Original article

Proc IMechE Part I:

J Systems and Control Engineering
2017, Vol. 231(3) 178–188
Robust control of nonholonomic Ó IMechE 2017
Reprints and permissions:
wheeled mobile robot with past sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0959651817691641

information: Theory and experiment journals.sagepub.com/home/pii

Spandan Roy1, Sambhunath Nandy2, Indra Narayan Kar1, Ranjit Ray2 and
Sankar Nath Shome2

Abstract
In this article, a robust hybrid control method is presented for efficient path tracking control of a nonholonomic wheeled
mobile robotic system under parametric and nonparametric variations. The present control law is a paradigm shift to
control a wheeled mobile robot over a predefined trajectory by fusing the best features of the switching control logic as
well as time-delayed control logic. The proposed hybrid control strategy aims at reducing the effort required for model-
ing the complex wheeled mobile robotic systems by approximating the unknown dynamics using input and feedback
information of past time instances. Furthermore, the proposed methodology significantly reduces the approximation
error arising from finite time-delay through the switching logic without any prior knowledge of the uncertainty bound. A
new stability analysis for the time-delayed control is proposed which establishes an analytical relation between the con-
troller performance and the approximation error. Performance of the proposed hybrid controller is tested with a real-
life wheeled mobile robot and improved tracking performance is observed compared to conventional robust control
strategies even with the incorporation of dynamic parametric uncertainties.

Keywords
Nonholonomic system, robust control, time-delayed control, wheeled mobile robot

Date received: 5 March 2016; accepted: 3 January 2017

Introduction Lyapunov function and proposed a sliding mode con-

trol (SMC) approach to control the robotic platform
Nonholonomic systems can be defined as the systems over the desired path. Jiang and Pomet3 designed adap-
which obey certain non-integrable constraints. These tive control for the nonholonomic systems to tackle
constraints may be results of some physical kinematic unknown parameters based on backstepping method.
constraints which are imposed on the system. The Hamel and Meizel4 reported a robust controller to
constraints, expressed in terms of the generalized negotiate the localization errors. Aguilar et al.5 pro-
system velocities, cannot be integrated over time. posed an SMC-based path-following controller to
Nonholonomic wheeled mobile robots with their ensure robustness against the curvature and localiza-
large application domains ranging from transporta- tion estimation error for a car-like robot. Chawa6 used
tion, planetary exploration, military applications to SMC for trajectory tracking as well as posture stabili-
human–machine interfaces for people with mobility zation of a WMR. Ryu and Agarwal7 proposed a
deficiency, have received a lot of attentions in the past tracking controller based on robust outer loop control
two decades.
The controllers designed for wheeled mobile robot 1
Department of Electrical Engineering, Indian Institute of Technology,
(WMR) to track a predefined path, as reported in the
Delhi, New Delhi, India
literature, can be broadly defined into two categories: 2
CSIR-Central Mechanical Engineering Research Institute, Durgapur, India
kinematic model based and dynamic model based. De
Wit and Khennouf1 proposed a kinematic model–based Corresponding author:
continuous piecewise controller to stabilize the system Spandan Roy, Department of Electrical Engineering, Indian Institute of
Technology, Delhi, A-24, Nilgiri Hostel, Hauz Khas, New Delhi 110016,
exponentially around the origin. Guldner and Utkin2 India.
generated the desired trajectories based on the Email: sroy002@gmail.com
Roy et al. 179

