Received August 17, 2020, accepted August 28, 2020, date of publication September 10, 2020, date of current

version September 29, 2020.

Digital Object Identifier 10.1109/ACCESS.2020.3023345

Energy Comparison of Controllers Used for a

Differential Drive Wheeled Mobile Robot
ALEXANDR STEFEK 1 , THUAN VAN PHAM 1, VACLAV KRIVANEK 2, (Member, IEEE),
AND KHAC LAM PHAM 3
1 Department of Air Defence, University of Defence, 66210 Brno, Czech Republic
2 Department of Military Robotics, University of Defence, 66210 Brno, Czech Republic
3 Department of Aircraft Technology, University of Defence, 66210 Brno, Czech Republic
Corresponding author: Thuan Van Pham ([email protected])
This work supported by the Czech Republic Ministry of Defence (University of Defence in Brno development program ‘‘Research of
sensor and control systems to achieve battlefield information superiority’’).

ABSTRACT In order to select the best controller for a Differential Drive Wheeled Mobile Robot (DDWMR),
an energy consumption comparison relating to tracking accuracy is used as a very strict criterion. Therefore,
this paper reviews some well-known controllers designed for the DDWMR. Furthermore, there are presented
several experiments with the extensible open-source code programmed in Python. Such an extensible
open-source code presentation could serve as a tool for simulating, comparing, and evaluating a set of
different control algorithms. The kinematic and dynamic models of the DDWMR and control algorithms
are implemented in this open-source code to determine a travel time, a distance between the robot’s position
and a given path, a linear velocity, an angular velocity, a travel path length, and a total kinetic energy loss
of the DDWMR. These simulation results are used to compare and evaluate the given control algorithms.
Moreover, the simulation results also enable to answer the question of whether a significant increase in
energy consumption is worth shortening the travel path by just a bit. Finally, this paper includes a direct
link to the stored experiments which are runnable and could serve as a proof. Besides, users can also easily
supplement with other controllers and different paths to evaluate robot tracking control algorithms.

INDEX TERMS Differential drive robots, energy model, fitness function, kinetic energy, robot’s energy,
wheeled mobile robot control.

I. INTRODUCTION platform and one passive castor wheel used for balance and
Robotic Science has been developing rapidly due to its var- stability.
ious useful applications in many aspects of common life, The motion control algorithm of the DDWMR: In princi-
industry, medicine, the military, and especially its ability ple, the movement of the DDWMR is based on two separately
to operate in hazardous and toxic environments. Recently, driven wheels placed on either side of the robot body. There-
the Differential Drive Wheeled Mobile Robot (DDWMR) fore, it changes its direction by changing the relative rotation
has been increasingly noticed and widely applied within the speed of the two wheels. For the DDWMR, the problems
scope of Robotic Science. It has a lot of advantages, such as of path planning and path tracking are the most important.
flexible motion capabilities, a simple structure, lower produc- However, the path tracking is more important because its
tion costs. Furthermore, it can operate independently for a accuracy directly affects the robot operability. Many authors
long time without a direct human control. have researched and published various control algorithms of
The DDWMR has some different forms, such as a 2-wheel, path tracking for the DDWMR, such as the adaptive output
a 3-wheel, or a 4-wheel type. We can find the 2-wheel type feedback control [2], the input-output feedback linearization
in [13], the 3-wheel type in [5], [7], [10], [11], [18]–[20], [23], method [4], the two-step feedback linearization control [3],
and the 4-wheel type in [15], [17]. The 3-wheel type is the the backstepping-based control [5], the PID control [6],
most popular form which comprises of two fixed powered the Lyapunov function-based control [8], [9], the adaptive
wheels mounted on both left and right side of the robot and sliding mode control [7], [10], [11], [18], the neural-
network-based control [19], and the robust adaptive-based
control [21]. All of the above studies refer only to continuous
The associate editor coordinating the review of this manuscript and or smooth curves, whereas the energy consumption of the
approving it for publication was Yingxiang Liu . robot is not mentioned. In reality, robots often have to work in

