Original Article

Proc IMechE Part G:

J Aerospace Engineering
Engagement time constrained 0(0) 1–19
! IMechE 2020

formation generation guidance Article reuse guidelines:

sagepub.com/journals-permissions

following a non-manoeuvring leader DOI: 10.1177/0954410020909209

journals.sagepub.com/home/pig

Shalini Suresh and Ashwini Ratnoo

Abstract
This paper addresses the problem of formation generation of two UAVs with constraints on engagement time. A leader–
follower scenario with the leader being non-manoeuvring is considered. The work uses track guidance method for
generating a class of follower spatial paths and combines it with a speed profile that works to minimize the error with
respect to the desired follower position. Path curvature analysis is carried out and set of feasible spatial paths which
abide by the UAV turn rate constraint is obtained. The equilibrium point of the resulting nonlinear system is found to be
the desired formation geometry and stability analysis using Lyapunov direct method guarantees convergence to the
equilibrium point. The set of achievable engagement times is deduced with closed-form limits. Extensive simulation
exercises are carried out incorporating follower speed limits, wind and vehicle dynamics, and presence of obstacles.
The work highlights a flexible guidance method which effectively uses one design parameter and analytic conditions for
formation generation.

Keywords
UAV formation generation, guidance with engagement time constraint

Date received: 23 January 2019; accepted: 8 January 2020

The leader–follower approach considers one UAV as

Introduction leader and the remaining as followers. This architecture
Unmanned aerial vehicle (UAV) formations address a is easy to implement as it simplifies the formation
wide range of applications like surveillance, geographic generation problem into a tracking problem for the fol-
mapping, rescue operations, fire detection, reconnais- lowers. Also, leader–follower approach allows decentra-
sance, etc. In dynamic and airspace restricted environ- lized control wherein the followers exclusively use the
ments, time taken in reaching the formation could be information obtained using the on-board sensors. Tahk
crucial for addressing the aforementioned applications. et al.9 and Betser et al.10 present leader–follower forma-
In that context, a guidance method which could provide tion generation algorithms using the line-of-sight infor-
flexibility in terms of controlling time-to-formation is mation wherein the follower speed and heading
desirable. guidance commands are generated based on the error
Methods for generating UAV formations can in the desired and actual range and line-of-sight angle,
be broadly classified into behaviour-based approach, respectively. Considering paths leading to desired fol-
virtual structure method, potential field-based approach lower relative location, vector field-based path following
and leader–follower framework. Behaviour-based research studies11–14 were also used to attain the forma-
approach employs individual behaviour for each vehicle tion. The vector field guidance commands are typically
and the final control is derived in such a way that the computed as local heading directions. Modified for path
entire group regains formation whenever disturbed, following, various classical guidance laws have also
avoids the obstacles and completes the task.1,2 The vir-
tual structure method considers the entire formation as Autonomous Vehicles Laboratory, Department of Aerospace
a rigid body with a reference point with respect to which Engineering, Indian Institute of Science, Bangalore, India
the desired position of each vehicle can be calculated
Corresponding author:
and the formation can be achieved by using a suitable
Shalini Suresh, Autonomous Vehicles Laboratory, Department of
controller.3,4 Potential field-based approaches use spe- Aerospace Engineering, Indian Institute of Science, Bangalore 560012,
cific direction and magnitude of a potential field to Karnataka, India.
achieve a desired formation shape and orientation.5–8 Email: [email protected]
2 Proc IMechE Part G: J Aerospace Engineering 0(0)

found application in formation generation considering noteworthy that the work considers a simplistic non-
leader–follower framework. These path following meth- manoeuvring leader motion model while deriving the
ods use pursuit guidance,15–19 line-of-sight guidance,20 key results. However, the significance of the work lies
non-linear guidance laws based on proportional naviga- in the fact that the proposed method directly relates the
tion21 and trajectory shaping guidance.22 These methods engagement time, turn rate constraint and obstacle
are based on guidance logics pursuing a virtual target avoidance feature to a single guidance parameter
which can potentially lead the follower to the desired using closed-form relations. This imparts high imple-
relative location with respect to the leader. A track guid- mentation value to the proposed method. Further
ance algorithm is discussed in Lee et al.23,24 for generat- strengthening the contributions, robustness of the
ing a class of exponential paths which converge to a method is verified using simulations considering kine-
desired trackline. By considering the trackline to pass matic autopilot models and the presence of wind.
through the desired relative location, the paths were Preliminary results of this work were presented in
used for attaining a formation. Suresh and Ratnoo30 comprising the basic idea and
While most of the relevant literature deals with gen- sample simulations. That work did not consider ana-
erating and maintaining UAV formations, time-to-for- lysis of the resulting dynamics, turn rate constraint and
mation control has received some attention. Koo extensive validation of the method.
et al.25 presented a control law design methodology The remainder of the paper is arranged as follows:
for UAV formation generation classified as two scen- the following section presents the problem statement
arios. The first one considers the followers reaching the followed by the proposed guidance method. The suc-
respective desired positions in a specified sequence. cessive sections discuss the engagement time control
And the other scenario considers a simultaneous arri- framework and simulation studies. Concluding
val at the desired relative positions. The method con- remarks are provided in the last section.
siders three independent acceleration control inputs for
controlling the respective position error in three dimen-
sions and the time of approach depends on the gain of
Problem statement
the error regulation-based controller. Ryoo et al.26 pro- Consider a planar initial engagement geometry with
pose a formation generation guidance law obtained as two UAVs as shown in Figure 1. Here, F and L
a feedback solution to a linear quadratic regulator denote the follower and the leader, respectively.
problem which minimizes a manoeuver induced cost A coordinate system is defined with its origin at the
together with constraint on terminal position and vel- initial follower location and the X axis along the lea-
ocity. While designing the guidance laws, Koo and der’s velocity direction. Initially, the leader is at a sep-
Shahruz25 and Ryoo et al.26 do not consider vehicle aration of ðxl0 , yl0 Þ from the follower. Speed of the
speed limits as well as turn rate constraint which is follower is denoted by vf, and vl represents the speed
important from a realistic point of view. of the non-manoeuvring leader. Heading directions of
The focus of this work is on flyable guidance solu- the follower and the leader are denoted by wf and wl,
tions which incorporate vehicle speed and turning rate respectively, as measured clockwise from the Y axis.
constraints into the problem. Research studies27–29 The desired formation geometry is defined in terms of
present a two-step planning solution to the engage- two parameters, that is, lateral clearance, lc and for-
ment-time constrained UAV coordination problem. ward clearance, fc between the vehicles. The desired
Each vehicle is assigned a feasible spatial path relative follower position with respect to the leader, D
described by algebraic polynomial of a virtual arc is marked in Figure 1. While abiding by its maximum
length parameter and a speed profile. The next step
involves solving a constrained optimization problem
considering vehicle dynamic constraints and the path
length. It is noteworthy that a two-vehicle planar
scenario would require evaluating eight polynomial
coefficients and a numerical solution for the optimiza-
tion problem following that approach.
In contrast to the existing two-step approach invol-
ving a numerical solution, this work proposes a unified
and easily computable guidance solution to the multi-
constrained problem. Specifically, the work proposes
time constrained UAV formation generation guidance
logic in a non-manoeuvring leader–follower frame-
work. Formation is achieved by combining feasible
follower spatial paths and an adaptive speed profile
that works independent of the path. Stability analysis
of the relative engagement kinematics guarantees
convergence to the desired formation geometry. It is Figure 1. Engagement scenario.
Suresh and Ratnoo 3