(ROC) to compensate the effect of slip. However, it uncertainties in actuator while the core controller was
requires a predefined bound of the slip angle. Shi et al.8 based on feedback linearization method ignoring the
reported a controller for the case when the wheel velo- system uncertainty. However, implementation of feed-
city is not available without considering any parametric back linearization–based control or inverse dynamics–
uncertainty. An integral sliding mode controller is based control is not always feasible in real-life scenario
designed in Bessas et al.9 based on the kinematic model as it requires exact knowledge of the hardware para-
of the system. Onat and Ozkan10 devised an adaptive meters. Shojaei22 reported a robust controller with
controller for parameter identification by considering known uncertainty bound considering actuator limita-
multiple kinematic models. The controller is switched tion. Time-delayed control (TDC)23–26 approximates
between various kinematic models based on the track- all the uncertainties using control input and state infor-
ing error. mation of the immediate past time instant. However,
The kinematic model of WMR system does not take performance of TDC gets affected by the time-delayed
into account the effects of parametric uncertainties. error (TDE) introduced by the finite delay time. Chen
Therefore, the performances of these controllers fade et al.27 adopted an adaptive SMC (ASMC) method
away when uncertainties come into effect or the sys- where adaptive control is used to approximate the
tem’s velocity varies or even when the WMR system is upper bounds of uncertainty. However, the adaptation
directed to follow complex path. Moreover, assump- law makes the switching gain a monotonically increas-
tions made during the kinematic modeling of the system ing function of errors and makes the system susceptible
are often responsible for modeling errors, which in turn to chattering due to high switching gain.
pose hurdles for efficient control of WMR. The above- This article has two major contributions. First, a new
mentioned circumstances make it necessary to consider stability analysis for TDC is provided based on the sec-
the dynamic model of WMR to improve the trajectory ond method of Lyapunov. The proposed stability
tracking of WMR.11 Gradually, many researchers approach helps in realizing the effect of TDE on the
designed controllers based on the dynamic model of the tracking performance of TDC analytically. The second
system. Yang and Kim12 represented the posture in polar contribution of the article augments the first contribution
coordinates and feedback-linearized the dynamic equa- by a novel hybrid control strategy, namely, time-delayed
tion of WMR using computed-torque approach. They robust outer loop control (TDROC). The proposed
also reported a robust trajectory tracking controller using methodology aims to amalgamate the inherent advan-
SMC. Corradini and Orlando13 used SMC in discrete tages of switching logic and time-delayed logic, while
time domain for trajectory tracking control of WMR. overcoming their respective shortcomings. The proposed
Lee et al.14 reported reaching law–based SMC for trajec- TDROC method leads to the following advantages:
tory tracking of WMR. Hung et al.15 used backstepping
strategy together with SMC for path-following purpose  Significant reduction in modeling effort as knowl-
with constant velocity. Lee et al.16 devised an SMC based edge of unknown payload variation and unknown
on approach angle in Cartesian coordinate. One major dynamics such as slip and friction is not required.
disadvantage of switching control (both ROC and SMC)  Again, approximation of uncertain terms reduces
approach is the possibility of chattering effect due to high the switching gain and consequently helps attain
switching gains which, in turn, may activate high- better tracking accuracy by reducing the error
frequency unmodeled dynamics causing harmful effect bandwidth.28
on robotic system. To add more, deciding the uncertainty  With the presence of switching logic, TDROC is
bounds necessitates nominal modeling of the system, further able to negotiate TDE using past data with-
which again is a tedious task for a complex system such out the knowledge of uncertainty bounds.
as WMR.
Oriolo et al.17 reported dynamic feedback lineariza- The proposed controller is tested on a real-life
tion control for trajectory tracking as well as posture WMR, ‘‘Drishti,’’26,29 and improved tracking perfor-
stabilization and compared the performance of the mance is noted compared to other controllers, namely,
same against linear feedback linearization and non- ROC,7 TDC26 and ASMC.27
linear feedback linearization. Sun18 proposed a combi- The organization of the rest of the article is as fol-
nation of pole placement and exact feedback lows: the dynamics of WMR is briefly described in sec-
linearization method for the purpose of devising a tra- tion ‘‘Dynamics of WMR,’’ section ‘‘Controller design’’
jectory tracking controller. Cho et al.19 reported a feed- describes the proposed control methodology in detail
back linearization–based control law using a least and section ‘‘Experimental results and comparison’’ ela-
square discrete friction observer. However, the control- borates the experimental results followed by conclusion
ler requires a predefined bound on the friction and the which is presented in section ‘‘Conclusion.’’
effect of discretization error is not considered. Shojaei
et al.20 designed an inverse dynamics and robust pro-
Dynamics of WMR
portional–integral–derivative (PID)-based approach
for the trajectory tracking control strategy. Arab and In Campion and Chung,30 different perspectives of
Fateh21 used fuzzy logic to approximate the parametric modeling of a general WMR (both omni-directional
180 Proc IMechE Part I: J Systems and Control Engineering 231(3)

