Int. J. Automation and Control, Vol. 18, No.

5, 2024 489

Bio-inspired evasion strategies under variable evader

speed

Lairenjam Obiroy Singh*

Manipur International University,
Imphal, Manipur, 795140, India
Email: [email protected]
*Corresponding author

Devanathan Rajagopalan
Hindustan Institute of Science and Technology,
Chennai, 603103, India
Email: [email protected]

Abstract: The pursuit evasion game (PEG) is a natural phenomenon with

implications for civilian and military applications. Researchers have developed
methods to analyse PEG dynamics under various bio-inspired strategies,
providing valuable insights into possible applications. Studies assume constant
speeds for both the pursuer and evader, but this can result in less agile pursuit.
This paper explores the trade-off between higher speed and less agility when
the evader increases speed as the pursuer approaches. By simulating PEG
trajectories, nine combinations of bio-inspired pursuer-evader strategies were
tested using computer simulations. Results showed that a higher average speed
may delay the evader’s capture or even result in their escape in some instances.
However, computer simulations indicate a mixed outcome for evader escape
performance in cases of sudden turns and non-reactive evader strategies. The
paper’s results contrast existing results that rely solely on numerical
computation of PEG outcomes under varying evader speeds.

Keywords: bio-inspired; closed-loop control; differential game theory;

feedback laws; pursuit evasion game; PEG.

Reference to this paper should be made as follows: Singh, L.O. and

Rajagopalan, D. (2024) ‘Bio-inspired evasion strategies under variable evader
speed’, Int. J. Automation and Control, Vol. 18, No. 5, pp.489–515.

Biographical notes: Lairenjam Obiroy Singh received his PhD and MTech
from Hindustan Institute of Technology and Science, Chennai, India and BE
degree from Manipur Institute of Technology, Imphal, Manipur. Currently, he
is working as an Assistant Professor cum Research officer in Electronics and
Computer Science department at Manipur International University. He is also
an Assistant Controller of Examinations at MIU. He is a member of IEEE and
IET. His research interest includes pursuit evasion game (PEG), differential
pursuit evasion game theory, bio-inspired evolutionary game theory,
autonomous vehicle, robotics and system engineering, wireless sensor network
(WSN), and artificial intelligence.

Devanathan Rajagopalan received his PhD and MSc (Engineering) from

Queen’s University, Kingston, Ontario, Canada and his BE and ME degrees
from the Indian Institute of Science, Bangalore, India. He has taught at

Copyright © 2024 Inderscience Enterprises Ltd.

490 L.O. Singh and D. Rajagopalan

Nanyang Technological University (NTU), Singapore for over two decades. He

has published over 180 papers in international and national conference
proceedings and journals and has received awards from the IEEE Education
Society and NTU. He has chaired and co-chaired international conferences
organised by IEEE and NTU. He is a Professor Emeritus in Electrical and
Electronic Engineering at the Hindustan Institute of Technology and Science,
Chennai, India. His interests include control systems, nonlinear systems,
robotics, image processing, discrete event systems and quantitative linguistics.
He is a life senior member of IEEE.

1 Introduction

The Pursuit evasion game (PEG) is found all across nature. PEG can be a single-player or
a multi-player game. PEG is seen in nature, for example, when a lion chases a deer or a
predator chases a prey. PEG is also seen in a jungle as predatory animals fight over
territory. PEG can also be seen in animal mating behaviours. PEG applications are used
in various military and civilian scenarios. For instance, they are used in missile guidance
(Shneydor, 1998), aircraft pursuit and evasion (Svec et al., 2012), military harbour
defence (Lee et al., 2013), harbour asset protection (Simetti et al., 2010), and UAV
collision avoidance (Mueggler et al., 2014; Lin and Saripalli, 2017).
The PEG problem is a fascinating one that has been extensively researched by various
disciplines for many years (Issacs, 1965; Howland, 1974; Pachter and Yavin, 1981;
Weihs and Webb, 1984; Pachter, 1987; Hedenström and Rosén, 2001; Sgall, 2001; Caro,
2005; Kopparty and Ravishankar, 2005; Gerkey et al., 2006). The problem is based on
predator-prey dynamics, where it is commonly assumed that the pursuer moves faster
than the evader. However, a faster pursuer has limited manoeuvrability due to a longer
turning radius at higher speeds, whereas the evader can manoeuvre more easily at lower
speeds. This makes the problem of capturing the evader by the pursuer a complex one.
Various analyses of PEG exist in terms of kinematics, motion geometry, and more.
However, these analyses do not provide a satisfying answer to how the pursuer should
move or what control should be applied in a given PEG situation. In particular, the
analysis of capture time in PEG is difficult and does not exist in complete form
(Glendinning, 2004).
Issacs (1965) contributed to developing pursuit-evasion games. He provided PEG
research based on game theory. John Von Neumann (Von Neumann and Morgenstern,
1945) developed a mathematical formulation of contemporary game theory. The
applicability of PEG can even be seen in economics. Each player or group used a distinct
strategy based on game theory in these cases. The strategies employed determine the
game’s outcome. Nature has given us countless lessons during the evolution of humanity.
Several strategies are developed that are inspired by nature. Although PEG theory has
been studied for a long time, research on bio-inspired PEG approaches is relatively young
compared to standard PEG theory. The quantity of studies on bio-inspired methods has
recently expanded dramatically.
In their paper (Justh and Krishnaprasad, 2006) propose a biologically inspired
problem and introduce a strategy called motion camouflage (CM). The proposed problem
is formulated using feedback control laws, and its performance is evaluated through
computer simulations. This study also briefly covers the relationship between the missile
 Bio-inspired evasion strategies under variable evader speed 491

defence system and the proposed strategy. Krishnaprasad and his colleagues (Wei et al.,
2009) investigated PEG using geometric control theory. The term ‘manifold’ in PEG
dynamics refers to an Euclidean curve corresponding to the pursuer-evader trajectory.
Once on the manifold, the baseline distance between the evader and pursuer is guaranteed
to reduce owing to the pursuer’s increased speed eventually leading to the evader’s
capture by the pursuer. The manifold is defined by criteria relating to the evader’s
relative distance and velocity. A feedback control lawallows the pursuer to stay on the
desired manifold. This study expands upon the previous work of Justh and
Krishnaprasad, 2006). The authors presented two other bioinspired strategies, viz.,
constant bearing (CB) and classical pursuit (CP), which are also prevalent in nature.
While CM is a stealth approach, CB (CP) refers to the pursuer forming an angle (being
aligned) to the baseline connecting the pursuer and evader. The authors model the issue
as a closed-loop feedback system. The performance of the intended strategies is evaluated
using computer simulation.
The work of Wei et al. (2009) has shown that CB and CP strategies always converge
to CM strategy under certain assumptions of evader motion, thus supporting what is
prevalently seen in nature. A study of evolutionary games carried out by Pais and
Leonard (2010) considers interacting particles moving at fixed speeds. Three strategies of
both the pursuer and the evader are studied in this study. Monte Carlo Simulation is used
to analyse the performance of different strategies. It is demonstrated that if the evader
follows a reactive (RE) (i.e., dependent on the pursuer and evader locations) control law,
the CB and CP strategies do not necessarily converge to CM always. The assumptions in
(Pais and Leonard, 2010) of fixed speed for the pursuer and evader are dropped in the
work of (Li, 2017), who introduced the case of the evader turning abruptly left or right
when the distance between pursuer and evader fell below a particular threshold value,
which is another strategy found in nature. When a particular threshold distance between
the pursuer and evader is reached, the work suggests either the evader will turn left or
right or if the evader reaches what is called an escape zone, then it will be regarded as a
successful escape from the pursuer.
The work (Li, 2016) is about confining of evader to a fixed predefined path by the
pursuer. In Obiroy Singh and Devanathan (2017) and Obiroy Singh and Devanathan
(2018), the authors show improvement in the performance of the CM strategies of the
pursuer in the PEG system by employing PID control action compared to primary control
action already optimised in Justh and Krishnaprasad (2006). This result shows the
performance of the PID controller in an essentially nonlinear dynamic setting as in PEG.
It is worth noting, in this context, that the function of PID control in advanced control
applications is well established (see, for example, Aboulem et al., 2019; Ardeshiri et al.,
2019; Pati and Negi, 2019; Ahamed et al., 2020; Tang et al., 2020; Yi et al., 2021; Goud
et al., 2022). The authors (Singh and Devanathan, 2019) demonstrate how the evader’s
performance in the PEG may be improved by employing the sudden turn (ST) strategy
and the surprise aspect connected with the technique. Wu et al. (2020) propose a guiding
approach to intercept a manoeuvring target in finite time in a study in the general domain
of target tracking. It is proposed to ensure the robustness and stability of guidance using a
Lyapunov function approach.
The motivation for the present study is based on the natural fact that the evader or
prey runs faster or slower depending on how far away the predator is from it. In other
words, it is more natural to think of the evader’s speed as a variable rather than being
fixed as is assumed in many existing works. While increasing speed may help the evader
492 L.O. Singh and D. Rajagopalan