turn rate and speed constraints, the follower should passing through the desired follower position along
reach the desired relative position with speed and the desired heading direction as shown in Figure 2.
heading direction identical to that of the leader. Track guidance method generates paths that exponen-
The objective here is to propose a prospective guid- tially converge to the track line. The paths are gov-
ance method for the follower to attain formation with erned by the spatial relationship
the aforementioned constraints together with a con-
trol over the time taken to reach the formation. y ¼ yfinal ð1  ex=k Þ ð1Þ

where (x, y), yfinal and k denote the instantaneous

Proposed guidance method position of the follower, the shortest distance of the
track line from the initial follower location and
Follower spatial trajectory the spatial guidance constant, respectively. Figure 2
This subsection briefly reviews the class of spatial also presents a typical path generated using equation
paths generated in Lee et al.23,24 A track line is defined (1). The spatial guidance constant, k determines the
rate at which the path converges to the desired track
line. Using equation (1), for x ¼ k, 3k, 4k and 5k, y
attains 63.2%, 95%, 98.2% and 99.3% respectively,
of yfinal . Accordingly, lower the value of k, steeper is
the convergence of the path to the track line. Figure 3
shows three such paths generated using different
values of k with yfinal ¼ 100 m. The instantaneous
heading of the follower, wf at any point on the path
can be obtained as

 
 1 dy
f ¼  tan ð2Þ
2 dx

From equation (1), it can be readily deduced that

Figure 2. Follower spatial path generated using track dy yfinal  y yfinal ex=k
guidance.
¼ ¼ ð3Þ
dx k k

Figure 3. Path variation with k.

4 Proc IMechE Part G: J Aerospace Engineering 0(0)

Using equation (3) in equation (2) leads to of the follower to the desired position in terms of the
  heading.
k
f ¼ tan1 ð4Þ
yfinal ex=k Curvature characteristics of track paths
At x ¼ 0, the initial follower heading is given by Curvature characteristics directly relate to the turning
  rate of the UAV and hence needs to be evaluated for
k designing flyable UAV paths. Curvature, C of a
f0 ¼ tan1 ð5Þ
yfinal planar curve can be expressed as

It can be seen from equation (4) that as x ! 1, j yðxÞ00 j

C¼ ð8Þ
wf ! 90 eventually driving the relative heading ð1 þ y0ðxÞ2 Þ
3=2
error to zero. Since, theoretically it takes infinite
horizontal displacement for convergence to the Taking the first and second derivatives of y(x)
track line, a practical final point with an error in given by equation (1) and substituting that in equa-
y with respect to yfinal as 0.01 m is considered. Using tion (8) results in
equation (1), this
 corresponds to a displacement of
x ¼ k ln 0:01
along x direction. Also, imposing a kyfinal ex=k
yfinal C¼ 3=2 ð9Þ
 2
k2 þ ðyfinal ex=k Þ
  angle constraint, wf589:99 at
desired heading
x ¼ k ln y0:01
final
in equation (4) results in a range
of feasible spatial guidance constants as Curvature characteristics of the track paths given
by equation (1) are plotted for different values of k in
 Figure 4. Evaluating @C
k  @x ¼ 0 leads to
   5 tanð89:99 Þ ð6Þ
     
yfinal ex=k
x¼k ln 0:01 y  y
yfinal final x=k 2 x=k 2 final x=k 2
 1þ e yfinal e k 2 e ¼0
k k
) k557:3 ð7Þ ð10Þ
 2 
The spatial paths conforming to equation (7) con- k k
stitute the proposed formation guidance algorithm )x¼ ln or x ! 1 ð11Þ
2 2y2final
feasible set of paths which ensure proper convergence

Figure 4. Curvature characteristics.

Suresh and Ratnoo 5

pffiffiffi
From equations (9) and (11), it can be deduced that maximum curvature for the paths with k 4 2yfinal
the curvature of the path increases monotonously
  is at x ¼ 0 as given by
k2
to reach the maximum value at x ¼ k
2 ln 2y2final
and
kyfinal
then decreases approaching zero as x ! 1. Cjðx¼0Þ ¼ 3=2
ð15Þ
Maximum k2 þ y2final
 curvature,
 Cmax can be obtained by using
k2
x ¼ k
2 ln 2y2final
in equation (9) as Using equation (5) in equation (15) leads to

    kyfinal
1
2 ln
k2 Cðx¼0Þ ¼ ð16Þ
kyfinal e
2y2
final ðk= sin f0 Þ3
Cmax ¼ 0    12 13=2 ð12Þ
0
1 k2
B 2 ln
C C Using equation (13) and equation (16), the max-
Bk2 þ B
@yfinal e
2y2
final
A C
@ A imum curvature on the follower track path can be
classified as
8 pffiffiffi
2 < p2ffiffi for k4 2yfinal
) Cmax ¼ pffiffiffi ð13Þ Cmax ¼
3 3k
pffiffiffi ð17Þ
3 3k : kyfinal 3 for k 4 2yfinal
ðk= sin f0 Þ
Using equation (11), the position at which the
curvature is maximum can be classified as A typical variation of maximum curvature with k
( as given by equation (17) is shown in Figure 5.
pffiffiffi
50 for k4 2yfinal Accordingly, the minimum value of the spatial guid-
xCmax pffiffiffi ð14Þ ance constant that satisfies a pre-specified curvature
50 for k 4 2yfinal
constraint, CmaxUAV can be obtained as
8 pffiffiffi
It is noteworthy that the present application con- < 3pffiffi3C2 for k4 2yfinal
maxUAV
siders only the part of the exponential curve satisfying kminC ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffi ð18Þ
x 5 0. Since the curvature decreases monotonically : yfinal sin3 f0 for k 4 2yfinal
CmaxUAV
after the maximum value, using equation (14), the

Figure 5. Variation of maximum curvature with k.