and nonholonomic) are mentioned based on different

types of wheels. Considering toroidal wheels, a planar
mechanism is adopted by Chakraborty and Ghosal31
for a three-wheeled nonholonomic WMR with lateral
movement capability. The WMR is modeled with
appropriately designed joints and links, and each wheel
was represented with a 3-degree-of-freedom (dof) pla-
nar joint. However, considering the rolling without
slipping condition, two nonholonomic constraints were
considered for each of the planar joints of the wheel
(thus each joint has one holonomic and two nonholo-
nomic constraints). Hence, Chakraborty and Ghosal31
termed such planar joints as 3-dof ‘‘nonholonomic
joints’’ rather than considering conventional 3-dof
spherical joint. Such mechanisms, which are themselves
different and one of its kind works, require dedicated
model-dependent path planning and control frame- Figure 1. Schematic diagram of the wheeled mobile robot.
work. However, in this work, we have adopted the
more commonly and widely used state-space model30,32 of the WMR. Again, to denote coordinates of any point
for ease of controller design which is applicable to on the wheels, two more variables ur and ul are defined
work in a planar terrain only. which specify rotation of the right and left wheels,
The Lagrange–Euler-based equations of motion of respectively. So the generalized coordinate vector of the
WMR derived in Nandy et al.32 subjected to constraints system containing the five generalized coordinates is
are reproduced below defined as q 2 <5 = (xc , yc , u, ur , ul )T .
The constraint equations (rolling without slipping)
m€ € sin u + u_ 2 cos u)
xc + K(u for WMR are given by
l1 sin u  cos u(l2 + l3 ) = 0 ð1Þ
2
y_ c cos u  x_ c sin u  ud
_ =0 ð6Þ
yc  K(u
m€ € cos u  u_ sin u)
b
+ l1 cos u  sin u(l2 + l3 ) = 0 ð2Þ x_ c cos u + y_ c sin u + u_  ru_ r = 0 ð7Þ
2
b b
€ + K(€
Iu xc sin u  y€c cos u)  dl1 + (l3  l2 ) = 0 x_ c cos u + y_ c sin u  u_  ru_ l = 0 ð8Þ
2 2
ð3Þ
Equations (1)–(5) are written compactly in the form
Iw €
ur + l3 r = t r ð4Þ
Iw €
ul + l3 r = t l ð5Þ M(q)€q + H(q, q) _ = Bt  AT (q)l ð9Þ
2 3
with m = mP + 2mw , and I = IP + mP d2 + 2mw m 0 K sin u 0 0
2 2 6 0 m K cos u 0 0 7
((b =4) + d ) + 2Im . These notations are explained 6 7
6
where M(q)= 6 K sin u K cos u 0 07
subsequently. I 7,
4 0 0 0 Iw 0 5
Let the WMR (Figure 1) driven by two motors con-
nected with the two wheels having radius r be separated 0 0 0 0 Iw
2 2 3 2 3
to a distance b. The length of the vehicle is considered Ku_ cos u 0 0
to be L. A moving body-fixed coordinate system is posi- 6 Ku_ 2 sin u 7 60 07  
6 7 6 7 tr
tioned along the x-axis from the axle of the robot at the _
H(q, q)= 6 0 7 , B(q) = 6 0 0 7 , t= ,
6 7 6 7 tl
center of mass (COM), and it is assumed that it has an 4 0 5 4 1 0 5
offset of distance d. Also, let mP represents the mass of 0 0 1
2 3 2 3
the WMR without the driving wheels, the motors’ rotor  sin u cos u d 0 0 l1
and gearbox rotor; mw be the mass of each driving A(q)= 4  cos u sin u b=2 r 0 5 and l = 4 l2 5.
wheel, and the corresponding motor-rotor, gearbox-  cos u sin u b=2 0 r l3
rotor; IP is the moment of inertia (MOI) of the platform Here, t = ½ t r t l  2 <2 is the vector of generalized
about a vertical axis passing through COM; Iw repre- forces, M 2 <5 3 5 is the mass matrix, H 2 <5 3 1 repre-
sents the equivalent wheel MOI with reference to wheel sents the Coriolis and centripetal matrix, B 2 <5 3 2 is
axis; Im is the MOI of each wheel along with the motor the control input matrix, A 2 <3 3 5 and l 2 <3 denote
and gearbox rotor which is about a wheel diameter constraint matrix and vector of constraint forces,
passing through the z-axis; l1 , l2 , l3 are the Lagrange respectively; expression of K is given in Nandy et al.32
multipliers. The position of the WMR can be specified Solving the Lagrange multipliers from equations (1),
by three generalized coordinates xc , yc , u. Here, (xc , yc ) (2) and (3) and utilizing the constraint equations (6)–(8)
is the position of the COM and u is the heading angle the following solutions are obtained
Roy et al. 181

(Kr  mrd)(€ ur  €
ul ) mr2 (u_ 2l  u_ 2r ) obtain, and determination of bound for unmodeled
l1 = + ð10Þ
b 2b dynamics is not always possible. So using system
  dynamics (15) and the time-delayed logic,23–26 H^ 1 is
O1 + u(I€  Kd) + Kbu_ 2
l2 = ð11Þ estimated from the input and measured state informa-
b
  tion of the past instant as
O2 + u(Kd
€  I) + Kbu_ 2
l3 = ð12Þ ^ 1 ðq(t), q(t),
b H _ q(t)Þ ffi H1 ðq(t  h),
€
ð17Þ
_  h), € q(t  h)Þ = u(t  h)  M ^ ðq(t  h)Þ€
q(t  h)
where O1 = mbr(€ur + €ul )=2 + (rmd2  rKd)(€ur  €ul )= q(t
2b  mbdu_ 2 + r2 (u_ 2r  u_ 2l )(md  K)=4b and O2 = where h is the delay time and generally selected as the
mbr(€ ur + € ul )=4 + (rmd2  rKd)(€ul  €ur )=b  mb2du_ 2  sampling time.23 The advantage of TDC is that only
2 _2 _ 2
r (ur  ul )(md  K)=2b. nominal knowledge of the mass matrix would suffice
Subtracting equation (8) from equation (7) yields the controller design purpose, which significantly
r(u_ r  u_ l ) reduces tedious modeling effort of complex robotic sys-
u_ = ð13Þ tems. However, this approximation method would
b
exactly hold when sampling frequency is very high or
Substituting equations (10)–(12) into equations (1)– h ! 0. Since sampling frequency is set based on an
(3) and putting equation (13) into equations (4) and (5), assumed computation delay, sensor response time and
the new system dynamics is written as follows so on, an infinitesimally small sampling time is not rea-
lizable in practice. Intuitively, this process can be
M(q)€ _ =u
q + H(q, q) ð14Þ
thought to be analogous to approximating any
2 3 curve using a collection of infinitesimally small
m 0 K sin u k1 k2
6 0 m K cos u k3 k4 7 straight lines. However, as the length of straight lines
6 7
6
where M(q)= 6 K sin u K cos u k5 k5 7 increases, approximation error rises. Keeping with the
I 7,
4 k1 k3 k5 Iw 0 5 same analogy, an approximation error, defined as
k2 k4 k5 0 Iw _ €q(t))  H1 (q(t  h), q(t
m(t)=H1 (q(t), q(t), _  h), €q(t  h)),
2 3 would always persist in TDC. Let qd (t) be the desired
mdu_ 2 cos u+mr2 sin u(u_ 2r  u_ 2l )=2b
6 7 position trajectory to be tracked and e1 = qd  q repre-
6 mdu_ 2 sin umr2 cos u(u_ 2r  u_ 2l )=2b 7 sents the position tracking error. With this consider-
6 7
_
H(q, q)= 6 Kr2 (u_ 2r  u_ 2l )=2b 7, k1 = ation, the auxiliary control input is defined as
6 7
6 7
4 Kru_ 2 =2 5
2 ua (t) = €qd (t) + K1 e1 (t) + K2 e_ 1 (t) ð18Þ
Kru_ =2
((mdrKr)=b sin umr=2 cos u), k2 =((Krmdr)= where K1 2 <5 3 5 and K2 2 <5 3 5 are two positive defi-
b sin umr=2 cos u), k3 =((Krmdr)=b cos umr= nite matrices. Putting equations (16) and (18) into equa-
2 sin u), k4 =((mdrKr)=b cos umr=2 sin u), tion (15), the error dynamics is obtained as
k5 =r(IKd)=b and u=Bt.
1
^ m
€e1 + K1 e1 + K2 e_ 1 = M ð19Þ
Controller design Again, error dynamics (19) can be expressed in state-
space format as
Background: TDC
The system dynamics (14) is further modified in a more e_ = Ae + Bd ð20Þ
compact form as follows    
T 0 I 0
where e = ½ eT1 e_ T1  , A = ,B=
^ q + (M  M)€
M€ ^ q+H=u K1 K2 I
ð15Þ and d = M ^ 1 m. Here, d represents the TDE. K1 and K2
^ q + H1 = u
M€
must be judiciously selected such that A is Hurwitz sta-
where H1 (q, q, ^
_ €q) = (M(q)  M(q))€ _ and M
q + H(q, q) ^ is ble. TDE remains bounded if M ^ is selected in a manner
the nominal value of M: The control input u is defined such that the condition jjM1 M ^  Ijj \ 1 is satis-
to be fied.23–26
The perturbation in M is primarily caused by
^ a +H
^1 payload variation, and it is always possible to have a
u = Mu ð16Þ
prior idea regarding the maximum allowable payload
where H^ 1 is the nominal value of H1 and ua represents for a WMR. Hence, one can always find a M ^ such that
an auxiliary control input to be defined later. the condition is satisfied.
Unmodeled dynamics and disturbances can be consid-
ered to be subsumed by H1 . The knowledge of M ^ and
^
H1 are required for the evaluation of nominal control Motivation
for the robust controllers such as ROC and SMC. The stability analysis for TDC, as carried out in Hsia
However, nominal model of uncertainties is difficult to and Gao,23 Shin and Kim24 and Roy et al.,25,26 does
182 Proc IMechE Part I: J Systems and Control Engineering 231(3)