delay capture or escape, it may also reduce their agility, thereby reducing the surprise
effect of STs on the pursuer. The question is whether such an effect of varying evader
speed can be supported through simulation or even through theoretical analysis.
The following are the primary contributions of the paper: In a two-dimensional
scenario, assuming a single pursuer and a single evader, the dynamics of PEG is
formulated. Evader is assumed to follow three different bio-inspired strategies with its
speed varying inversely as the baseline distance between pursuer and evader. The pursuer
is assumed to follow three well-known bio-inspired strategies under closed loop control
which seeks to maintain the pursuit on the respective pursuit manifold. The objective is to
investigate the capture time of the evader under nine different combinations of
pursuer-evader strategies, with varying evader average speeds and different pursuer
starting positions. Computer simulations of the pursuer-evader trajectories are studied
and analysed. Analytical results have also been derived for all nine scenarios, which
support the simulation results in principle. The findings of the paper contrast with the
current result (Li, 2017), in which efficient evader escape strategies are studied
numerically but no analytical solution is offered.
The rest of the paper is organised as follows: Section 2 provides a brief overview of
the system dynamics of pursuer and evader in the PEG system, as well as the pursuit
manifold and the cost function associated with the various pursuer strategies. Section 3
discusses system modelling and feedback control systems. It also has separate control
laws for the pursuer and the evader. Section 4 presents simulation results of PEG
trajectories under various combinations of pursuer and evader strategies together with an
analysis of the results. In Section 5, a theoretical analysis is presented to support the
empirical findings of Section 4. The conclusion is provided in Section 6.

2 Background

2.1 Dynamics of PEG

When gyroscopically interacting particles are provided, the baseline trajectory connecting
the evader and pursuer locations can be defined by a curve in Euclidean space
represented by the Frenet-Serret equation (Boothby, 1975), as shown below.
dT 
= kN , 
ds

dN 
= − kT + τB,  (1)
ds 
dB 
= − τN , 
ds 
d
where is the derivative with respect to arc length, k is the curvature (i.e., deviation
ds
from a straight line) and τ is torsion (twisting or deviation from a plane curve). T is the
unit vector tangent to the curve pointing in the direction of motion, N is the normal unit
 Bio-inspired evasion strategies under variable evader speed 493

vector and B is the binormal unit vector. Considering only a planar motion, we let the
ds
τ(s) = 0 in (1), and considering the unit speed along the arc length = 1, (1) becomes
dt
dT 
= kN 
dt 
 (2)
dN
= − kT 
dt 

Denoting r ∈ ℝ2 to be the position of the particle and the tangent T at r to be x and

identifying k as u and the normal N as y, we can then write
dr 
=x 
dt

dx 
= uy  (3)
dt 
dy 
= −ux 
dt 
Thus we arrive at the notation of Wei and Krishnaprasad (Justh and Krishnaprasad, 2006)
and Justh and Krishnaprasad (2005), where PEG is described by a planar motion given by
rp = x p , 

x p = y p u p ,  (4)
y p = − x p u p , 

where rp ∈ ℝ2 is the position of the pursuer, xp and yp are the velocity and acceleration of
the pursuer, respectively. up is the scalar steering control of the pursuer. ( x ) denotes the
ds
derivative of x with respect to time t. For the case of an evader, we can put = ν < 1,
dt
where v is the evader’s speed. The dynamics of evader planar motion are then given by
re = vxe , 

xe = vye ue ,  (5)
y e = −vxe ue , 

where re ∈ ℝ2 is the position of the evader, xe and ye are the velocity and acceleration of
the evader, respectively. ue is the scalar steering control of the evader. We define the
‘baseline’ between the pursuer and evader as
r = rp − re (6)

2.2 Pursuit manifolds and cost functions

Following (Justh and Krishnaprasad, 2006), we define the cost functions F on the pursuit
manifold G ∈ ℝ2 as F: G × G → ℝ with different pursuit strategies as in Issacs (1965)
494 L.O. Singh and D. Rajagopalan

where (x • y) denotes the dot product between two vectors and |x| stands for the Euclidean
norm. The cost functions are defined as follows.
d |r|
 r r 
Γ= ⋅  = dt ( Motion camouflage) (7)
 | r | r  dr
dt

 r 
Λ= ⋅ Rx p  (Constant bearing ) (8)
| r | 

cosθ − sinθ   π π
where R =   for θ ∈  − ,  . For R = I2 the identity matrix of order 2, we
 sinθ cosθ   2 2
define
 r 
Λ0 =  ⋅ x p  (Classical pursuit ) (9)
| r | 
to be the cost function associated with CP. The cost functions Γ, Λ and Λ0 define the
respective pursuit manifolds, and they take the values in the interval [–1, 1]. Figure 1
provides an intuitive view of the effect of different pursuer strategies studied on the
geometric orientation of the pursuer and the evader. The CM manifold is defined by the
condition, which corresponds to the case of the angle between r and r being π [see
Figure 1(a)]. The CB manifold is represented by the condition Λ = –1 when the heading
of the pursuer makes an angle θ with the baseline vector, as shown in Figure 1(b).
Similarly, the CP manifold is defined by the condition Λ0 = –1 where the heading of the
pursuer is aligned (in the opposite direction) with the baseline [see Figure 1(c)].

3 System modelling

3.1 Pursuit-evasion system

Representing
t t
rp =  rpx , rpy  ; re =  rex , rey 

Equations (4) and (5) can be written in terms of state equations as follows:
x1 = rpx ; x7 = rex ; 
x2 = rpy ; x8 = rey ; 
x3 = x px ; x9 = xex ; 
 (10)
x4 = x py ; x10 = xey ; 
x5 = y px ; x11 = yex ; 

x6 = y py ; x12 = yey ;
 Bio-inspired evasion strategies under variable evader speed 495

x1 = rpx = x px = x3 
x2 = rpy = x py = x4 

x3 = x px = y px u p = x5u p 

x4 = x py = y py u p = x6u p 
x5 = y px = − x px u p = − x3u p 

x6 = y py = − x py u p = − x4 u p 
 (11)
x7 = rex = νxex = νx9 
x8 = rey = νxey = νx10 

x9 = xex = νyex ue = νx11ue 
x10 = xey = νyey ue = νx12 ue 

x11 = y ex = −νxex ue = −νx9 ue 
x12 = y ey = −νxey ue = −νx10 ue 

 x  x   x − x 
r = rp − re =  1  −  7  =  1 7  (12)
 x2   x8   x2 − x8 

 x   x   x  x 
r = rp − re =  1  −  7  =  3  − ν  9  (13)

x x
 2  8  4 x  x10 

3.2 Feedback laws

Pursuing strategies in terms of feedback control laws are formulated in this sub-section.
The respective cost function Γ, Λ and Λ0 associated with different pursuer strategies are
maintained at the reference value of –1 on the individual manifold. The expressions for
the control laws of different pursuit and evasion strategies and their expressions in state
variables are given in propositions 1 and 2, whose proofs are available in an earlier
publication of the authors (Singh and Devanathan, 2019).