6 Proc IMechE Part G: J Aerospace Engineering 0(0)

Follower speed profile with initial conditions satisfying vfx > 0, f 4 0 and
The track guidance philosophy presents a class of spa- xd  x5  , where d is a small positive value.
tial paths converging to the desired path. However, it The system has only one equilibrium state satisfying
ignores the time at which the follower reaches a Fðx e Þ ¼ 0 which can be readily deduced using equa-
desired position as well as any speed constraints for tion (23) as
maintaining the formation thereafter. The follower 2 3
speed profile along the x direction, vfx, is proposed vl
6 7
as a function of the position error with respect to x e ¼ 4 =2 5 ð24Þ
the desired location xd ¼ xl  fc , as 0
vl  xd x

vfx ¼ vl þ 1  e kv ð19Þ
r Stability analysis. The stability of the system with
respect to the equilibrium state can be analysed
where vl and kv represent magnitude of the leader’s using a candidate Lyapunov function given by
velocity and the follower velocity guidance constant,
respectively. The positive constant, r, known as the 1
L¼ ðxd  xÞ2 þ ð=2  fÞ
2
þ ðvl  vfx Þ2 ð25Þ
speed advantage factor limits the maximum and min- 2
imum values that vfx can attain. It can be chosen con- h  
sidering the fixed wing UAV speed constraint. Here, where ðxd  xÞ 2 ,  k ln 0:01
, ð=2  fÞ 2 ½0, =2Þ
yfinal
xd denotes the desired x position of the follower.
and ðvl  vfx Þ 2 ðvl =r, vl Þ. It can be seen that
From equation (19), the follower speed is theoretically
bounded with a maximum value of rþ1   
r vl for a very 0:01
large error ðxd  xÞ. At the desired location, x ¼ xd, L50 for ðxd  xÞ 2 ,  k ln ,
yfinal
which using equation (19) implies vfx ¼ vl . The con-
stant, kv determines the rate of convergence (with ð=2  fÞ 2 ½0, =2Þ and ðvl  vfx Þ 2 ðvl =r, vl Þ
respect to ðxd  xÞ) of vfx to vl. ð26Þ

To obtain the range of kv that ensures practical where the equality holds if and only if x, f , vfx ¼
convergence of the follower velocity, a terminal ðxd , =2, vl Þ. Hence, the Lyapunov function is positive
speed constraint is imposed as definite over the respective domains. Differentiating
equation (25) leads to
ðvfx  vl Þjðxd xÞ¼0:01 40:01 ð20Þ : :
L_ ¼ ðxd  xÞðxd xÞ _ þ ð=2  f Þð fÞ
: ð27Þ
Using equation (19) in equation (20) results in þ ðvl  vfx Þð vfx Þ

1 xd x
 0:01 0:01 Using equation (23) in equation (27) becomes
1þ 1  e kv 4 þ 1 ) kv 4  
r vl ln 1  0:01r  
vl vfx cos f sin f
L_ ¼ ðxd  xÞðvl  vfx Þ  ð=2  f Þ
ð21Þ k
: 
xd x_ xdkx
 ðvl  vfx Þ vl e v
Equation (21) presents the constraint on the choice rkv
of velocity guidance constant satisfying the final speed ð28Þ
requirements.
Equation (28) can be modified as
v  
Equilibrium point and stability analysis L_ ¼ ðxd  xÞ
l xd x
1  e k v
r 
Equilibrium point. The kinematic model of the leader vfx cos f sin f
follower formation geometry subject to equation (1)  ð=2  f Þ ð29Þ
k
and equation (19) can be expressed in the form of  
non-linear state space equation as vl xdkx
 ðvl  vfx Þ2 e v
rkv
x_ ¼ FðxÞ
 ð22Þ   
_ 0:01
) L40 for ðxd  xÞ 2 ,  k ln ,
where yfinal
ð=2  fÞ 2 ½0, =2Þ and ðvl  vfx Þ 2 ðvl =r, vl Þ
2 3 2 : x x 3
xd x_  dkv ð30Þ
vfx rkv vl e
6 7 6 7
x ¼ 4 f 5 and  ¼6
FðxÞ 4
vfx cos f sin f 7
5 ð23Þ The equality holds if and only if ðxd  xÞ ¼ 0,
k
xd  x ð f  =2Þ ¼ 0 and ðvfx  vl Þ ¼ 0 which is the equilib-
vl  vfx rium state defined by equation (24).
Suresh and Ratnoo 7

Controlling the engagement time Algorithm 1: Algorithm for engagement time con-
 
With xl0 5k ln y0:01 þ fc , the proposed velocity pro- strained formation generation guidance
final
1: Use kmin as maxð57:3, kminC Þ to compute tfmin
file together with the spatial path given by equation using equation (33)
(1) can bring the follower to the
 desired
 location with 2: Define set St ¼ ½tfmin , 1Þ
0:01
negligible errors at x ¼ k ln yfinal . The leader loca- 3: Choose a t 2 St
tion at that instant can be expressed as 4: Compute k using equation (32)
5: Choose a kv satisfying equation (21) and an r > 0
 
0:01
xl ¼ k ln þ fc ð31Þ
yfinal Feasible leader initial position
where fc is the forward clearance between the vehi- Note that the equation (32) assumes that the follower
cles in the desired formation geometry. The time reaches the desired position at the instant the leader
taken by the leader to reach this position can be position satisfies equation (31). This would require a
obtained as greater follower speed advantage for leader initial
h  i positions close to
k ln y0:01
final
þ fc  xl0  
tf ¼ ð32Þ 0:01
vl x ¼ k ln þ fc ð34Þ
yfinal