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 8, 2020 170915
 A. Stefek et al.: Energy Comparison of Controllers Used for DDWMR

complex conditions (environments with many obstacles, slip-

ping wheels, acceleration of robots varies when they move)
over a long time. So in many cases, it is too difficult for
the robot to track the given path smoothly with saved energy
consumption.
A requirement for the DDWMR is that it has to move
smoothly, with no or minimum vibration, and detects a minor
tracking error while tracking the sharp turn path or the discon-
tinuous path. In order to achieve that requirement, the robot
needs to have a constant rotation of the two wheels or it
has to move on a path in a circular arc format. However,
it should be mentioned that the circular arc function does
not fit all possible paths, which the robot might be required
to take [12]. These problems have been studied and solved
by several different control algorithms that were presented in
some papers, such as the controllers proposed by Kanayama
and Robins Mathew [1], a feedback-based control for cir- FIGURE 1. (a) a 2-wheel differential drive mobile robot, (b) a 3-wheel
cular path [12], the Lyapunov-based method [25], a clever differential drive mobile robot, (c) a 4-wheel differential drive mobile
robot.
trigonometry-based control [24], a fuzzy logic control [24],
and a Dubins path-based control [16].
The comparison and evaluation of the given control algo- In this study, we mention only the 3-wheel DDWMR
rithms can be based on many different criteria, such as the type because it is widely used in many applications, such
distance between the robot’s position and the given path (the as industrial automated guided vehicles and service robots.
cross-track error), the travel distance (the length of the path), Furthermore, the robots of this type are controlled more easily
the total energy loss, the travel time, etc. For the mobile robot than others and provide high maneuverability and rotation
motion control, accuracy and energy consumption are always around the center of the robot [11]. Besides, the 3-wheel
strict criteria. So it is necessary to have a simulator to verify DDWMR has a simple structure and is suitable for many
and compare these criteria when the different controllers are practical applications as well as being popular and prevailing
used. Therefore, this paper aims at simulating, and comparing in the market.
the results of some well-known control algorithms for the 3-
wheel DDWMR to track the sharp turn path or the discon- B. MODELS OF A 3-WHEEL DIFFERENTIAL DRIVE MOBILE
tinuous path in the same working conditions. Then the best ROBOT TYPE
controller will be designed to improve it so that the robot suits 1) KINEMATIC MODEL
each working condition and specific working environment.
The Differential Drive Wheeled Mobile Robots (DDWMRs)
This paper is organized as follows: some different shapes,
usually have two independently driven wheels and one or
a kinematic model, a dynamic model, and an energy model
more unpowered wheels at the rear as a balance. An important
of the DDWMR are introduced in Section II; a review of
issue of the differential driving of mobile robots which needs
some well-known control algorithms used for the DDWMR is
considering is that their motion controller design is mostly
provided in Section III; simulation, result collection, compar-
based on kinematic models. The main reason is that dynamic
ison, and the used control algorithms evaluation in Section IV.
models are more complex than kinematic models and mobile
A summary and a conclusion complete the paper in Section V.
robots usually use only the low speed of the motor to control
the loop [1].
II. DIFFERENTIAL DRIVE WHEELED MOBILE ROBOTS
The geometry and kinematic parameters of this robot are
Wheeled mobile robots are employed for inventory manage-
shown in Fig. 2.
ment, factory automation, military surveillance, etc. These
In Fig. 2, this robot is the DDWMR which has important
robots are classified into different categories, such as car-like
parameters as below:
robots, omnidirectional robots, and differential drive robots;
of which differential drive robots are a prominent class [1]. v is the linear velocity of the DDWMR (ms−1 ),
θ is the orientation of the DDWMR (rad),
A. POPULAR SHAPES OF THE DIFFERENTIAL DRIVE ωr is the angular velocity of the right wheel (rads−1 ),
WHEELED MOBILE ROBOT ωl is the angular velocity of the left wheel (rads−1 ),
Currently, the structure of the DDWMR has many different vr is the linear velocity of the right wheel (ms−1 ),
types, such as a 2-wheel, a 3-wheel, or a 4-wheel differential vl is the linear velocity of the left wheel (ms−1 ),
drive mobile robot, as shown in Fig. 1. Each one has different r is the radius of the right and the left wheels (m),
advantages but the 3-wheel type has more advantages than the b is the distance between the right and the left wheels
others. (m),

170916 VOLUME 8, 2020

A. Stefek et al.: Energy Comparison of Controllers Used for a DDWMR

2) MOTOR DYNAMICS
A dynamic model of motor behavior has to be modelled to
consider its characteristic on the robot chassis [12].
The DC motors are usually used for the DDWMR. The
motor dynamics is modelled as follows [30]:
The electrical circuit and the free body diagram of the
motor rotor of a DC motor are depicted in Fig.3.

FIGURE 3. A basic electrical circuit of a DC motor and a free body

diagram of the motor rotor.

FIGURE 2. A geometry of a 3-wheel differential drive mobile robot.