Figure 1 Geometric representation of pursuit manifolds, (a) motion camouflage (CM) pursuit
(b) CB pursuit (c) CP (see online version for colours)

(a) (b) (c)

3.2.1 Control laws for the pursuer

Proposition 1
496 L.O. Singh and D. Rajagopalan

The control expressions for different pursuer strategies are given below. μ and η are the
proportionality coefficients. The derivation of (1 to 3) is done by (Obiroy Singh and
Devanathan, 2018).
1 For motion camouflage,

 ( x1 − x7 , x2 − x8 ) 
up = μ   ⋅ ( νx10 − x4 ) , ( x3 − νx9 )  (14)
( x1 − x7 ) + ( x2 − x8 ) 
2 2


2 For CB,

( x1 − x7 , x2 − x8 )  cosθ − sinθ 
 ( x5 , x6 )
t
up = η ⋅
 sinθ cosθ 
( x1 − x7 ) + ( x2 − x8 )
2 2

(15)
1 ( x1 − x7 , x2 − x8 )
+ ⋅ ( νx10 − x4 ) , ( x3 − νx9 )
( x1 − x7 ) + ( x2 − x8 ) ( x1 − x7 ) + ( x2 − x8 )
2 2 2 2

3 For CP, the expression for control law up is the same as that of (15), with the
substitution of R = I2.

3.2.2 Control laws for the evader

Proposition 2
The control expressions for evader for different strategies assumed are given below. The
derivation of (1 to 3) is done by (Obiroy Singh and Devanathan, 2018)
1 For non-reactive (NR) case,
ue = cost (16)

In this case, the evader’s motion is assumed to follow a cyclical path irrespective of
the pursuer’s strategy.
2 For RE,

 ( x1 − x7 , x2 − x8 ) 
ue =    ( −νx10 , νx9 )  (17)
( x1 − x7 ) + ( x2 − x8 ) 
2 2


An evader’s RE strategy is based on information about the pursuer’s location.

3 For ST,

 ( x1 − x7 , x2 − x8 ) 
ue =   [ νx9 , νx10 ] (18)
( x1 − x7 ) + ( x2 − x8 ) 
2 2


In this case, the evader is assumed to suddenly turn 90 degrees left or right to the baseline
vector to avoid the evader.
 Bio-inspired evasion strategies under variable evader speed 497

3.3 Variable speed of evader

There are no existing results on the modelling of variation of evader speed. Without such
results, it is thought best to let the evader’s speed vary inversely to the pursuer’s distance.
Evader’s variable speed is represented by vvariable. The magnitude of the distance between
the pursuer and the evader determines the variable speed value. The closer the pursuer
gets to the evader, the greater the value of vvariable. Formula (19) is a direct and intuitive
way to implement that requirement.
The expressions for variable speed and its expressions in state variable form are
assumed below.
Ω Ω
νvariable = + (1 − Ω)  νvariable = + (1 − Ω ) (19)
| r | +1 ( x1 − x7 )
2
+ ( x2 − x8 ) + 1
2

with 0.1 ≤ Ω ≤ 0.5. When |r| = 0, vvariable reaches its highest value of 1 for all Ω. For large
values of |r|, vvariable reaches in the limit (1 – Ω). Ω is a tuneable parameter.

4 Simulation

In this section, we present and analyse the comparative performance of the evader in
constant and variable speed cases. Different combinations of the pursuer (CM, CB and
CP) and evader (RE, NR and ST) strategies will be studied. We employ the system
dynamic equations from Section 3 for PEG simulation. We utilise (19) to modify the
variable speed vvariable. To test the performance of the PEG system, we assume a typical
time range of 0 to 100 seconds. Figures 2 to 4 aim to compare the effect of variable
evader speed vvariable against the lower constant evader speed.

4.1 Baseline threshold ρ

The time the pursuer requires to catch the evader for the first time is the performance
metric to be employed. Capture is considered to have occurred when the baseline distance
falls below a particular threshold, represented as ρ > 0. When dealing with asymptotic
responses, comparing the times to capture for the two instances of a fixed lower speed
and a greater average speed of the evader leads to the selection of a value for ρ. Except
for easing the comparison of the variable and constant speed instances in terms of capture
speed, the parameter ρ has no systemic significance. Initial experiments have revealed
that a value of 0.5 may be appropriate for ρ. However, the threshold ρ is decreased in
some circumstances to provide a more accurate comparison.

4.2 Comparative performance of evader – pursuer strategies

Figures 2 to 4 provide ‘time to capture’ data for the evader strategies RE, NR and ST
based on simulation runs. Subsections (a), (b) and (c) in Figures 2 to 4 depict distinct
strategies used by the pursuer: motion camouflage (CM), CB and constant pursuit (CP),
respectively. The x axis of Figures 2 to 4 represents the average speed Av of the evader
498 L.O. Singh and D. Rajagopalan

t
1
t 
where Av = νvariable . The y axis in Figures 2 to 4 corresponds to the time to capture of
0
evader (tr) by the pursuer for the first time. Here the term ‘capture’ of evader by pursuer
refers to the magnitude of |r| ≤ ρ > 0. The legends correspond to six different initial
starting coordinates for rp and xp. re and xe are assumed to be (0, 0)t and (1, 0)t
respectively. If the tr value is not plotted in a certain case, which means the distance
between the pursuer and the evader does not reach the threshold value within the
specified simulation time limit so as to be considered as the capture of the evader by the
pursuer. Under the CB pursuit strategy, we used θ = 3.2º. The reason for choosing this is
to make the performance data lay within 100 seconds so that a comprehensive
comparison can be made across the strategies and control.

4.2.1 Evader strategy RE

Figure 2(a) shows the graph for tr vs. Av for CM of the pursuer and RE of the evader. It is
seen that as the evader’s average speed Av increases, the time to capture the evader tr also
increases. Consider, for example, the initial starting position at (4, –6), where the time to
capture is found to be 5.2 seconds when the lower constant speed of the evader is 0.6.
Similarly, when the average speed of the evader is varied from 0.64 to 0.73 and then
further to 0.79, the time to capture the evader increases from 5.7 seconds to 6.2 seconds
and then to 6.7 seconds, respectively.
Figure 2(b) shows the effect of variable speed vvariable on time to capture tr for CB of
pursuer and RE of evader. In this case, the time to capture the evader tr generally
increases with the increase in the evader’s average speed Av. Let us consider, for
example, the initial starting position at (–1, 1). It is seen that as the average speed of the
evader increases, the time to capture the evader also increases. The time to capture the
evader increases from 2.5 to 2.6 to 28.9 seconds when the average speed varies from 0.6
to 0.63 and to 0.79, respectively. However, there are certain cases when the time to
capture the evader is not reflected with the change in the average speed. For instance, the
initial starting position at (10, –1) indicated by the green colour in the graph tr is not
reflected for Av values of 0.76, 0.69 and 0.62. The reason is due to the failure of the
pursuer to capture the evader within the time frame of 100 seconds, and thus the evader is
considered to have technically escaped from the pursuer.
Figure 2(c) corresponds to the effect of variable speed vvariable on time to capture tr for
CP of the pursuer and RE of the evader. Similar results, as in Figures 2(a) and 2(b) also
apply in this case. As the average speed Av of the evader increases, the time to capture the
evader also increases. For example, let us consider the initial starting position at (–2, 10).
It is observed that the time to capture the evader is 8.5 seconds, 11.5 seconds and
12.5 seconds when the average speed is 0.6, 0.89 and 0.92, respectively.