Equation (32) shows that formation could be gen- Secondly, a leader initial position too far away
erated for a set of engagement times by varying the from its position at tf would require a much lower
spatial guidance constant, k 2 ½kmin , 1Þ where kmin is follower speed as governed by equation (19). In prac-
maxð57:3, kminC Þ, satisfying the constraint given by tice fixed wing UAVs have minimum and maximum
equation (7) and the curvature constraint given by operational speed limits and that would affect the
equation (18). feasible leader initial positions for a particular value
Hence, the minimum achievable engagement time of k for the proposed guidance method to work.
can hence be obtained as Figure 6 presents the engagement geometry marking
h  i the feasible leader initial positions.
kmin ln y0:01
final
þ fc  xl0
tfmin ¼ ð33Þ Follower maximum speed constraint. Owing to the max-
vl imum speed constraint, the follower will fail to
achieve formation at the time given by equation (32)
Algorithm 1 summarizes the guidance logic for for initial leader locations closer to the point given by
generating engagement time constrained formations. equation (34). Hence, xl0max is defined as an upper

Figure 6. Scenario with feasible leader initial positions.

8 Proc IMechE Part G: J Aerospace Engineering 0(0)

limit to the feasible leader initial x position. The max- determined by imposing.
imum feasible leader initial x position can be obtained
by iteratively checking the time taken by the two vehi- vfx
vf ðt ¼ 0Þ ¼ ðt ¼ 0Þ5vfstall ð35Þ
cles to reach the desired formation. Algorithm 2 pre- sin
sents a method for obtaining xl0max for a given speed
advantage factor r and guidance constant, k. Equations (4) and (19) in equation (35) leads to
 xd0

Algorithm 2: Algorithm for obtaining xl0max vl þ vrl 1  e kv
1: Inputs 5vfstall ð36Þ
k, yfinal , fc, r, x   pffiffiffiffiffiffiffiffiffiffiffiffi
2
k
2
ffi
k þyfinal
2: Set xl0 ¼ k ln y0:01 þ fc .
final
0 1
3: Compute tf using equation (32). B rkvfstall C
) xd0 5  kv ln@ð1 þ rÞ  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA ð37Þ
4: Compute time taken
  by the follower, tfollower to vl k2 þ y2final
reach x ¼ k ln y0:01
final
,

5: if jtf  tfollower j 5 104 s then, output xl0max ¼ xl0 . Adding positive forward clearance to the min xd0
6: else, update xl0 ¼ xl0  x, results in
7: Repeat steps 2 and 3. 0 1
8: end if
B rkvfstall C
xl0min ¼ kv ln@ð1 þ rÞ  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA þ fc
Figure 7(a) shows the variation of xl0max with the vl k2 þ y2final
speed advantage factor, r. The variation considers
yfinal ¼ 75 m, fc ¼ 25 m and k ¼ 57.3. The effect of ð38Þ
r on the maximum follower velocity is shown in
Figure 7(b). It can be seen that with a greater speed Figure 8(a) shows the effect of r on xl0min for a fixed
advantage, the follower can achieve formation for stall speed of 8 m/s. A lower value of xl0min is attained
leader initial positions
 closer
 to the theoretical max- for lower choice of r. Consider r ¼ 2. Figure 8(b)
imum of x ¼ k ln 0:01
þ fc ¼ 536:3 m. shows the variation in xl0min with vfstall. It can be
yfinal
seen that a lower stalling speed leads to reduction in
xl0min . It is noteworthy that, overall, a reduction in r
Follower stalling speed constraint. It is evident from equa- increases the range of feasible leader initial positions.
tion (19) that the follower speed decreases below the The framework considered in this work assumes a
leader speed when ðxd  xÞ becomes negative. Further two-UAV scenario. With the present work being a
negative values of ðxd  xÞ may lead to stalling of the building block, multiple track lines can be used for
follower UAV. Considering the stall speed of the fol- developing a multi-vehicle formation generation
lower, the minimum leader initial x position can be method. In that regard, a collision-avoidance

(a) (b)

vf max
Figure 7. xl0max characteristics. (a) Variation of xl0max with r; (b) Variation of vl with r.
Suresh and Ratnoo 9

(a) (b)

Figure 8. xl0min characteristics. (a) Variation of xl0min with r; (b) Variation of xl0min with vfstall.

guarantee would be an additional requirement for scal- computed using equation (32) as k ¼ 81.25 which,
ing up the existing framework. using equation (5), results in f0 ¼ 47:3 .
Considering a non-manoeuvring leader is a limita- The resulting leader and follower trajectories are
tion of this work. Manoeuvring leader scenarios can shown in Figure 9(a). The follower traces the expo-
potentially use an instantaneous track-line construc- nential curve defined by equation (1) with k ¼ 81.25
tion for updating the guidance commands with the and reaches the desired location with respect to the
changing speed and heading direction of the leader. leader at the desired time t ¼ 60 s. Figure 9(b) shows
This could be a future work direction using the pre- the variation of follower heading which smoothly
sent work as a basis. converges to the desired leader heading as the for-
mation is achieved. The follower x speed reduces
with reducing position error ðxd  xÞ as shown in
Simulation studies Figure 9(c). The overall follower speed variation is
Simulations are carried out using Algorithm 1 and plotted in Figure 9(d). Both the heading and the
ideal dynamics presented in section titled ‘Proposed speed of the follower converge to those of the
Guidance method’. In addition, realistic simulations leader, simultaneously while reaching the desired
are carried out on kinematic guidance model bor- location. Follower lateral and longitudinal acceler-
rowed from Beard and McLain.31 which incorporates ations are plotted in Figure 10(a) and (b), respect-
kinematic autopilot dynamics and wind disturbances. ively. It can be seen that the accelerations smoothly
The leader is set to be moving with vl ¼ 12 m/s, reduce to zero as the follower approaches the