not reflect the effect of TDE on the system stability as possible since its selection depends on system hardware
well as controller performance explicitly. However, to configuration and their responses.
the best knowledge of the authors, this has been the
only mean of stability analysis of TDC reported in the
literature. A new stability analysis of TDC is presented TDROC
in this article in the sense of uniformly ultimately The absence of any term to negotiate TDE in TDC
bounded (UUB). The proposed stability notion estab- results in sustained approximation error and reduces
lishes a direct relation between the tracking error and tracking accuracy as can be seen from equation (19) and
TDE; thus, it reflects the effect of TDE on system sta- further substantiated by equation (21). Furthermore,
bility and the controller performance which provides the TDC-based control29 requires predefined bound of
an important insight to the performance of TDC which TDE. So the aim is to envisage a control strategy that
was previously unaddressed in the literature. would provide robustness against the uncertainties with
minimal modeling effort and eliminate the approxima-
Theorem 1. The system given by equation (15) employ- tion errors to the best extent possible to improve track-
ing the control input (16) and having the approximation ing accuracy without any prior knowledge of the bound
logic (17) is UUB. of TDE. Toward achieving this goal, TDROC is con-
ceptualized, which amalgamates TDC as well as switch-
ing logic. The structure of the proposed control law is
Proof. Let the Lyapunov function candidate be selected as equations (16) and (17); however, the auxili-
V(e) = eT Pe, where P is the solution of the Lyapunov ary control input ua is designed as
equation AT P + PA =  Q with Q . 0. Using equa-
tion (20), time derivative of V yields ua = ^ua + Dua ð22Þ
_ = e_ T Pe + eT P_e
V(e) where ^ua is the nominal control input and it is selected
T T
= e (A P + PA)e + 2e PBd T the same as equation (18); Dua is the switching control
input and it is responsible to counter the TDE d: Let
=  eT Qe + 2eT PBd
1 1
^ ^
ł  lmin (Q)kek2 + 2kekkPBkkdk kdk = M m ł M :kmk :¼ k ð23Þ
=  kekflmin (Q)kek  2kPBkkdkg
which is possible as TDE is bounded. Following Roy
As mentioned earlier, the TDE (d) remains bounded et al.,29 Dua can be designed as
if jjM1 M ^  Ijj \ 1. So V_ \ 0 will be established if
8 s
jjejj . (2jjPBjj jjdjj)=lmin (Q) :¼ -1 . So the system >
< k(e, t) for ksk ø e
would be UUB with the following ultimate bound25 ks k
Dua = ð24Þ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : k(e, t) s for ksk \ e
>
 ffi e
lmax (P)-21
ke k ł ð21Þ
lmin (P) where k is the switching gain, e . 0 is a small scalar to
avoid chattering and s = BT Pe. Determination of
where lmin (  ) and lmax (  ) denote the minimum and switching gain k using equations (23) and (24) requires
maximum eigenvalues of (  ), respectively, P and jj  jj predefined bound of TDE. Conventional robust con-
denotes the Euclidean norm of (  ). Let denotes the trollers such as ROC and SMC also require predefined
smallest level surface of V containing the ball B-1 with bound on uncertainties. However, bound estimation
radius P -1 centered at e = 0. For initial P time t0 , if requires uncertainty modeling which is not always pos-
e(t0 ) 2 P then the solution remains in 8tPø t0 . If sible in practical circumstances. So the switching con-
e(t0 ) 62 , then V decreases as long as e(t) 62 P. The trol law for the proposed TDROC is modified as below
time requiredP to reach -1 is zero when e(t0 ) 2 , and
8 s
while e(t0 ) 62 the finite time tr1 required to reach the >
< k(e, t) for ksk ø e
bound is given by Khalil33 ks k
Dua = ð25Þ
>
:  s
ðVðe(t0 )Þ  V(-1 )Þ k(e, t) for ksk \ e
tr1  t0 ł e
c0
where
_ ł  c0 .
where V(e) h
P
n P 
^ 1 jj n ni
 = jjM
k ani (O(i)  O(i + 1)) = a ;
i=1 i=1
Remark 1. Through this stability analysis, it can be ^
O(i) = u(t  ih)  M(q(t  ih))€q(t  ih); 0 \ a ł 1 is a
inferred that -1 is proportional to the TDE d. Thus, discount factor and n . 1 indicates the number of pre-
tracking accuracy of TDC gets deteriorated as the ceding input and measured state information (or length
approximation error TDE increases. One can reduce h of the window for past data) to determine the switching
to reduce TDE, but arbitrary reduction in h is not gain k
.
Roy et al. 183