4.2.2 Evader strategy NR

Figure 3(a) shows the effect of variable speed vvariable on time to capture tr for the
pursuer’s CM and the evader’s NR. Notably, as the average speed Av increases, so does
the time required to capture evader tr. Consider the initial starting position at (10, –1) in
the figure, represented by green colour. When the average speed Av is 0.6, the time to
 Bio-inspired evasion strategies under variable evader speed 499

capture the evader tr is found to be 9.2 seconds while Av is 0.75 and 0.92, and the time to
capture the evader tr is 10.6 and 12.1 seconds, respectively.

Figure 2 Effect of variable speed vvariable on time to capture tr for (a) CM (b) CB (c) CP under RE
(see online version for colours)

(a)

(b)

(c)
500 L.O. Singh and D. Rajagopalan

Figure 3 Effect of variable speed vvariable on time to capture tr for (a) CM, (b) CB (c) CP under
NR (see online version for colours)

(a)

(b)

(c)

Figure 3(b) depicts the effect of variable speed vvariable on time to capture tr for the
pursuer’s CB and the evader’s NR strategies. In this scenario, the pursuer exhibits a
mixed result in that as Av increases, so does tr for a few initial starting positions.
However, in a few other cases, the relationship between Av and tr is not so related.
 Bio-inspired evasion strategies under variable evader speed 501

Considering the initial starting position of (–1, 1), it is observed that as Av is increased, so
does tr. When the average speed Av is increased from 0.6 to 0.8 and 0.93, the time to
capture the evader tr increases from 2.8 seconds to 21.2 seconds and then to 24.6 seconds.
However, as Av is increases from 0.6 to 0.79 and then to 0.92 for the initial starting
position at (–2, 10), the time to capture the evader tr decreases from 17.7 seconds to 17.3
seconds and then to 17.1 seconds. This is probably because in the CB-NR case, NR,
being of a cyclic motion, does not give the advantage of increased baseline distance with
increased speed (as is uniformly available with RE of an evader). Further, the pursuer
strategy of CB not being as predictable as CM results in a mixed outcome.
Figure 3(c) corresponds to the study of the effect of variable speed vvariable on time to
capture tr for CP of the pursuer and NR of the evader. In this case, the time to capture the
evader tr increases as average speed Av increases, only until the pursuer’s speed limit is
reached, after which tr decreases as Av increases. Consider the initial starting position at
(0, 8), which is represented in grey in the same figure. It is clear that as Av increases, tr
increases as well, but only up to a certain speed limit, after which tr starts decrease as Av
increases. When Av is 0.5, tr is found to be 11.6 seconds. When Av is increased to 0.7, tr
increases to 15.5 seconds. However, as Av increases to 0.85 and 0.94, tr decreases to 12.7
seconds and then to 11.7 seconds, respectively. The reason for this case’s results is
similar to that of CB’s pursuer strategy [Figure 3(b)], of which CP is a special case.

4.2.3 Evader strategy ST

Figure 4(a) depicts the effect of variable speed vvariable on time to capture tr for the CM of
the pursuer and the ST of the evader. In some cases tr increases when Av increases, while
in others, tr decreases when Av increases. Considering the initial starting position at
(–4, 9), tr increases from 22.1 seconds to 24.9 seconds and then to 25.4 seconds as Av
increases from 0.6 to 0.61 and then to 0.91. However, for the initial starting position of
(–2, 10), increasing Av to 0.6, 0.68 and 0.84 results in tr being 27.2 seconds, 25.1 seconds
and 24.3 seconds, respectively. The following is the explanation for the mixed results.
Because the ST is orthogonal to the baseline, increasing the evader’s speed results in only
a marginal increase in base length. This, combined with the pursuer’s efficient CM
strategy and possibly a reduced surprise effect due to fewer sharp turns at higher speed,
results in cases where time to capture does not improve as Av increases.
Figure 4(b) depicts the effect of variable speed vvariable on time to capture tr for the
pursuer’s CB and the evader’s ST. The figure clearly shows that as Av increases, tr
increases as well as decreases for almost all of the initial starting positions considered.
CB is less effective than CM in reducing baseline distance, whereas the ST results in a
marginal or no increase in baseline distance. With the two opposing effects, the evader’s
ST strategy results in increased capture time only in nearly 60% of the cases presented.
Figure 4(c) depicts the effect of variable speed vvariable on time to capture tr for the
pursuer’s CP and the evader’s ST. When Av increases, tr is found to increase as well as
decrease. This is because CB and CP pursuer strategies are less effective in reducing the
baseline distance when compared to the CM strategy, with a marginal increase in baseline
distance when using the ST strategy, resulting in opposing effects as in the case of
CB-ST. The results of Figure 4(c) are similar to those of Figure 4(b).
502 L.O. Singh and D. Rajagopalan

Figure 4 Effect of variable speed vvariable on time to capture tr for (a) CM (b) CB (c) CP under ST
(see online version for colours)

(a)

(b)

(c)
 Bio-inspired evasion strategies under variable evader speed 503

4.3 Higher variable speed, longer capture time

Logically, as the evader’s average speed increases, the evader can run away from the
pursuer faster. It is then expected that the pursuer will take longer to capture the evader,
implying a larger value of tr depending on the value of the threshold ρ used. However, in
some cases, possibly due to a loss of surprise effect with a less sharp turn and only a
marginal or no increase in baseline distance with increased evader speed, the time to
capture is even lower than with a fixed lower speed.
Figure 2 for the evader RE strategy clearly shows that a faster average speed of the
evader almost always results in a longer time to capture, regardless of the pursuer
strategy used. This is understandable because increased evader speed directly impacts RE
evader strategy (which assumes complete information about pursuer and evader
locations), resulting in increased baseline distance and, as a result, a longer time to
capture regardless of the pursuer strategy used. The CM strategy is already optimised in
terms of the specific effect of pursuer strategies (in the sense that the pursuer motion is
guaranteed to reduce the length of the baseline). As a result, the increased speed of the
evader has little effect on the increase in time to capture in the case of the pursuer’s CM
strategy. However, in the case of the CB strategy, the impact of increased speed on the
time to capture with the RE strategy of the evader is greater because the CB strategy is
not optimised to decrease baseline distance like the CM strategy. In other words, the
combination of RE strategy and increased speed gives the evader an advantage over the
pursuer, who is following a sub-optimal (in terms of ability to reduce baseline distance)
CB strategy. The evader appeared to avoid being captured in several cases examined
technically. CP, being a special case of CB, lies between CM and CB. Furthermore, due
to the evader’s RE strategy, the baseline distance increases consistently with increased
speed, resulting in a uniform increase in time to capture with increasing evader speed in
all cases of the pursuer strategy.
Figures 3 and 4 correspond to the evader’s NR and ST strategies. In Figure 4(a), the
ST is orthogonal to the baseline, so increasing the evader’s speed results in a marginal
increase in base length. This, combined with the pursuer’s efficient CM strategy (and
perhaps with less surprise effect due to less agility of the evader at higher speeds), results
in cases where time to capture does not improve as Av increases, as shown in Figure 4(a).
Figures 4(b) and 4(c), corresponding to CB and CP pursuer strategies, show the benefit of
faster evader speed to a greater extent. This is because the CB and CP pursuer strategies
are less effective in reducing baseline distance than the CM strategy. The ST strategy, on
the other hand, results in only a marginal increase in baseline distance. Due to the
opposing effects, the evader’s ST strategy only results in increased capture time in nearly
60% of the cases presented.
Figure 3 represents a NR evader strategy with a cyclic movement of the evader.
Hence, increasing the evader’s speed has only a limited effect on increasing the baseline
distance between the RE and ST cases. In terms of the number of cases with an increase
in capture time, the results of Figure 3 fall between those of Figures 2 and 4. Figure 3
shows an improvement in capture time in nearly 70% of the cases studied.
In conclusion, when the evader follows a RE control law, the increased speed of the
evader is beneficial in improving the ‘capture’ time, regardless of the pursuer strategy
used. Increased evader speed is less effective when using the ST and NR strategies
because increased speed does not increase the baseline distance as effectively in these
cases, and possibly due to reduced agility with less sharp turns at higher speeds. Thus, the
504 L.O. Singh and D. Rajagopalan

benefit of increased speed in terms of increased capture time results only in nearly two-
thirds of the cases considered in the case of the evader’s ST and NR strategies combined.