l ¼ 90 , xl0 ¼ 30 m, yl0 ¼ 100 m and the desired for- desired location.
mation corresponds to fc ¼ 25 m and lc ¼ 25 m. For a Further simulations are carried out to generate for-
minimum follower radius of turn of 25 m, kminC is mations with flexibility in time-to-formation. Three
computed as kminC ¼ 9:62. Accordingly, the minimum engagement times, that is, tf ¼ 60 s, 100 s and 150 s
possible engagement time is computed using kmin ¼ are considered. Corresponding spatial guidance con-
57:3 in equation (33) as tfmin ¼ 42:2 s. The speed stants are computed using equation (32) as
advantage factor is taken as r ¼ 10. The velocity guid- k ¼ 81:25, 135:05, and 202.25, respectively and result-
ance constant, kv ¼ 1 is selected satisfying equation ing trajectories are plotted in Figure 11(a). A lower
(21). The actual engagement time is computed as the desired time-to-formation, as governed by equation
time taken to reach a closing distance less than 0.01 m (32), leads to a lower value of k. This, using equation
from the desired relative location. (1) results in a steeper approach to the desired loca-
tion. Accordingly, paths corresponding to smaller
Case 1: Ideal dynamics and different values of k use higher curvatures for a quick conver-
gence to the desired location as shown in Figure 11(b).
engagement times
The error in time-to-formation is less than 104 s in all
This case considers the ideal engagement dynamics gov- the three engagements. Figure 11(c) and (d) shows
erned by equations (22) and (23). As a sample case, a that the follower speed and heading converge
desired engagement time, tf ¼ 60 s, (tf 4 tfmin ¼ 42:2 s) smoothly to the final desired values for all the three
is considered. The spatial guidance constant k is desired engagement times.
10 Proc IMechE Part G: J Aerospace Engineering 0(0)

(a) (b)

(c) (d)

Figure 9. Results for Case 1: Ideal dynamics (tf ¼ 60 s). (a) Leader and follower trajectories; (b) Follower heading variation;
(c) Variation of follower speed in x direction with ðxd  xÞ; (d) Follower speed variation.

(a) (b)

Figure 10. Results for Case 1: Acceleration profiles (tf ¼ 60 s). (a) Follower lateral acceleration; (b) Follower longitudinal
acceleration.
Suresh and Ratnoo 11

(a) (b)

(c) (d)

Figure 11. Results for Case 1: Ideal dynamics (tf ¼ 60, 100 and 150 s). (a) Leader and follower trajectories; (b) Curvature variation;
(c) Follower speed variation; (d) Follower heading variation.

Case 2: Realistic validation with vehicle dynamics

where wfd and vfad denote the desired follower heading
This subsection considers realistic validation of the and airspeed, respectively while wf and vfa are the
proposed guidance method on a kinematic autopilot actual follower heading and air speed, respectively.
model considering vehicle dynamics and effect of Constants k , k _ and kva are the proportionality con-
wind. The presence of wind causes the ground speed stants and _ fd and _ f denote the desired and actual
and course angle of the follower to deviate from the follower heading rates. The wind vector in the body
airspeed, vfa and heading, wf. The desired airspeed reference frame is expressed as
and heading profiles for formation generation can 2 3
be obtained from the desired ground speed and head- wn
ing profiles after incorporating the effect of wind. The vw ¼ 4 we 5 ð43Þ
model is borrowed from Beard and McLain.31 and 0
described subsequently. A planar kinematic model
of the follower motion is considered as Desired heading of the follower can be obtained
from the desired course angle vfd as given as
x_ ¼ vfa cos f þ wn ð39Þ
 
1 wn sin fd þ we cos fd
y_ ¼ vfa sin f þ we ð40Þ fd ¼ fd  sin ð44Þ
vfad
€f ¼ k ð fd  fÞ þ k _ ð _ fd  _ f Þ ð41Þ
where vfd corresponds to the spatial path given by
v_fa ¼ kva ðvfad  vfa Þ ð42Þ equation (1) and is obtained from equation (4).
12 Proc IMechE Part G: J Aerospace Engineering 0(0)

Rearranging and differentiating equation (44) gives speed and heading directions are controlled to approach
the desired rate of change of follower heading as their respective commanded values. However, since the
spatial path is fixed, virtual target-based UAV path fol-
_ fd vfad cosðfd  fd Þ þ wn cos fd  we sin fd lowing method is considered for implementation. The
_ fd ¼
vfad cosðfd  fd Þ virtual target moves along the spatial path governed by
wn sin fd  we cos fd equation (1) at a fixed lead distance of R ¼ 30 m and

vfad cosðfd  fd Þ the UAV uses the path following logic presented in
ð45Þ Thakar and Ratnoo.18 Simulation is carried out for a
desired engagement time of 100 s. The corresponding
Figure 12 shows the heading hold autopilot model spatial guidance constant is computed using equation
block diagram. From equation (19), the desired (32) as k ¼ 135. The other engagement parameters are
ground speed of the follower along the spatial trajec- the same as considered in Case 1. The follower has an
tory can be obtained as arbitrary initial heading of 20 with an initial speed,
 xd x
 10 m/s. The engagement scenario is simulated in the pres-
vl þ vrl 1  e kv ence of wind, modelled arbitrarily as
vfgd ¼ ð46Þ 2 3
sin fd 1
vw ¼ 4 2 5m=s ð49Þ
Using the desired ground speed, the corresponding 0
airspeed can be deduced as
2 3 2 3 2 3 Figure 14(a) shows the results wherein the follower
vfad cos fd vfgd cos fd wn smoothly converges to the desired path and thereafter
6 7 6 7 6 7 remains on it and achieves formation. The follower
4 vfad sin fd 5 ¼ 4 vfgd sin fd 5  4 we 5 ð47Þ
airspeed and ground speed and the respective desired
0 0 0
values are shown in Figure 14(b) and (c), respectively.
Follower course angle converge to the desired value as
which upon simplification leads to shown in Figure 14(d). In this simulation case, the
formation is achieved at tf ¼ 99:8 s. Error in achieving
vfad ¼ a predefined engagement time is considered as
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
v2fgd þ w2n þ w2e  2vfgd wn cos fd  2vfgd we sin fd tfactual  tf
et ¼  100 ð50Þ
ð48Þ tf

Figure 13 shows the airspeed hold autopilot model where tfactual is the actual time of the engagement.
block diagram. With closed-loop autopilot dynamics, the Figure 15 shows the variation of et for various k,

Figure 12. Autopilot block diagram for heading hold.

Figure 13. Autopilot block diagram for airspeed hold.