Remark 2. It can be noticed from equation (23) that Case (2): when jjsjj \ e.
^ 1 . So the designer needs to select
TDE (d) involves M
the nominal parameters in a way such that M^ 1 always The term 2sT d can be written as
exists.
sT s
2sT d ł 2kskkdk ł 2kksk = 2k ð28Þ
ks k
Remark 3. According to equation (25), switching gain k 
attempts to emulate the approximation error dynami- Again, using equations (25) and (28), time derivative
cally from the measured states and control input of past of V yields
instances. Thus, TDROC removes any prerequisite of 
uncertainty bound estimation and, further, removes the _ ł  eT Qe + 2sT Dua + k s
V(e)
burden of uncertainty modeling. The choice of a = 1 ks k
 ð29Þ
would ensure equal weightage to all the previous data 2 T s s
ł  lmin (Q)kek + 2s  k +k
while selection of a closer to 1 would emphasize on the e ks k
recent data than that of the past instances. The value of
n is determined according to the available computa- The second term of equation (29) takes the maxi-
tional capacity. mum value (ek2 )=(2
k) when jjsjj = (ek)=(2
k). Therefore
2
_ ł lmin (Q)kek2 + ek
V(e)
Stability of TDROC 2
k
The stability analysis of TDROC is carried out in terms Again, _ \ 0 would be established if
V(e)
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
of UUB and is stated in the following theorem. jjejj . (ek2 )=(2 klmin (Q)) :¼ -3 . Therefore, the sys-
tem is UUB with the following ultimate bound33
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
Theorem 2. The system given by equation (15) employ-
lmax (P)-23
ing control input (16) with auxiliary control input (22) ke k ł ð30Þ
and having the switching law represented by equation lmin (P)
(25) is UUB. Similarly, the time to reach -3 can found as
tr3  t0 ł (V(e(t0 ))  V(-3 ))=c0 . h
Proof. Using the Lyapunov function candidate
V(e) = eT Pe, the stability aspect of TDROC is investi- Remark 4. It can easily be comprehended from equa-
gated for different cases as follows: tions (21), (27) and (30) that -3 \ -2 \ -1 , and thus
the accuracy that can be obtained from TDROC is bet-
Case (1): when jjsjj ø e. ter than TDC. Moreover, in practice, e is selected very
small, and thus -3 can be further reduced by proper
Utilizing equations (20) and (25), time derivative of V tuning of e and Q.
gives
_ = e_ T Pe + eT P_e
V(e) Experimental results and comparison
T T T
= e (A P + PA)e + 2e PB(Dua + d) The true essence of the proposed hybrid robust control-

T T s ler can only be realized in the real-world environment
=  e Qe + 2s  k +d
ksk as unmodeled dynamics and practical uncertainties
cannot be appropriately simulated. So as a proof of
_ yields
Further simplification of V(e) concept for the proposed control law, experimentation
_ ł  eT Qe  2 is carried out with Drishti robotic platform (Figure
V(e) kksk + 2kskk
2(a)).29 The experimental architecture of WMR is
ł  lmin (Q)kek2 + 2(k  k
 )ksk ð26Þ shown in Figure 2(b). The hardware parameters for the
  robotic platforms are selected as m = 36 kg,
ł  kek lmin (Q)kek  2(k  k ) BT P
d = 0:12 m, r = 0:096 m, b = 0:414 m, I = 5:72 kg m2 ,
So _ \0
V(e) would be established if Iw = 0:07 kg m2 , K = 0:34 kg m.32 The communication
jjejj . (2(k  k  )jjBT Pjj)=lmin (Q) :¼ -2 . Thus, the sys- between the server and the client of Drishti is main-
tem would be UUB with the following ultimate bound33 tained through a ‘‘D-link’’ USB modem. The sampling
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi time h is judiciously assumed to be 100 ms considering
 ffi
lmax (P)-22 communication delay between the robot and the con-
ke k ł ð27Þ trol station, the associated sensor suite and hardware.
lmin (P)
The wheel encoders (incremental quadrature-type enco-
The time to reach -2 can found as der with 500 ppr) are used during the experiments to
tr2  t0 ł (V(e(t0 ))  V(-2 ))=c0 .33 get the wheel position and velocity. Furthermore, upon
184 Proc IMechE Part I: J Systems and Control Engineering 231(3)