4.4 Less turning effect at a higher speed

An object’s ability to manoeuvre left or right is roughly inversely proportional to its
speed of motion. Higher evader speed in the ST or NR case may result in less ability to
turn sharply and thus less surprise effect or agility, with only a marginal increase in
baseline distance affecting the evader’s escape performance. In other words, contrary to
expectation, increasing evader speed results in either little or no change in evader
‘capture’ time. This is evident in the following examples:
1 Figure 3(a) (CM vs. NR) for initial conditions of (0, 8), (–1, 1) and (–2, 10).
2 Figure 3(b) (CB vs. NR) for initial conditions of (0, 8), (–2, 10) and (10, –1).
3 Figure 3(c) (CP vs. NR) for all initial conditions.
4 Figure 4(a) (CM vs. ST) for initial conditions of (0, 8), (–1, 1) and (–2, 10).
5 Figure 4(b) (CB vs. ST) for initial conditions of (0, 8), (–1, 1).
6 Figure 4(c) (CP vs. ST) for all initial conditions.
In the above cases, the effects of increased evader speed and decreased agility, with its
less surprising effect, appear to have had opposite effects on the time to capture. In the
NR and ST evader cases, this demonstrates decreasing returns on increasing speed for the
evader.

5 Theoretical support

This section attempts to offer theoretical justification for the empirical data-based
conclusions presented in Part 4. The empirical findings are in terms of capture time. The
analysis of capture time in PEG is difficult and does not exist in its complete form
(Glendinning, 2004). As a result, there is no one-to-one correlation between theoretical
analysis and actual data. However, the conditions under which the cost function tends to
decrease (towards the lower limit of –1 corresponding to the manifold) and the
guaranteed time period during which this decline (towards the manifold) can occur can be
demonstrated. The pursuit manifold is achieved when the cost function is set to –1. Once
on the manifold, the baseline distance is assured to drop due to the pursuer’s faster speed,
culminating in the eventual capture of the evader by the pursuer. The result is provided in
the form of two propositions, one to show the condition under which the cost function is
guaranteed to decrease and the other to show the guaranteed time period during which
this decrease will happen. The outcomes of the two propositions will assist in
conceptually explaining the empirical findings in Section 4 in general. As previously
established, there are nine combinations of pursuer and evader strategies. Motion
camouflage (CM), CB, and CP are the pursuer strategies. There are three types of evader
strategies: RE, NR, and ST. Proposition 3 provides theoretical solutions for each of the
nine scenarios independently. The outcome of Proposition 4 applies to all nine cases.
 Bio-inspired evasion strategies under variable evader speed 505

The cost function used in Propositions 3 and 4 is Γ in the CM case since it has a direct
bearing on the reduction of the baseline distance between pursuer and evader. The
derivative of Γ being non-positive will ensure that Γ will eventually reach –1 resulting in
a decrease in the baseline distance and the eventual capture of the evader. The derivative
of Γ is considered using the control input for the respective pursuer and evader strategies
employed to give the correct comparison between the different strategies.
Proposition 3: In the PEG considered in the paper, we can state that the cost function Γ is
such that Γ ≤ 0 provided the following conditions are satisfied for the respective
pursuer-evader strategies as below.
1 CM-RE

 1+ ν   1 + ν  (1 + ν)2
μ > β ν2   1 + + ; | r |> r0 > 0; β > 0; 0 < ν < 1; ε > 0 (20)
1− ν   ε (1 − ν) 2  (1 − ν)r0

where μ is the pursuer controller gain, β is the evader control gain, v is the evader
speed and 1 – Γ2 ≤ ε > 0.
2 CM-ST

 1 + v  1 + v γ | r | (1 + v)v 
2
μ>  +  ; | r |> r0 > 0; γ > 0; 0 < ν < 1; ε > 0 (21)
 1 − v   r0 (1 − v) ε 

where μ is the pursuer controller gain, γ is the evader control gain, v is the evader
speed and 1 – Γ2 ≤ ε > 0.
3 CM-NR

 1 + ν  1 + ν ψv (1 + v) 
2
μ>  +  ; | r |> r0 > 0; ψ > 0; 0 < ν < 1; ε > 0 (22)
 1 − ν   r0 (1 − v) 2 ε 

where μ is the pursuer controller gain, ψ is the evader control gain, v is the evader
speed and 1 – Γ2 ≤ ε > 0.
4 CB-RE
−1
 (1 − v)  r  (1 + v )r0  1 + v  (1 + v)v 2 (1 − v)v 2 
η>  .Ry p  −   + β  +  ;
 (1 − v)   r0 (1 + v) 
2 2 (23)
 ε (1 + v)  | r |  ε (1 − v)
| r |> r0 > 0; β > 0; 0 < ν < 1; ε > 0

where η is the pursuer controller gain, β is the evader control gain, v is the evader
speed and 1 – Γ2 ≤ ε > 0.
5 CP-RE
For the case of CP-RE substitute R = I2 in the case CB-RE above.
6 CB-ST
506 L.O. Singh and D. Rajagopalan

−1
 (1 − v)  r  (1 + v)  1 + v γ | r | (1 + v) 
η> 2 
.Ry p  −   + ;
 ε (1 + v)  | r |  r0 (1 − v )   r0 ε (1 − v)  (24)
| r |> r0 > 0; γ > 0; 0 < ν < 1; ε > 0

where η is the pursuer controller gain, γ is the evader control gain, v is the evader
speed and 1 – Γ2 ≤ ε > 0.
7 CP-ST
For the case of CP-ST substitute R = I2 in the case CB-ST above.
8 CB-NR
−1
 (1 + v)  1 + v ψv (1 + v) 
2
 (1 − v)  r
η> 2 
.Ry p −   + ;
 ε (1 + v)  | r |  r0 (1 − v )   r0 ε (1 − v) 2  (25)
| r |> r0 > 0; γ > 0; 0 < ν < 1; ε > 0

where η is the pursuer controller gain, ψ is the evader control gain, v is the evader
speed and 1 – Γ2 ≤ ε > 0.
9 CP-NR
For the case of CP-NR, substitute R = I2 in the case of CB-NR above.
Proof: See Appendix 1.
Remark 1: In the case of CM-RE, CM-NR, CB-RE and CB-NR, the high gain condition
on the pursuer is such that the higher the evader speed v, the higher the bound on the
RHS of (20), (22), (23) and (25) respectively, and hence a higher pursuer control gain μ
or η is required for Γ ≤ 0. In the case of CP, the high gain condition corresponds to
R = I2. For Γ to decrease towards (–1), a higher gain μ and η is required for a larger v. For
a larger v, if Γ ≠ 0 or Γ > 0, PEG moves away from manifold, thus increasing the
chance of a higher capture time or even the escape of the evader.
Remark 2: In particular, in the case of CM-ST and CB-ST, the high gain condition on the
pursuer is such that the larger |r|, the baseline distance, the larger the bound on the RHS
of (21) and (24) respectively, and hence a higher pursuer control gain μ and η are
required for Γ ≤ 0. For Γ to decrease towards (–1), higher gain μ or η is required for large
|r|. For large |r|, if Γ ≠ 0 or Γ > 0, PEG moves away from manifold and hence has higher
chances of increased capture time or even escape. For smaller |r| however, the bound on
the RHS of (21) and (24) decreases, thus increasing the capture at a shorter time for a
given pursuer control gain. The comment also applies to the pursuer’s CP case with
R = I2.
Proposition 4: With the condition stated in Proposition 3 ensuring that
Γ ≤ 0 (26)
 Bio-inspired evasion strategies under variable evader speed 507

for the respective pursuer-evader strategies case, we can further state that the period T for
which (26) holds can be guaranteed to be such that
r (0) − r0
T= (27)
1+ v
Proof: See Appendix 2.
Remark 3: The conditions derived in Proposition 3 guarantee that Γ ≤ 0. That is, the cost
function tends towards –1. As seen in (27), as v increases with higher evader speed, T
decreases. This means that the guaranteed time in which Γ ≤ 0 decreases as v increases.
This means less chance of reaching the respective manifold, resulting in PEG becoming
more evasive, thus indirectly supporting the case of a larger capture time as the evader
speed v increases.