Suresh and Ratnoo 13

(a) (b)

(c) (d)

Figure 14. Results for Case 2: Realistic validation with vehicle dynamics. (a) Leader and follower trajectories; (b) Follower airspeed;
(c) Follower ground speed; (d) Follower course angle.

Case 3: Formation generation with

obstacle avoidance
While reaching the formation, the follower may
have to avoid any obstacle or airspace restrictions.
This subsection analyses and incorporates preliminary
obstacle avoidance characteristics into the proposed
guidance design. Consider the scenario shown in
Figure 16(a). The obstacles are considered as circular
regions. The dashed line marks the follower initial
heading direction corresponding to kmin . Any obstacle
to the left of this line (for example, the light grey circle)
is not a hindrance. Accordingly, any obstacle inside the
region bounded by the track line, the dashed line and
the x axis needs to be considered for avoidance. The
dark grey circle represents one such obstacle. The
Figure 15. Results for Case 2: Error in engagement time.
centre of the obstacle is denoted as (xo, yo) and
radius as ro. Figure 16(b) represents the limiting fol-
ðk557:3Þ. It can be seen that error remains well within lower paths avoiding the obstacles. Conservative obs-
a negligible range. tacle avoidance conditions can be expressed as,
This study highlights the ease of application of the
yðx ¼ xo  ro Þ5yo þ ro or yðx ¼ xo þ ro Þ4yo  ro
proposed guidance method on realistic vehicle model
and environmental conditions. ð51Þ
14 Proc IMechE Part G: J Aerospace Engineering 0(0)

(a) (b) Y

((x0 − r0), (y0 + r0))

r0
(x0, y0)

((x0 + r0), (y0 − r0))

yfinal 1 − e(−x/kl )

yfinal 1 − e(−x/ku)

(0, 0) X

(c)

Figure 16. Results for Case 3: Formation generation with obstacle avoidance. (a) Obstacle avoidance schematic; (b) Limiting UAV
paths avoiding the bounding square; (c) Simulation result.

These conditions guarantee that no follower path A sample simulation is carried out considering an
passes through the square bounding the obstacle. obstacle with a radius of 10 m having centre at (200 m,
Applying the constraints of equation (51) in equation 55 m). Other engagement parameters are the same as
(1) leads to those considered in the realistic scenario in Case 2
8 with kinematic autopilot model and the presence of
>
> 4kl ¼ ðxo ro ÞÞ wind. The limiting safe values of k are obtained using
>
< ln 1 yyo þro
final equation (52) as kl ¼ 94:3 and ku ¼ 229:2, respectively.
k ð52Þ
> ðx þr o ÞÞ The paths corresponding to kl and ku are as shown in
> 5ku ¼  yo ro 
> o
: ln 1 Figure 16(c) which avoid the obstacle. Using equations
yfinal
(53) and (54), the achievable time-to-formation can be
as the feasible choice of guidance constant k avoiding obtained as tf 2 ½42:2s, 69:7s [ ½170s, 1Þ.
the obstacle. Accordingly, the achievable time-to-for- Further multi-obstacle simulations are carried out
mation is now given as tf 2 ½tfmin , t1  [ ½t2 , 1Þ, where considering three obstacles, with centre at (125 m,
t1 and t2 are obtained by using kl and ku, respectively 85 m), (200 m, 55 m) and (400 m, 35 m) and radii 8 m,
in equation (32) as 10 m and 15 m, respectively as shown in Figure 17(a).
h  i The limiting safe values of k are obtained using equa-
kl ln y0:01 þ fc  xl0 tion (52) for each of the obstacles, individually.
final
t1 ¼ ð53Þ Other engagement parameters are the same as those
vl considered in the realistic scenario in Case 2 with kine-
matic autopilot model and the presence of wind.
and Since, the obstacle, O1 is located above the track
h  i line, it is not a hindrance for the UAV. The limiting
ku ln y0:01 þ fc  xl0 safe values of k to avoid obstacle O2, as obtained using
final
t2 ¼ ð54Þ equation (52) are kl ¼ 94:3 and ku ¼ 229:2, respect-
vl ively. The corresponding achievable time-to-formation
Suresh and Ratnoo 15

(a) (b)

Figure 17. Results for Case 3: Multiple obstacles. (a) Multiple obstacles scenario; (b) Simulation result.

is tf2 2 ½42:2 s, 69:7 s [ ½170s, 1Þ. Similarly, for obs- Using equation (57) in equation (56) leads to
tacle O3, kl ¼ 350:4 and ku ¼ 1338 give the limiting
feasible paths and corresponding achievable time-to-  
0:01
formation is tf3 2 ½42:2s, 260:2s [ ½994:5s, 1Þ. 0:01t2f þ 2vl0 tf þ 2k ln  2fc þ 2xl0 ¼ 0
yfinal
Combining the feasible ranges of k for the two
obstacles that need to be avoided, the achievable ð58Þ
time-to-formation is
tf 2 ½42:2 s, 69:7 s [ ½170 s, 260:2 s [ ½994:5 s, 1Þ. Solving equation (58) for k
Two desired engagement times, tf ¼ 50 s and 200 s
are chosen from the achievable set. The corresponding fc  xl0  0:005t2f  vl0 tf
k¼   ð59Þ
values of k are obtained using equation (32) as
ln y0:01
k ¼ 67.8 and k ¼ 269.5, respectively. Resulting UAV final

trajectories are shown in Figure 17(b). The UAV is

able to attain formation at the desired times while A desired engagement time of 120 s is considered.
safely avoiding the obstacles present. Other simulation parameters are the same as those
considered in the realistic scenario in Case 2 with
kinematic autopilot model and the presence of wind.
Case 4: Varying leader speed in realistic scenario
The guidance constant corresponding to the desired
This case studies the effectiveness of the proposed engagement time is obtained from equation (59) as
guidance method in a sample scenario considering k ¼ 170. The actual engagement time from the simu-
time varying leader speed. Simulation considers the lation is found to be 121.165 s which shows a close
leader speed as match with the theoretical value.
Figure 18(a) shows the corresponding leader
vl ¼ vl0 þ 0:01t ð55Þ and follower trajectories. The variations of leader
speed and follower speed are shown in Figure 18(b).
where the initial leader speed is taken as, vl0 ¼ 12 m/s. The plot shows that the follower speed converges to
Here, the engagement time can be obtained by using the varying leader speed as the vehicle approaches the
the average leader speed vlavg in equation (32) as desired location. Figure 19 shows the variation of per-
centage error in computed engagement time over a
  range of guidance constants from kmin to 1000.
0:01
k ln yfinal þ fc  xl0 It can be seen that the error is bounded within 3.5%
tf ¼ ð56Þ
vlavg of the desired value. The slight error is due to the
realistic vehicle dynamics and the presence of wind
considered.
where
Case 5: Comparative studies
Z tf
1 0:01tf A comparative study with existing methods that
vlavg ¼ ðvl0 þ 0:01tÞdt ¼ vl0 þ ð57Þ
tf 0 2 address the problem of UAV formation generation
16 Proc IMechE Part G: J Aerospace Engineering 0(0)

(a) (b)

Figure 18. Results for Case 4: Time varying leader speed. (a) Leader and follower trajectories; (b) Leader and follower speeds.