Figure 2. (a) Drishti wheeled mobile robot and (b) experimental architecture of WMR-Drishti.

obtaining the wheel velocity u_ r , u_ l , the other state vari-

ables are obtained by integrating the following kine-
matic relation32
2 3
x_ c
6 7
4 y_ c 5
u_
2r r 3
(b cos u  d sin u) (b cos u + d sin u)
6b b 7
6 7
= 6 r (b sin u + d cos u) r (b sin u  d cos u) 7
4b b 5
(r=b) (r=b)
" #
_ur
ð31Þ
u_ l
Figure 3. System mass variation.
Upon obtaining xc , yc from equation (31), the con-
trol/performance deviation of an individual controller
is evaluated by computing the position error achieve various COMs and variations in inertia at vari-
(xdc  xc , ydc  yc ). The desired positions, that is, xdc , ydc , ous times. Two different paths, circular and lawn mower
are defined subsequently for individual trajectories. paths, are selected for the WMR to track.
Furthermore, the above-mentioned choice of system
parameters renders the invertibility of M. ^ To verify the
Circular path
performance of the proposed TDROC, it is felt prudent
to compare the performance of TDROC with The trajectory definition for circular trajectory with
ASMC,27 TDC26 and ROC7 employing them in radius R1 = 1:5 m and angular velocity u_ d = 0:25 rad=s
Drishti. The controller gains are chosen as is defined as
K1 = K2 = diagf 4 4 4 4 4 g to achieve critically
xdc = R1 cos (u_ d t + u0 ),
damped error convergence rate. The other control
parameters for TDROC are selected as ydc = R1 + R1 sin (u_ d t + u0 ),
Q = diagf 1 1 1 1 1 g, a = 0:9, n = 4, e = 0:1. u_ d (R1 + b=2)
Similar values of K1 , K2 and Q have been used for ud = u0 + u_ d t, udr = ,
r
ROC and TDC as well for fair comparison.
u_ d (R1  b=2)
Furthermore, the sliding surface required for ASMC is udl =
defined to be the same as ‘‘s’’ (as defined in equation r
(24)) for parity. ROC needs a predefined bound on the The initial position is selected as
perturbation and these bounds are selected to be 30% ½xc0 yc0 u0 T = ½1:5 1:5 0T . Figure 4 depicts the circular
above the nominal values of the hardware parameters of path tracked by the vehicle when four different control-
Drishti defined earlier. The robot carries a 6-kg laser ran- lers are employed.
gefinder (LRF) as extra payload. To create a time- The corresponding positional error plots, in terms of
varying payload, a further 3 kg load was added and positional percentage error, are provided in Figures 5
removed periodically on the WMR platform as shown in and 6, respectively. The percentage errors are calcu-
Figure 3. The variable payloads are kept and removed lated with respect to the diameter of the circular path.
around periodically as depicted in Figure 3. However, Due to the absence of any robustness property against
the additional load was placed rather randomly to the approximation error, TDC lacks tracking accuracy
Roy et al. 185

Figure 7. Right wheel switching gain comparison for circular

path.
Figure 4. Circular path tracking comparison.

Figure 8. Left wheel switching gain comparison for circular

Figure 5. xc position error comparison. path.

ROC susceptible to unmodeled dynamics. TDROC

abridges this gap between ROC and TDC and provides
superior tracking accuracy among the four different
control strategies under consideration. While ASMC
also removes the requirement of modeling of the uncer-
tainties, its adaptive switching gain is evaluated by inte-
grating the absolute value of a predefined error
function. Therefore, the switching gain becomes a
monotonically increasing function of error. To demon-
strate this fact, the right and left wheel switching gain
of ASMC, ROC and TDROC are provided through
Figures 7 and 8 for circular path. This very nature of
the switching gain of ASMC not only reduces its track-
ing accuracy but also makes it vulnerable to chattering
due to very high switching gain. Apart from reducing
Figure 6. yc position error comparison. the modeling effort and approximating the unknown
dynamics, the embedded time-delayed logic of TDROC
over TDROC in spite of having the capability to nego- helps it attain the least switching gain among TDROC,
tiate the unknown dynamics. On the other front, exclu- ASMC and ROC. Low switching gain of TDROC fur-
sion of friction, slip and so on in system modeling as ther augments its tracking supremacy over the other
well as in computation of the switching gain makes three controllers.
186 Proc IMechE Part I: J Systems and Control Engineering 231(3)

Figure 9. Desired lawn mower path. Figure 10. Desired wheel velocities.