6 Conclusions

Typically, in bio-inspired studies of pursuit evasion games (PEG), the evader and pursuer
speeds are considered constant, with the pursuer speed being greater than the evader
speed. To mimic nature, it is assumed in this paper that the evader’s speed (albeit less
than the pursuer’s speed) changes roughly inversely proportional to the baseline distance
between the pursuer and the evader. The goal is to determine the effect of increasing
evader speed on the PEG capture time.
A competent analysis of the above problem has been attempted in this paper, both
through computation and theoretically. Nine possible combinations of bio-inspired
pursuer-evader strategies have been assumed and analysed. Further, different initial
starting positions of the pursuer relative to the evader are considered in the simulation to
exclude any directional bias in the results. In addition, technical capture is assumed when
the baseline distance falls below a particular threshold to avoid asymptotic responses
distorting the results. Computer simulation of continuous trajectories of the pursuer and
evader help to capture the initial capture time. In the case of an evader’s RE strategy, it is
concluded that higher evader speed provides an evader with a natural advantage in terms
of longer capture time or even complete escape in a technical sense. However, in the case
of other evader strategies, such as NR or STs, the increased evader speed is balanced
against the decreased agility resulting in evader’s less sharp turn, combined with a
marginal or no increase in baseline distance, yielding a mixed result in terms of the
evader’s escape performance. The empirical results of the simulation show that faster
speed and the advantages of agile turns may not go together. The theoretical analysis
indirectly supports the empirical proof that increasing evader speed reduces the certainty
of reaching the manifold, implying a longer capture time. Further study might examine
the practical scenario when the speeds of the pursuer and evader are uncertain owing to
possible obstacles.
508 L.O. Singh and D. Rajagopalan

References
Aboulem, S., Boufounas, E. mahout and Boumhidi, I. (2019) ‘Intelligent proportional-integral
sliding mode control of wind turbine systems based particle swarm optimisation’,
International Journal of Automation and Control, Vol. 13, No. 3, p.347,
https://doi.org/10.1504/IJAAC.2019.098585.
Ahamed, S.R. et al. (2020) ‘A comparative evaluation of PID-based optimisation controller
algorithms for DC motor’, International Journal of Automation and Control, Vol. 14, No. 5/6,
p.634, https://doi.org/10.1504/IJAAC.2020.110076.
Ardeshiri, R.R., Kashani, H.N. and Ahrabi, A.R. (2019) ‘Design and simulation of self-tuning
fractional order fuzzy PID controller for robotic manipulator’, International Journal of
Automation and Control, Vol. 13, No. 5, p.595, https://doi.org/10.1504/IJAAC. 2019.101912.
Boothby, W.M. (1975) An Introduction to Differentiable Manifolds and Riemannian Geometry,
Academic Press, New York, USA.
Caro, T.M. (2005) Antipredator Defenses in Birds and Mammals, Illustrate, University of Chicago
Press, Chicago, USA.
Gerkey, B.P., Thrun, S. and Gordon, G. (2006) ‘Visibility-based pursuit-evasion with limited field
of view’, The International Journal of Robotics Research, Vol. 25, No. 4, pp.299–315,
https://doi.org/10.1177/0278364906065023.
Glendinning, P. (2004) ‘The mathematics of motion camouflage’, Proceedings of the Royal Society
B: Biological Sciences, Vol. 271, No. 1538, pp.477–481, https://doi.org/10.1098
/rspb.2003.2622.
Goud, E.C., Chidambaram, M. and Rao, A.S. (2022) ‘Novel design of PID controllers for minimum
and non-minimum phase time delay processes for enhanced performance’, International
Journal of Automation and Control, Vol. 16, No. 6, p.689, https://doi.org/10.1504
/IJAAC.2022.126082.
Hedenström, A. and Rosén, M. (2001) ‘Predator versus prey: on aerial hunting and escape
strategies in birds’, Behav. Ecol., p.12, https://doi.org/10.1093/beheco/12.2.150.
Howland, H.C. (1974) ‘Optimal strategies for predator avoidance: the relative importance of speed
and manoeuvrability’, Journal of Theoretical Biology, Vol. 47, No. 2, pp.333–350,
https://doi.org/10.1016/0022-5193(74)90202-1.
Issacs, R. (1965) Differential Games, Wiley, New York.
Justh, E.W. and Krishnaprasad, P.S. (2005) ‘Natural frames and interacting particles in three
dimensions’, Proceedings of the 44th IEEE Conference on Decision and Control, and the
European Control Conference, CDC-ECC ‘05, pp.2841–2846, https://doi.org/10.1109
/CDC.2005.1582594.
Justh, E.W. and Krishnaprasad, P.S. (2006) ‘Steering laws for motion camouflage’, Proceedings of
the Royal Society A: Mathematical, Physical and Engineering Sciences, Vol. 462, No. 2076,
pp.3629–3643, https://doi.org/10.1098/rspa.2006.1742.
Kopparty, S. and Ravishankar, C.V. (2005) ‘A framework for pursuit evasion games in Rn’,
Information Processing Letters, Vol. 96, No. 3, pp.114–122, https://doi.org/10.1016
/J.IPL.2005.04.012.
Lee, S., Polak, E. and Walrand, J. (2013) ‘A receding horizon control law for harbor defense’, 51st
Annual Allerton Conference on Communication, Control and Computing, IEEE, pp.70–77,
https://doi.org/10.1109/Allerton.2013.6736507.
Li, W. (2016) ‘The confinement-escape problem of a defender against an evader escaping from a
circular region’, IEEE Transactions on Cybernetics, Vol. 46, No. 9, pp.2166–2172,
https://doi.org/10.1109/TCYB.2016.2541158.
Li, W. (2017) ‘A dynamics perspective of pursuit-evasion: capturing and escaping when the
pursuer runs faster than the agile evader’, IEEE Transactions on Automatic Control, Vol. 62,
No. 1, pp.451–457, https://doi.org/10.1109/TAC.2016.2575008.
 Bio-inspired evasion strategies under variable evader speed 509