An equation governing the mechanical dynamics of the
Q is the center of the axis between the right and the motor is
left wheels, J θ̈ + Bθ̇ = Kt ia (10)
G is the center of gravity of the DDWMR,
a is the distance between Q and G (m). The armature current has its own dynamics. From Fig. 3,
we can write
The kinematic model equations depend on the geometrical dia
structure of the DDWMR [3]. However, most of the 3-wheel La + Ra ia = V − Ke θ̇ (11)
dt
DDWMRs have the same kinematic equation which is con-
structed as follows [3], [5], [12], [13], [15]: where J is the moment of inertia, θ is the angle of rotation of
the output shaft, T = Kt ia is the mechanical torque developed
xG = xQ − a cos θ (1) in the motor rotor, Kt is the armature constant, ia is the
yG = yQ − a sin θ (2) armature current, B is the damping coefficient, V is the input
voltage, Ke is the motor constant, Ra stands for the armature
We assume that [3], [13]: resistance, and La is the armature inductance.
- The wheels are rolling without slipping, The Laplace transform of (10) and (11) are given as:
- The center of gravity G coincides with the point Q,
- The guidance axis is perpendicular to the robot plane. Js2 θ (s) + Bsθ (s) = Kt Ia (s) (12)
Based on Fig. 2 we get: La sIa (s) + Ra Ia (s) = V (s) − Ke sθ (s) (13)
b
vr = v + θ̇ (3) From (12) and (13) we get
2
b V (s) − Ke sθ (s)
vl = v − θ̇ (4) Js2 θ (s) + Bsθ (s) = Kt (14)
2 Ra + sLa
Adding and subtracting (3) and (4) we get: The transfer function of the angular velocity ω(s) to the
1 input voltage V (s) is
v = (vr + vl ) (5)
2 ω (s) Kt
1 F (s) = = (15)
θ̇ = (vr − vl ) (6) V (s) (Ra + La s) (Js + B) + Kt Ke
b Two identical motors are used for one robot chassis, so only
Due to the non-slipping assumptions we have vr = rωr one transfer function of the motor will be defined.
and vl = rωl . Kt
From Fig. 2 we get: ω (s) La J
F (s) = = (16)
V (s) s2 + Ra J +La Ra B+Kt Ke
ẋ = ẋQ = v cos θ (7) La J s + La J

ẏ = ẏQ = v sin θ (8) Or:

θ̇ = ω (9) F (s) =
b1
s2 + a0 s + a1
Equations (7), (8) and (9) are a kinematic model of the
Kt Ra J +La Ra B+Kt Ke
3-wheel DDWMR. where b1 = La J ; a0 = La J ; a1 = La J

VOLUME 8, 2020 170917

 A. Stefek et al.: Energy Comparison of Controllers Used for DDWMR

C. FITNESS FUNCTION In this paper, (19) and (20) are used as the fitness functions
There are many kinds of fitness functions for a mobile robot, to calculate energy consumption and travel distance that are
but this paper only presents behavioral fitness functions used to compare and evaluate the controllers based on the
for the 3-wheel Differential Drive Wheeled Mobile Robots. energy consumption in relation to the travel distance of the
In this case, we consider the behavior of the robots in relation robot.
to their energy consumption and their travel distance because
the amount of energy consumption and the travel distance III. WELL-KNOWN MOTION CONTROL ALGORITHMS
affects the uptime and ability to perform their tasks. Motion is an important activity of the Differential Drive
Wheeled Mobile Robot. Controlling the DDWMR movement
1) ENERGY MODEL OF DIFFERENTIAL DRIVE WHEELED depends on the structure of the robot, on the task that it
MOBILE ROBOT will perform, and on the environment in which it will move.
Overall Energy Model of the Differential Drive Wheeled Therefore, choosing suitable algorithms for robot motion
Mobile Robot: After analyzing all lost components, the com- control depends on each specific case.
plete energy model equation is presented as [22]: Currently, there are a lot of different motion control meth-
ods for the DDWMR, such as problems of position (posture)
Ebattery = Edc + Ek + Efriction + Eelect (17)
tracking, trajectory (path) tracking, point to point tracking
where Edc is the energy loss for DC motor, Ek is the kinetic and leader following. Among them, the path tracking is of
energy losses, Efriction is the energy losses due to friction, more concern. In the trajectory tracking problems, a reference
Eelect is the energy losses in the electronics, Ebattery is the point (the midpoint between the left and right wheel) on the
energy of battery used for the robot. robot must follow a desired trajectory in the Cartesian space
In this study, we only consider the kinetic energy losses of starting from a given initial configuration. Some prevalent
the robot and ignore the other components. algorithms of the DDWMR path tracking are the adaptive out-
The kinetic energy of the robot at any time can be expressed put feedback control algorithm, the input-output feedback lin-
as [29]: earization method, the backstepping control method, the PID
1 1 control, the Lyapunov-based control, the sliding mode-based
Ek = m (v(t))2 + I (ω(t))2 (18) control, the robust control method, the adaptive fuzzy control,
2 2
the neural-network control, and the vision-based methods,
From (18), the total kinetic energy loss from the initial time
etc.
to the final time is
Point to point motion control or waypoint tracking of
tn  
X X 1 1 a robot: the robot must move to a desired goal configu-
Ek = m (v(t))2 + I (ω(t))2 ration from a given initial configuration. As mentioned in
t0
2 2
z   the introduction of this paper, the DDWMR needs to move
X 1 1 smoothly and minimize error tracking even when it tracks
= m (v(t0 + i1T )) + I (ω(t0 + i1T ))
2 2
2 2 the sharp turn path or the discontinuous path. Some well-
i=0
(19) known point to point motion controllers have been proposed,
such as the feedback-based controller for circular path, the
where t, t0 , tn represent time, initial time, and final time Dubins path-based controller, the Lyapunov-based controller,
respectively, m and I denote the mass and the moment of the clever trigonometry-based controller, the advance con-
inertia of the robot, i = (0, 1, 2, 3, ....,z = (tn − t0 )/1T ),1T troller by Robins Mathew, and by the Kanayama’s controller,
is step time. etc.