Lawn mower path

The next path selected for study is a lawn mower path
which is very useful for survey applications. This path
can be thought to be a mixture of a square wave and a
sinusoid wave. But square wave is discontinuous at the
turns and sinusoid wave gets narrower at the peaks. So
an approximated lawn mower path (Figure 9) with the
following trajectory definition is selected

xdc = 0:1t + 0:3 sin (0:4t),

ydc = 1:2 + 1:2 cos (0:2t  T)

The wheel velocities are chosen to be square wave of

the same magnitude and the left wheel velocity is half of
the time period delayed from the right wheel. As a per-
fect square signal is discontinuous in nature, they are
approximated applying Fourier series as Figure 11. Lawn mower path tracking performance
 X  pp   comparison.
‘
sin 2 cos 2pp
T t
_ur = (a + b) + 2 (a  b) ,
2 p p = 1, 3
p over this type of path thus becomes a challenge when
 X ‘
 pp 2pp  the system is simultaneously subjected to time-varying
(a + b) (a  b) sin cos T t
u_ l = 2 2 payload like in Figure 3. The trajectory tracked by the
2 p p = 1, 3
p WMR onboard with different controllers against the
predefined lawn mower path is demonstrated in
Here, a and b signify the maximum and minimum value Figure 11. The percentage path error (Euclidean error
of the signal, respectively, and T is the time period of distance in the Cartesian coordinates) incurred by the
the signal. a = 3, b = 1.6 and T = 25 s are selected for vehicle while following the desired path is illustrated in
the lawn mower path. The desired velocity profile is Figure 12. For percentage calculation, the maximum
depicted through Figure 10. Trajectories for other gen- path was considered that comprises the individual max-
eralized coordinates are chosen as imum position information in xdc and ydc . It can be
ð ð ð noticed that starting from the second loop of the lawn
ud = c (u_ r  u_ l )dt, udr = u_ r dt, udl = u_ l dt mower path, ASMC starts lacking its tracking accuracy
due to its ever increasing switching gain and it is sub-
stantiated through the switching gain plots between
The initial condition is taken as
ASMC, ROC and TDROC illustrated in Figures 13
qd0 = ½ 0 0 0 0 0 T . Unlike the circular path,
and 14 for the right and left wheels, respectively.
where the rate of changing the heading of the vehicle is
constant, periodic switching in the heading of the vehi-
cle (clockwise to anticlockwise and vice versa) in lawn
Conclusion
mower path intensifies the impact of the unmodeled
dynamics such as friction, slip and skid on the vehicle In this article, a new stability analysis for TDC is pro-
maneuverability. To control the movement of WMR posed which provides an analytical framework in
Roy et al. 187

tracking problem of WMR. In TDROC, the switching

logic and the time-delayed logic are judiciously fused
to obtain better tracking accuracy than conventional
TDC and switching control laws. The embedded
time-delayed logic of TDROC takes care of the
unknown dynamics and parametric uncertainties by
approximating them using past state and control
information while the inbuilt switching control logic
provides robustness against the approximation error
without any prior information of the uncertainty
bound. It needs to be mentioned here that it is possi-
ble to improve the control performance further,
through incorporation of the high-end processors
leading to increased sampling frequency which
enables TDROC render better approximation of the
Figure 12. Path error comparison for various controllers.
unknown dynamics and thereby further reduction in
the approximation error.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest
with respect to the research, authorship and/or publica-
tion of this article.

Funding
The author(s) received no financial support for the
research, authorship and/or publication of this article.

References
1. De Wit CC and Khennouf H. Quasi continuous stabiliz-
Figure 13. Right wheel switching gain comparison for lawn ing controllers for nonholonomic systems: design and
mower path. robustness consideration. In: Proceedings of the 3rd Eur-
opean control conference, Rome, 5–8 September 1995,
pp.2630–2635: IEEE.
2. Guldner J and Utkin VI. Stabilization of non-holonomic
mobile robots using Lyapunov functions for navigation
and sliding mode control. In: Proceedings of the 33rd
IEEE conference on decision and control, Lake Buena
Vista, FL, 14–16 December 1994, pp.2967–2972. New
York: IEEE.
3. Jiang ZP and Pomet B. Combining backstepping and
time-varying techniques for a new set of adaptive con-
trollers. In: Proceedings of the 33rd IEEE conference on
decision and control, Lake Buena Vista, FL, 14–16
December 1994, pp.2207–2212. New York: IEEE.
4. Hamel T and Meizel D. Robust control laws for wheeled
mobile robots. Int J Syst Sci 1996; 27(8): 695–704.
5. Aguilar LE, Hamel T and Soueres P. Robust path fol-
lowing control for wheeled robots via sliding mode tech-
niques. In: Proceedings of the 1997 IEEE/RSJ
Figure 14. Left wheel switching gain comparison for lawn international conference on intelligent robots and systems,
mower path. Grenoble, 11 September 1997, pp.1389–1395. New York:
IEEE.
6. Chawa D. Sliding mode tracking control of nonholo-
nomic wheeled mobile robots in polar coordinates. IEEE
realizing the effect of the approximation error on the T Contr Syst T 2004; 12(4): 637–644.
controller performance. Furthermore, a novel hybrid 7. Ryu J-C and Agarwal SK. Differential flatness-based
robust control strategy, TDROC, has been proposed robust control of mobile robots in the presence of slip.
and experimentally validated for the efficient path Int J Robot Res 2011; 30(4): 463–475.
188 Proc IMechE Part I: J Systems and Control Engineering 231(3)