Lin, Y. and Saripalli, S. (2017) ‘Sampling-based path planning for UAV collision avoidance’, IEEE
Transactions on Intelligent Transportation Systems, Vol. 18, No. 11, pp.3179–3192,
https://doi.org/10.1109/TITS.2017.2673778.
Mueggler, E., Faessler, M., Fontana, F. and Scaramuzza. D. (2014) ‘Aerial-guided navigation of a
ground robot among movable obstacles’, 12th IEEE International Symposium on Safety,
Security and Rescue Robotics, SSRR 2014 - Symposium Proceedings, [Preprint],
https://doi.org/10.1109 /SSRR.2014.7017662.
Obiroy Singh, L. and Devanathan, R. (2017) ‘Performance improvement of bio-inspired strategies
through feedback laws’, 8th International Conference on Innovations in Bio-Inspired
Computing and Applications, Marrakech, Morroco, Springer, pp.143–156, https://doi.org
/10.1007/978-3-319-76354-5_14.
Obiroy Singh, L. and Devanathan, R. (2018) ‘Comparative study of the performance of application
of bio-inspired strategies to pursuit evasion game under feedback laws’, in The 18th
International Conference on Control, Automation and Systems, Pyeongchang, South Korea,
Pyeongchang, South Korea, IEEE, pp.276–281, https://ieeexplore.ieee.org/document/8571948.
Pachter, M. (1987) ‘Simple-motion pursuit-evasion in the half plane’, Computers & Mathematics
with Applications, Vol. 13, Nos. 1–3, pp.69–82, https://doi.org/10.1016/0898-1221(87)90094-
0.
Pachter, M. and Yavin, Y. (1981) ‘A stochastic homicidal chauffeur pursuit-evasion differential
game’, Journal of Optimization Theory and Applications, Vol. 34, No. 3, pp.405–424,
https://doi.org/10.1007/BF00934680.
Pais, D. and Leonard, N. (2010) ‘Pursuit and evasion: evolutionary dynamics and collective
motion’, AIAA Guidance, Navigation, and Control Conference, pp.1–14, https://doi.org
/10.2514/6.2010-7584.
Pati, A. and Negi, R. (2019) ‘An optimised 2-DOF IMC-PID-based control scheme for real-time
magnetic levitation system’, International Journal of Automation and Control, Vol. 13, No. 4,
p.413, https://doi.org/10.1504/IJAAC.2019.100468.
Sgall, J. (2001) ‘Solution of David Gale’s lion and man problem’, Theoretical Computer Science,
Vol. 259, Nos. 1–2, pp.663–670, https://doi.org/10.1016/S0304-3975 (00)00411-4.
Shneydor, N.A. (1998) ‘Missile guidance and pursuit: kinematics, dynamics and control’, The
Aeronautical Journal, Vol. 102, No. 1017, p.406, https://doi.org /10.1017
/S0001924000065210.
Simetti, E., Turetta, A., Casalino, G., Storti, E. and Cresta, M. (2010) ‘Protecting assets within a
civilian harbour through the use of a team of USVs: interception of possible menaces’, in
IARP Workshop on robots for Risky Interventions and Environmental Surveillance
Maintenance (RISE’10).
Singh, L.O. and Devanathan, R. (2019) ‘Study of performance of bio-inspired strategies applied to
pursuit evasion game under feedback laws’, Advances in Science, Technology and
Engineering Systems, Vol. 4, No. 3, pp.207–219, https://doi.org/10.25046/aj040328.
Svec, P., Thakur, A., Shah, B.C. and Gupta, S.K. (2012) ‘USV trajectory planning for time varying
motion goals in an environment with obstacles’, Proceedings of the ASME 2012 International
Design Engineering Technical Conferences and Computers and Information in Engineering
Conference, 36th Mechanisms and Robotics Conference, Parts A and B, Chicago, Illinois,
USA, 12–15August, Vol. 4, pp.1297–1306, ASME, https://doi.org/10.1115/DETC2012-
71239.
Tang, T.F., Chong, S.H. and Pang, K.K. (2020) ‘Stabilisation of a rotary inverted pendulum system
with double-PID and LQR control: experimental verification’, International Journal of
Automation and Control, Vol. 14, No. 1, p.18, https://doi.org/10.1504 /IJAAC.2020. 103799.
Von Neumann, J. and Morgenstern, O. (1945) ‘Theory of games and economic behavior’, J.
Philos., September, Vol. 42, No. 20, p.550, doi: 10.2307/2019327.
Wei, E., Justh, E.W. and Krishnaprasad, P.S. (2009) ‘Pursuit and an evolutionary game’,
Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences,
Vol. 465, No. 2105, pp.1539–1559, https://doi.org/10.1098/rspa.2008.0480.
510 L.O. Singh and D. Rajagopalan

Weihs, D. and Webb, P.W. (1984) ‘Optimal avoidance and evasion tactics in predator-prey
interactions’, Journal of Theoretical Biology, Vol. 106, No. 2, pp.189–206,
https://doi.org/10.1016/0022-5193(84)90019-5.
Wu, J., Wang, H., He, S. and Lin, D. (2020) ‘Composite finite-time convergent guidance law for
maneuvering targets with second-order autopilot lag’, Asian Journal of Control, Vol. 22,
No. 1, pp.556–569, https://doi.org/https://doi.org/10.1002/asjc.1922.
Yi, L., Xu, X., Tan, M., Zhang, Z., Xiao, W. and Fan, L. (2021) ‘PID self-tunning method based on
deep belief network and improved firefly algorithm’, International Journal of Automation and
Control, Vol. 15, No. 3, pp.363–377.

Appendix 1

Proof of Proposition 3:
1 CM – RE:
 r ⊥   r ⊥
up = −μ  . r  , μ > 0; ue = β  . re  , β > 0;
| r |  | r | 
⊥ ⊥ ⊥
r = rp − re ; r = rp − re ; r = rp − re

 r   r   r   r 
∴ ue = β  . re ⊥  = β  ( )
. rp ⊥ − rr ⊥  ; ue = β  . rp ⊥  − β  . re ⊥ 
| r |  | r |  | r |  | r | 
From (Justh and Krishnaprasad, 2006), we can write,
2
 r ⊥ 
 | r |. r 
1 − Γ2 =  
| r |2

and

r  1  r ⊥   1  1  r ⊥  
2
Γ =  2
| r |  | r |  | r |
. r   +
  | r |  | r |  | r | 
(
 2  . r   1 − ν ( x p .xe ) u p )
1  1  r ⊥ 
+
| r |  | r |  | r |
(
2
)
 2  . r   ν − ( x p .xe ) ν ue

that is

r  1  r ⊥  μ  1  r ⊥ 
2 2
Γ = 
| r |  | r |2| r | . r −
  | r |   2  | r |
 
.(
r )
  1 − ν ( x p . xe )
  | r |   
1   r   r ⊥   1  r ⊥ 
+ β
| r |   | r |
. rp ⊥  − β 
  | r |
(
   | r |  | r | 
)
. r    2  . r   ν − ( x p . xe ) ν 2
 Bio-inspired evasion strategies under variable evader speed 511

r [ μ [
Γ = 1 − Γ2 ] − 1 − Γ 2 ] 1 − ν ( x p . xe )
( )
|r| 
|r|
β  r  β [
+ 2
1 − Γ2  ( )
. rp ⊥  ν − ( x p . xe ) ν 2 − 1 − Γ 2 ] ν − ( x p . xe ) ν 2
( )
| r | | r |  | r |

We note that
−1 ≤ x p . xe ≤ 1; 1 − ν ≤ 1 − ν ( x p . x ) ≤ 1 + ν; ν − 1 ≤ ν − ( x p . xe ) ≤ 1 + ν
e

 r μ β  β
∴ Γ ≤ (1 − Γ 2 )  − (1 − ν) − (ν − 1)ν 2  + 2 1 − Γ 2 (1 + ν)ν 2 ,
 | r | | r | | r |  | r |
Since

 r 
 . r ⊥  ≤ 1
| r | 
also
1 − ν ≤ | r |≤ 1 + ν, let | r |> r0 > 0

1 + ν μ(1 − ν) β (1 − ν)ν 2  β 1 − Γ 2 (1 + ν)ν 2

∴ Γ ≤ (1 − Γ 2 )  − + +
 r0 1+ ν | r |  | r |2

1 + ν μ (1 − ν) β (1 − ν)ν 2  β 1 − Γ 2 (1 + ν)ν 2
≤ (1 − Γ 2 )  − + +
 r0 1+ ν 1− ν  (1 − ν) 2

Let

 β (1 − Γ ) (1 + ν)ν
2 2
1 + ν μ(1 − ν)
1 − Γ 2 ≤ ε > 0; Γ ≤ (1 − Γ 2 )  − + β ν2  +
 r0 1+ ν 
2
(1 − ν) ε

1 + ν μ (1 − ν) β (1 + ν)ν 2 
Γ ≤ (1 − Γ 2 )  − + β ν2 + 
 r0 1+ ν (1 − ν) 2 ε 

Let

 1− ν  1+ ν β (1 + v)v 2   C 
C1 =  μ− − β ν 2 ; C2 = ; Γ ≤ − (1 − Γ 2 ) C1 − 2 
 1+ ν  r0 (1 − v) 2
 ε
To ensure,
C2  1− ν  (1 + ν) β (1 + ν)ν 2
C0 = C1 − > 0,   μ− − β ν2 >
ε  1+ ν  r0 ε (1 − ν) 2

or
512 L.O. Singh and D. Rajagopalan

 1+ ν   1 + ν  (1 + ν )2
μ > β ν2   1 + + ; | r |> r0 > 0; β > 0; 0 < ν < 1; ε > 0
1− ν   ε (1 − ν) 2  (1 − ν)r0

With the high gain condition on μ as above, Γ ≤ − (1 − Γ 2 ) C0 ; | Γ |≤ 1; 1 − Γ 2 ≥ 0.