2) TRAVEL DISTANCE A. FEEDBACK-BASED CONTROLLER FOR CIRCULAR PATH

The robot travel distance is an important quantity used for According to [12] and kinematic model (7), (8), (9) we get:
evaluating the quality of moving along the track of the given
path. The less the difference between the travel distance ẋ = ẋQ = v cos θ
and the length of the given path is, the less the average ẏ = ẏQ = v sin θ
of cross-track error is. In this study, we expect the average vr − vl 21
cross-track error to be as small as possible. So, the travel θ̇ = ω = =
distance is as similar to the length of the given path as b b
possible. The function of the travel distance is where v = vr +v vr −vl
2 ;1 =
l
2 , b is the distance between the
Ztn right and the left wheel.
S= |v(t)|dt (20) The DDWMR moves from point A to point B. Based on
feedback control, the [12] shows that:
t0

where S is the travel distance by the robot, t0 and tn represent 21 (AB)x sin θ − (AB)y cos θ
θ̇ = ω = = 2v (21)
the initial time and final time, respectively. b |AB|2
170918 VOLUME 8, 2020
A. Stefek et al.: Energy Comparison of Controllers Used for a DDWMR

In the (21)
r
−
→ 
AB = (AB)x ; (AB)y , |AB| = (AB)2x + (AB)2y

(AB)x is the projection of AB along the x-axis. (AB)y is the

projection of AB along the y-axis.
The kinematic model with the angular velocity of the
robot in (21) is a form of feedback-based controller for the
DDWMR moving on a circular path segment.
The paper [12] introduces one advanced control law by FIGURE 5. A geometry of a 3-wheel differential drive wheeled mobile
adding a coefficient Kc to (21), so we get: robot moves to a goal base on trigonometry.

21 (AB)x sin θ − (AB)y cos θ

θ̇ = ω = = Kc 2v (22)
b |AB|2 we align the robot with our target destination θ = θG (here
Coefficient Kc is greater than zero. (22) allows the running θ is the orientation of the robot, θG is the orientation of the
of multiple experiments and fine-tuning a path of a robot on goal) then X will pass over the location of X̃ shortly after X̃
a single line segment [12]. gets to next location [23].
From Fig. 5, we get:
B. THE LYAPUNOV-BASED CONTROLLER
The DDWMR moves from point Q with linear velocity v and x̃ = x + l cos θ ⇒ x̃˙ = ẋ − l sin θ (25)
orientation θ to point G, depicted in Fig. 4. ỹ = y + l sin θ ⇒ ỹ˙ = ẏ + l cos θ (26)

By substituting ẋ, ẏ in (25), (26) for ẋ, ẏ in (7), (8) and

presented in [23] we get:

v = vri cos(θG − θ) (27)

vri
ω= sin(θG − θ) (28)
l
where vri is the desired forward velocity, an expected input
for steering the robot towards a goal, l is the distance from X
to X̃ . It is considered as a coefficient of the controller.
The l needs to tune. In general, a shorter value of l makes
the robot turn quicker, but a longer l allows the robot to run
more smoothly [23].