Figure 19. Results for Case 4: Error in engagement time.

is carried out. Table 1 presents the detailed qualitative Further to the qualitative comparison, a compara-
comparison study. Details from the Table highlights tive simulation studies with Lee et al.23 is carried out.
that in addition to vehicle constraints, the proposed Results are plotted in Figure 20 wherein all the initial
method also considers a direct control over engage- conditions are same as considered in Case 1 except the
ment time and an obstacle avoidance feature derived presence of a circular obstacle with centre and radius
as a closed-form function of the guidance parameter. being (200 m, 50 m) and 20 m, respectively. Using
It is noteworthy that Ryoo et al.26 also considers a equation (52), the limiting values of k are obtained
given desired engagement time as an input to the opti- as kl ¼ 66:5 and ku ¼ 430:67 which are considered
mization problem. However, in the absence of a for generating the follower paths. These two specific
closed-form relationship, it is not possible to deter- values of k, using equations (53) and (54), corres-
mine the feasible limits of engagement time in that ponds to engagement times 49 s and 319.8 s, respect-
approach. ively. For the method presented in Lee et al.,23 k ¼ 175
Suresh and Ratnoo 17

Table 1. Qualitative comparison with existing works.

Vehicle turn-rate
Guidance constraint and Closed-form expression
method Number of design parameters velocity constraints Obstacle avoidance for engagement time
23
2: Spatial guidance gain, k and heading Considered Not considered No
rate proportional control gain kp
25
2: Controller gains, k1 and k2 Not considered Not considered No. Simultaneous arrival
of UAVs can be
achieved, but the time
of arrival is not defined.
26
2: Guidance gain, N and Not considered Not considered No. Terminal time is taken
terminal time, tf as a design parameter
for obtaining the opti-
mal guidance law; how-
ever, there is no
closed-form
expression.
29
8 polynomial coefficients per vehicle Considered Not considered No
for the feasible path
Proposed 3: Spatial guidance constant, Considered Considered. Set of feas- Yes. Engagement time is
method k, velocity guidance ible guidance constants expressed in terms of
constant, kv and speed with closed-form limits the initial conditions
advantage factor, r is deduced considering and a single design par-
local axis aligned rect- ameter, k.
angular obstacles.

Figure 20. Result for Case 5: Comparison with Lee et al.23

is considered for generating the trajectories. It can be

Conclusions
seen that for the considered choice of k in Lee et al.,23
the follower trajectory passes through the obstacle Combining a class of exponential spatial paths and a
region. There can be choices of k which correspond desired speed profile, a new guidance method is pre-
to avoiding the obstacle. However, Lee et al.23 does sented for generating leader–follower formation with
not provide any results or guarantees for satisfying a engagement time constraint. Lyapunov stability analysis
given engagement time or avoiding obstacles. guarantees convergence of the engagement states to the
18 Proc IMechE Part G: J Aerospace Engineering 0(0)