8. Shi S, Yu X and Khoo S. Robust finite-time tracking 22. Shojaei K. Saturated output feedback control of uncer-
control of nonholonomic mobile robots without velocity tain nonholonomic wheeled mobile robots. Robotica
measurements. Int J Control 2016; 89(2): 411–423. 2015; 33: 87–115.
9. Bessas A, Benalia A and Boudjema F. Integral sliding 23. Hsia TC and Gao LS. Robot manipulator control using
mode control for trajectory tracking of wheeled mobile decentralized linear time-invariant time-delayed joint con-
robot in presence of uncertainties. J Control Sci Eng trollers. In: Proceedings of the IEEE international confer-
2016; 2016: 7915375. DOI: 10.1155/2016/7915375. ence on robotics and automation, Cincinnati, OH, 13–18
10. Onat A and Ozkan M. Dynamic adaptive trajectory May 1990, pp.2070–2075. New York: IEEE.
tracking control of nonholonomic mobile robots using 24. Shin Y-H and Kim K-J. Performance enhancement of
multiple models approach. Adv Robotics 2015; 29(14): pneumatic vibration isolation tables in low frequency
913–928. range by time delay control. J Sound Vib 2009; 321:
11. Fierro R and Lewis FL. Robust practical point stabiliza- 537–553.
tion of a nonholonomic mobile robot using neural net- 25. Roy S, Nandy S, Shome SN, et al. Robust position con-
works. J Intell Robot Syst 1997; 20: 295–317. trol of an autonomous underwater vehicle: a comparative
12. Yang J-M and Kim J-H. Sliding mode control for trajec- study. In: Proceedings of the IEEE international confer-
tory tracking of nonholonomic wheeled mobile robots. ence on automation science and engineering, Fort Worth,
IEEE T Robotic Autom 1999; 15: 578–587. TX, 17–20 August 2013, pp.1002–1007. New York:
13. Corradini ML and Orlando G. Control of mobile robots IEEE.
with uncertainties in the dynamical model: a discrete time 26. Roy S, Nandy S, Ray R, et al. Robust path tracking con-
sliding mode approach with experimental results. Control trol of nonholonomic wheeled mobile robot: experimental
Eng Pract 2002; 10: 23–34. validation. Int J Control Autom 2015; 13(4): 897–905.
14. Lee JH, Lin C, Lim H, et al. Sliding mode control for 27. Chen C, Li TS, Yeh Y, et al. Design and implementation
trajectory tracking of mobile robot in the RFID sensor of an adaptive sliding-mode dynamic controller for
space. Int J Control Autom 2009; 7(3): 429–435. wheeled mobile robots. Mechatronics 2009; 19: 156–166.
15. Hung N, Im NS, Jeong S-K, et al. Design of a sliding 28. Lee H and Utkin VI. Chattering suppression methods in
mode controller for an automatic guided vehicle and its sliding mode control systems. Annu Rev Control 2007; 31:
implementation. Int J Control Autom 2010; 8(1): 81–90. 179–188.
16. Lee JK, Choi YH and Park JB. Sliding mode tracking 29. Roy S, Nandy S, Ray R, et al. Time delay sliding mode
control of mobile robots with approach angle in Carte- control of nonholonomic wheeled mobile robot: experi-
sian coordinates. Int J Control Autom 2015; 13(3): 718– mental validation. In: Proceedings of the IEEE interna-
724. tional conference on robotics and automation, Hong Kong,
17. Oriolo G, De Luca A and Vendittelli M. WMR control China, 31 May–7 June 2014, pp.2886–2892. New York:
via dynamic feedback linearization: design, implementa- IEEE.
tion, and experimental validation. IEEE T Contr Syst T 30. Campion G and Chung W. Wheeled robots. In: Sicilliano
2002; 10(6): 835–852. B and Khatib O (eds) Springer handbook on robotics. New
18. Sun S. Designing approach on trajectory-tracking control York: Springer, 2008, pp.391–410.
of mobile robot. Robot CIM: Int Manuf 2005; 21: 81–85. 31. Chakraborty N and Ghosal A. Dynamic modeling and
19. Cho HC, Fadali MS, Lee KS, et al. Adaptive position simulation of a wheeled mobile robot for traversing
and trajectory control of autonomous mobile robot sys- uneven terrain without slip. J Mech Design 2004; 127(5):
tems with random friction. IET Control Theory A 2010; 901–909.
4(12): 2733–2742. 32. Nandy S, Chakraborty G, Kumar CS, et al. A modular
20. Shojaei K, Shahari AM and Tabibian B. Design and approach to detail dynamic formulation and control of
implementation of an inverse dynamics controller for wheeled mobile robot. In: Proceedings of the IEEE inter-
uncertain nonholonomic robotic systems. J Intell Robot national conference on mechatronics and automation, Beij-
Syst 2013; 71: 65–83. ing, China, 7–10 August 2011, pp.1471–1478. New York:
21. Arab A and Fateh MM. An uncertainty compensator for IEEE.
robust control of wheeled mobile robots. Adv Robotics 33. Khalil H. Nonlinear systems, 3rd ed. Hoboken, NJ: Pre-
2015; 29(20): 1303–1313. ntice Hall, 2002.