1417 applications; 2015 December 14-16; Bucheon, Korea; Cham: Syst Technol 2018;27(2):915–22. 1484
1418 Springer; 2015. 63. Ajjanaromvat N, Parnichkun M. Trajectory tracking using online 1485
1419 41. Xiu C, Chen H. A behavior-based path planning for autonomous learning LQR with adaptive learning control of a leg-exoskeleton 1486
1420 vehicle. The 2010 international conference on intelligent robotics for disorder gait rehabilitation. Mechatronics 2018;51:85–96. 1487
1421 and applications; 2010 November10-12; Shanghai, China; Berlin: 64. Li Z, Deng J, Lu R, et al. Trajectory-tracking control of mobile 1488
1422 Springer; 2010. robot systems incorporating neural-dynamic optimized model 1489
1423 42. Ye B, Zhao M, Wang Y. Research of path planning method for predictive approach. IEEE Trans Syst Man Cybern Syst 2016;46 1490
1424 mobile robot based on artificial potential field. 2011 International (6):740–9. 1491
1425 conference on multimedia technology; 2011 July 26-28; Hangzhou, 65. Peng H, Li F, Liu J, et al. A Symplectic instantaneous optimal 1492
1426 China; Piscataway: IEEE Press; 2011. control for robot trajectory tracking with differential-algebraic 1493
equation models. IEEE Trans Ind Electron 2020;67(5):3919–4829. 1494

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020
CJA 1668 No. of Pages 19
13 July 2020
A review on carrier aircraft dispatch 19

1495 66. Moreno-Valenzuela J, Aguilar-Avelar C, Puga-Guzmán SA, et al. planning of connected and automated vehicles. J Intell Transport 1535
1496 Adaptive neural network control for the trajectory tracking of the Syst 2019;23(6):1–12. 1536
1497 Furuta pendulum. IEEE Trans Cyber 2016;46(12):3439–52. 78. Lázaro JL, Garcı´a JC, Mazo M, et al. Distributed architecture for 1537
1498 67. Lungu R, Lungu M. Automatic landing system using neural control and path planning of autonomous vehicles. Microprocess 1538
1499 networks and radio-technical subsystems. Chin J Aeronaut 2017;30 Microsyst 2001;25(3):159–66. 1539
1500 (1):399–411. 79. Luo Q, Duan H. Distributed UAV flocking control based on 1540
1501 68. Zhao X, Wang X, Li Ma, et al. Fuzzy-approximation-based homing pigeon hierarchical strategies. Aerosp Sci Technol 1541
1502 asymptotic tracking control for a class of uncertain switched 2017;70:257–64. 1542
1503 nonlinear systems. IEEE Trans Fuzzy Syst 2020;28(4):632–44. 80. Zhang P, Liu X, Zhao B, et al. Asteroid landing via onboard 1543
1504 69. Wang Y, Chang Y, Alkhateeb AF, et al. Adaptive fuzzy output- optimal guidance based on bidirectional extreme learning 1544
1505 feedback tracking control for switched nonstrict-feedback nonlin- machineInternational joint conference on neural networks; 2016 1545
1506 ear systems with prescribed performance. Circ Syst Signal Pr 2020. July 24-29; Vancouver. Piscataway: IEEE Press; 2016. p. 609–15. 1546
1507 https://doi.org/10.1007/s00034-020-01466-y. 81. Klima R, Bloembergen D, Savani R, et al. Space debris removal: 1547
1508 70. Bae J, Chung W. A heuristic for path planning of multiple Learning to cooperate and the price of anarchy. Front Robot AI 1548
1509 heterogeneous automated guided vehicles. Int J Precis Eng Man 2018;5:54. 1549
1510 2018;19(12):1765–2711. 82. Izzo D, Sprague C, Tailor D. Machine learning and evolutionary 1550
1511 71. Wang J, Zhang Y, Geng L, et al. A heuristic mission planning techniques in interplanetary trajectory design. Springer Optim 1551
1512 algorithm for heterogeneous tasks with heterogeneous UAVs. Appl 2019;144:191–210. 1552
1513 Unmanned Syst 2015;3(3):205–19. 83. Sombolestan SM, Rasooli A, Khodaygan S. Optimal path- 1553
1514 72. Jia Z, Yu J, Ai X, et al. Cooperative multiple task assignment planning for mobile robots to find a hidden target in an unknown 1554
1515 problem with stochastic velocities and time windows for hetero- environment based on machine learning. J Amb Intel Hum Comp 1555
1516 geneous unmanned aerial vehicles using a genetic algorithm. 2019;10(5):1841–50. 1556
1517 Aerosp Sci Technol 2018;76:112–25. 84. Gireesh KT, Poornaselvan KJ, Sethumadhavan M. Fuzzy support 1557
1518 73. Giardini G, Kalmar-Nagy T. Centralized and distributed path vector machine-based multi-agent optimal path planning 1558
1519 planning for multi-agent exploration. AIAA guidance, navigation approach to robotics environment. Defence Sci J 2010;60 1559
1520 and control conference and exhibit; 2007 August 20-23; Hilton (4):387–91. 1560
1521 Head, South Carolina; Reston: AIAA; 2007. 85. Lu Y, Yi S, Liu Y, et al. A novel path planning method for 1561
1522 74. Wang Y, Li X, Yao H. Review of trajectory optimisation for biomimetic robot based on deep learning. Assembly Autom 1562
1523 connected automated vehicles. IET Intell Transp Syst 2019;13 2016;36(2):186–91. 1563
1524 (4):580–6. 86. Benhlima S, Lamini C, Fathi Y. H-MAS architecture and 1564
1525 75. Jose K, Pratihar DK. Task allocation and collision-free path reinforcement learning method for autonomous robot path 1565
1526 planning of centralized multi-robots system for industrial plant planning. 2017 Intelligent Systems and Computer Vision (ISCV); 1566
1527 inspection using heuristic methods. Robot Auton Syst 2017 April 17-19; Fez, Morocco. IEEE Press; 2017. 1567
1528 2019;80:34–42. 87. Tai L, Paolo G, Liu M. Virtual-to-real deep reinforcement 1568
1529 76. Borrelli F, Subramanian D, Raghunathan AU, et al. MILP and learning: Continuous control of mobile robots for mapless 1569
1530 NLP techniques for centralized trajectory planning of multiple navigation. 2017 IEEE/RSJ international conference on intelligent 1570
1531 unmanned air vehicles. American control conference; 2006 June 14- robots and systems (IROS); 2017 September 24-28; Vancouver. Pis- 1571
1532 16; Minneapolis. Piscataway: IEEE Press; 2006. cataway: IEEE Press; 2017. 1572
1533 77. Li B, Jia N, Li P, et al. Incrementally constrained dynamic 1573
1534 optimization: A computational framework for lane change motion

Please cite this article in press as: WANG X et al. A review on carrier aircraft dispatch path planning and control on deck, Chin J Aeronaut (2020), https://doi.org/
10.1016/j.cja.2020.06.020

You might also like