formation equilibrium. The time-to-formation is pre- 7. Kokume M and Uchiyama K. Guidance law based on
sented as a closed-form expression of a spatial guidance bifurcating velocity field for formation flight. In: AIAA
constant and the leader initial conditions. Simulation Guidance, Navigation, and Control Conference, August
studies demonstrate the ability of the proposed guidance 2010, Toronto, Canada, AIAA paper 2010-8081.
8. Suzuki M and Uchiyama K. Autonomous formation
method in achieving time constrained formation with
flight using bifurcating potential fields. In: 27th
ideal as well as realistic vehicle dynamics in the presence
Congress of the international council of the aeronautical
of wind. Apart from the theoretical guarantee of achiev- science, 2010, pp.1–8.
ing the formation, the proposed guidance method pre- 9. Tahk MJ, Park CS and Ryoo CK. Line-of-sight guid-
sents greater flexibility in terms of easily incorporating ance laws for formation flight. J Guid Control Dyn 2005;
realistic constraints like the UAV turn rate, obstacle 28: 708–716.
avoidance and operating speed limits. The application 10. Betser A, Vela PA, Pryor G, et al. Flying in formation
of the method is limited to non-manoeuvring leader using a pursuit guidance algorithm. In: Proceedings of
moving on a straight line path and the obstacle avoiding the American control conference, June 2005, Portland,
feature is conservative in nature. Considering multiple USA, pp.5085–5090. Piscataway: IEEE.
variable curvature track paths, potential future works 11. Nelson DR, Barber DB, McLain TW, et al. Vector field
path following for miniature air vehicles. IEEE Trans
include extension of the work to a general multiple
Robotics Autom 2007; 23: 519–529.
vehicle scenario with a manoeuvring leader.
12. Pothen AA and Ratnoo A. Curvature-constrained lya-
punov vector field for standoff target tracking. J Guid
Declaration of Conflicting Interests Control Dyn 2017; 40: 2729–2736.
The author(s) declared no potential conflicts of interest with 13. Lawrence DA, Frew EW and Pisano WJ. Lyapunov
respect to the research, authorship, and/or publication of vector fields for autonomous unmanned aircraft flight
this article. control. J Guid Control Dyn 2008; 31: 1220–1229.
14. Reddy CV and Ratnoo A. Vector field based formation
Funding generation guidance for arbitrary initial conditions.
IFAC-PapersOnline 2014; 47: 443–447.
The author(s) disclosed receipt of the following financial
15. B Medagoda ED and Gibbens PW. Synthetic-waypoint
support for the research, authorship, and/or publication
guidance algorithm for following a desired flight trajec-
of this article: This effort is supported by IMPRINT India
tory. J Guid Control Dyn 2010; 33: 601–606.
initiative via Project Grant 6921.
16. Park S, Deyst J and How JP. Performance and
lyapunov stability of a nonlinear path following
ORCID iD guidance method. J Guid Control Dyn 2007; 30:
Shalini Suresh https://orcid.org/0000-0002-7690-2655 1718–1728.
17. Cho N, Kim Y and Park S. Three-dimensional non-
References linear differential geometric path-following guidance
1. Giulietti F, Innocenti M and Pollini L. Formation law. J Guid Control Dyn 2015; 38: 2366–2385.
flight control-a behavioral approach. In: AIAA guid- 18. Thakar S and Ratnoo A. A tangential guidance logic for
ance, navigation, and control conference and exhibit, virtual target based path following. In: AIAA Guidance,
August 2001, Montreal, Canada, AIAA paper Navigation, and Control Conference, January 2017,
2001–4239. Grapevine, USA, AIAA paper 20177–1042.
2. Gaudiano P, Bonabeau E and Shargel B. Evolving 19. Zhang J, Li Q, Cheng N, et al. Path-following con-
behaviors for a swarm of unmanned air vehicles. In: trol for fixed-wing unmanned aerial vehicles based on
Proceedings of IEEE swarm intelligence symposium, SIS a virtual target. Proc IMechE, Part G: J Aerospace
2005, June 2005, Pasadena, USA, pp.317–324. Engineering 2014; 228: 66–76.
Piscataway: IEEE. 20. Ambrosino G, Ariola M, Ciniglio U, et al. Algorithms
3. Li NH and Liu HH. Formation UAV flight control using for 3d UAV path generation and tracking. In:
virtual structure and motion synchronization. In: Proceedings of IEEE Conference on Decision and
Proceedings of the American control conference, June Control, December 2006, San Diego, USA,
2008, Seattle, USA, pp.1782–1787. Piscataway: IEEE. pp.5275–5280. Piscataway: IEEE.
4. Low CB and San Ng Q. A flexible virtual structure for- 21. Park S, Deyst J and How J. A new nonlinear guidance
mation keeping control for fixed-wing UAVs. In: logic for trajectory tracking. In: AIAA Guidance,
Proceedings of international conference on control and Navigation, and Control Conference and Exhibit,,
automation (ICCA), December 2011, Santiago, Chile, August 2004, Providence, Rhode Island, USA, AIAA
pp.621–626. Piscataway: IEEE. paper 2004–4900.
5. Paul T, Krogstad TR and Gravdahl JT. Modelling of 22. Ratnoo A, Hayoun SY, Granot A, et al. Path following
UAV formation flight using 3D potential field. Simul using trajectory shaping guidance. In: AIAA Guidance,
Model Pract Theory 2008; 16: 1453–1462. Navigation, and Control Conference and Exhibit, August
6. Han K, Lee J and Kim Y. Unmanned aerial vehicle 2013, Boston, USA, AIAA paper 2013–5233.
swarm control using potential functions and sliding 23. Lee D, Kim S and Suk J. Formation flight of unmanned
mode control. Proc IMechE, Part G: J Aerospace aerial vehicles using track guidance. Aerosp Sci Technol
Engineering 2008; 222: 721–730. 2018; 76: 412–420.
Suresh and Ratnoo 19

24. Lee D, Kim SK and Suk J. Design of a track guid- Appendix

ance algorithm for formation flight of UAVs.
In: AIAA Guidance, Navigation, and Control Notation
Conference, January 2015, Kissimmee, USA, AIAA
C curvature of the path
paper 2015–1315.
25. Koo TJ and Shahruz SM. Formation of a group of
Cmax maximum curvature of the path
unmanned aerial vehicles (UAVs). In: Proceedings of fc desired forward clearance between the
the American control conference, June 2001, Arlington, follower and the leader
USA, pp.69–74. Piscataway: IEEE. k spatial guidance constant
26. Ryoo CK, Kim YH and Tahk MJ. Optimal UAV for- kav autopilot gain for velocity hold
mation guidance laws with timing constraint. Int J Syst kmin minimum spatial guidance constant
Sci 2006; 37: 415–427. kv velocity guidance constant
27. Kaminer I, Yakimenko O, Pascoal A, et al. Path gener- k , k_ autopilot gains for course hold
ation, path following and coordinated control for time lc desired lateral clearance between the
critical missions of multiple UAVs. In: Proceedings of the follower and the leader
American Control Conference, Minneapolis, USA, June
tf time-to-formation
2006, pp. 4906–4913. Piscataway: IEEE.
28. Kaminer I, Yakimenko O, Dobrokhodov V, et al.
tfmin minimum time-to-formation
Coordinated path following for time-critical vf, vl follower and leader velocities,
missions of multiple UAVs via l1 adaptive output respectively
feedback controllers. In: AIAA Guidance, Navigation, vfa, vfad actual and desired follower airspeeds,
and Control Conference and Exhibit, August 2007, respectively
Hilton Head, South Carolina, USA, AIAA paper vfg, vfgd actual and desired follower ground
2007-6409. speeds, respectively
29. Xargay E, Dobrokhodov V, Kaminer I, et al. Time- vw wind velocity
critical cooperative control of multiple autonomous wn, we components of wind velocity along
vehicles: Robust distributed strategies for path-following north and east directions, respectively
control and time coordination over dynamic communica-
ðxl0 , yl0 Þ leader initial position
tions networks. IEEE Control Syst Magazine 2012; 32:
49–73.
ðx, yÞ instantaneous follower position
30. Suresh S and Ratnoo A. UAV formation generation xd desired x displacement of the follower
guidance with engagement time constraints. In: 2018 yfinal desired y displacement of the follower
AIAA Information Systems-AIAA Infotech@ Aerospace, vfd, wfd desired ground velocity and air velocity
January 2018, Kissimmee, Florida, USA, AIAA paper headings of the follower
2018-2144. wf, wl follower and leader headings,
31. Beard RW and McLain TW. Small unmanned aircraft: respectively
theory and practice. Princeton: Princeton Univ. Press, f0 initial follower heading
2012, pp.21–27, pp.175–177.