If C0 ≥ 0, we have Γ ≤ 0, with |r| > r0 > 0. Hence the result. ■
2 CM – ST:
 r ⊥
up = −μ  . r  , μ > 0; ue = γ ( r . r ) , γ > 0.
| r | 
From (Justh and Krishnaprasad, 2006),
2
 r ⊥ 
 | r |. r 
1 − Γ2 =  
| r |2

and

r  1  r   1  1  r
2

Γ =  2
| r |  | r |  | r |
. r ⊥   +  2 
  | r |  | r |  | r |
 
(
. r ⊥   1 − ν ( x p . xe ) u p)
1  1  r 
+  2 
| r |  | r |  | r |
( (
. r ⊥   ν − x p . xe

)) ν u
2
e

That is,

r  1  μ  1  r ⊥ 2 
2
 r ⊥
Γ = 
| r |  | r |2 | r | . 
r
 
(
  − | r |   2  | r | . r   1 − ν ( x p . xe ) )
  | r |   
1  1  r 
+
| r |  | r |2  | r |
( 
)
. r ⊥   ν − ( x p . xe ) ν 2 ( γ r . r )

r [ μ [ γ r . r
Γ = 1 − Γ2 ] − 1 − Γ 2 ] 1 − ν ( x p . xe ) +
( ) ( )
1 − Γ 2 ν − ( x p . xe ) ν 2
|r| | r | | r |2

r [ μ [
Γ = 1 − Γ2 ] − 1 − Γ 2 ] 1 − ν ( x p . xe )
( )
|r| | r |
γ | r || r | rˆ . rˆ
+
| r |2
(
1 − Γ 2 ν − ( x p . xe ) ν 2 )
We note that,

−1 ≤ x p . xe ≤ 1; −1 ≤ rˆ . rˆ ≤ 1,

since both are unit vectors,

 Bio-inspired evasion strategies under variable evader speed 513

1 − ν ≤ 1 − ν ( x p . x ) ≤ 1 + ν; ν − 1 ≤ ν − ( x p . xe ) ≤ 1 + ν;
e
1 − v ≤ r ≤ 1 + ν; | r |> r0 > 0

1+ v ( μ (1 − Γ 2 ) (1 − v) γ | r |
Γ ≤ 1 − Γ2 ) − + (1 + v)v 2 1 − Γ 2
r0 1+ v 1− v

Let

1 − Γ2 1 − Γ2
1 − Γ 2 ≤ ε > 0;  1 − Γ 2 = ≤
1 − Γ2 ε

1 + v μ (1 − v) γ | r | (1 + v)v 2 
Γ ≤ (1 − Γ 2 )  − + 
 r0 1+ v ε (1 − v) 

 μ(1 − v) 1 + v γ | r | (1 + v)v 2  2  C2 
Γ ≤ − (1 − Γ 2 )  − −   Γ ≤ − (1 − Γ ) C1 − 
 1 + v r0 ε (1 − v )   ε

where

μ (1 − v) 1+ v γ | r | (1 + v)v 2 C
C1 = − ; C2 =  C0 = C1 − 2 > 0, ,
1+ v r0 1− v ε

then

C2
C1 >
ε

μ (1 − v) 1 + v γ | r | (1 + v)v 2  1 + v  1 + v γ | r | (1 + v)v 
2
− >  μ>  +  ■
1+ v r0 (1 − v) ε  1 − v   r0 (1 − v) ε 

3 CM – NR: Following the derivation similar to cases (1) and (2), we obtain

 μ(1 − v) 1 + v ψv 2 (1 + v)   C 
Γ ≤ − (1 − Γ 2 )  − − 2
 Γ ≤ − (1 − Γ 2 ) C1 − 2 
 (1 + v) r0 ε (1 − v)   ε

μ(1 − v) 1 + v ψv 2 (1 + v) C
C1 = − ; C2 = 2
; C0 = C1 − 2 > 0, ,
(1 + v) r0 (1 − v) ε

then

C2
C1 > for the condition (21).
ε
514 L.O. Singh and D. Rajagopalan

4 CB – RE: Following the derivation similar to cases (1) and (2), we obtain
 1  η(1 − v)  r  β (1 + v)v
2

  2 
. Ry p  − 2 
 ε  (1 + v)  | r |  (1 − v) 
Γ ≤ − (1 − Γ 2 )  
 −  1 + v + η(1 + v) + β v (1 − v) 
2

  r0 r0 (1 − v) (1 + v)  
 
C 
 Γ ≤ − (1 − Γ 2 )  2 − C1 
 ε 

1 + v η(1 + v) β v 2 (1 − v) η(1 − v)  r  β (1 + v)v

2
C1 = + + ; C2 =  . Ry p − ;
r0 r0 (1 − v) (1 + v) (1 + v) 2  | r |  (1 − v) 2
C
C0 = 2 − C1 > 0,
ε
C2
then > C1 for the condition (22).
ε
5 CP – RE: The result follows as in the case in CB-RE with R = I2.
6 CB – ST: Following the derivation similar to cases (1) and (2), we obtain.
 1  η(1 − v)  r  γ r (1 + v)   1 + v η(1 + v)  
Γ ≤ − (1 − Γ 2 )   2 
. Ry p  − − + 
 ε  (1 + v)  | r |  (1 − v)   r0 r0 (1 − v)  
C 
 Γ ≤ − (1 − Γ 2 )  2 − C1 
 ε 
1 + v η(1 + v) η(1 − v)  r  γ | r | (1 + v) C
C1 = + ; C2 = 2 
. Ry p  − ; C0 = 2 − C1 > 0,
r0 r0 (1 − v) (1 + v)  | r |  (1 − v ) ε
C2
then > C1 for the condition (23).
ε
7 CP – ST: The results follow from the case CB-ST putting R = I2.
8 CB – NR: Following the derivation similar to cases (1) and (2) we obtain.
 1  η(1 − v)  r  ψv 2 (1 + v)   1 + v η(1 + v)  
Γ ≤ − (1 − Γ 2 )    . Ry − − + 
r0 (1 − v)  
2 p
 ε  (1 + v)  | r |  (1 − v) 2   r0
C 
 Γ ≤ − (1 − Γ 2 )  2 − C1 
 ε 
2
1 + v η(1 + v) η(1 − v)  r  ψv (1 + v) C
C1 = + ; C2 = 2 
. Ry p  − 2
; C0 = 2 − C1 > 0, ,
r0 r0 (1 − v) (1 + v)  | r |  (1 − v) ε
C2
then > C1 for the condition (24).
ε
9 CP – NR: The results follow from the case CB-NR putting R = I2. ■
 Bio-inspired evasion strategies under variable evader speed 515

Appendix 2

Proof of Proposition 4:
Assume the gain condition as given in Proposition 3. To estimate how long the Γ ≤ 0 can
be guaranteed, we proceed as follows (Justh and Krishnaprasad, 2006)
d |r|
≥ −(1 + v).
dt
Integrating both sides
r (t ) − r (0) ≥ −(1 + v)t , ∀t ≥ 0  r (t ) ≥ r (0) − (1 + v) t ≥ r0

Limiting the value of r0 is reached at t = T > 0. That is

r (0) − r0
r (0) − (1 + v)T = r0 ; T = ;
1+ v
T is the period when |r| ≥ r0. Hence the result. ■