D. ADVANCED CONTROLLER BY KANAYAMA AND ROBINS

FIGURE 4. A geometry of a 3-wheel differential drive wheeled mobile MATHEW
robot moves to a goal.
In Fig. 6, the DDWMR follows a path connecting the way-
points from a source qw0 = (xw0 , yw0 ) to a destination
Based on the Lyapunov function, the linear velocity and qwn = (xwn , ywn ) through a set of intermediate waypoints
the angular velocity of the DDWMR are calculated as fol- qwk = (xwk , ywk ), k = 1, 2, . . ., n − 1 [1]. In the paper [1],
lowing [13], [25]: Robins Mathew has mentioned the controller proposed by
v = vri cos θer (23) Kanayama, which is one of the most popular controllers in
vri cos θer sin θer (θer + k2 θG ) case of trajectories with continuous curvature. This control
ω = k1 θer + (24) equation is given in (29) and (30). Further, Robins Mathew
θer
proposes a new control method to make the DDWMR
where vri is a linear reference velocity of the robot, coeffi-
tightly track the line connecting the waypoints with a minor
cient k1 , k2 are greater than zero, θer , θ, θG = atan2(ly , lx )
cross-track error.
are the steering angle, the angle, and the robot orientation,
The control law of Kanayama is given below:
respectively.
vci = vri cos θie + k1 xie (29)
C. CLEVER TRIGONOMETRY-BASED CONTROLLER
ωci = ωri + k2 vri yie + k3 vri sin θie (30)
In this case, we need to move the DDWMR from the starting
point to the destination point, as in Fig. 5. The control law of Robins Mathew is given below:
A new point X̃ = (x, ˜ ỹ) is positioned a small distance l
directly in front of X . The point X̃ can simultaneously move vci = vri cos θie (31)
forward, perpendicular and rotate. X is close to X̃ and once ωci = ωri + k1 vri Tie + k2 vri sin θie (32)

VOLUME 8, 2020 170919

 A. Stefek et al.: Energy Comparison of Controllers Used for DDWMR

FIGURE 7. The Dubins path tracking.

FIGURE 6. The robot tracking waypoints trajectory.

TABLE 1. Input parameters of the robot.

In (29) – (32), vci , ωci , vri , ωri , are the linear, the angular
velocity, the linear reference, the angular reference velocity
respectively, (k1 , k2 , k3 ) > 0 are the control gains, (xi ,
yi , θi ) is the current robot position, and (xwk , ywk , θwk ) is
waypoint posture, (xie , yie , θie ) is the robot posture error,
where xie = (xwk − xi ),yie = (ywk − yi ),θie = (θwk −θi ),
Tie = (xi −xwk )sinφ−(yi −ywk )cosφ is the cross-track error,
φ is defined as the angle made by the line connecting current
waypoint with the previous waypoint.

E. DUBINS PATH – BASED CONTROLLER

Dubins curve consists of several circular segments and
straight segments. The shortest Dubins curve consists of three TABLE 2. Coefficients of controllers.

circular segments and straight segments. The primary forms

are: (RSL, LSL, RSR, LSR, RLR, LRL). Where L, R repre-
sent the circular turning to the left and right respectively, S
represent the tangent connecting two turning circulars [16].
The primary forms (RSL, LSL, RSR, LSR, RLR, LRL) of
the Dubins path form are presented in Fig. 7.

IV. RESULTS AND DISCUSSION

A. SIMULATING AND RESULTS COLLECTING
The input parameters of the robot used for simulation in this
study are shown in Table 1.
The coefficients of the controllers are given in Table 2. Two
different given paths are used to simulate the behavior of the
robot. The first one is a square path with parameters given in
Table 3 and the second one is a straightforward path that has a
different angle at any waypoints of the path. The parameters
of the straightforward path are given in Table 4.
The movement of a mobile robot will be directly affected
by the speed that its actuator can provide. If the provided slowly so it may affect its performance. Therefore, in this
velocity is too large, it will cause some difficulty in tracking, study, the velocity will be limited to evaluate and compare the
a high tracking error, and a high energy loss. Conversely, quality of the used controllers. Then the value of the provided
if the provided velocity is too small, the robot will move too velocity will be changed to investigate and evaluate the effect

170920 VOLUME 8, 2020

A. Stefek et al.: Energy Comparison of Controllers Used for a DDWMR

TABLE 3. Parameters of a square path.

TABLE 4. Parameters of a straightforward path.

FIGURE 9. Travel distance over time of the robot when tracking the
square path.

FIGURE 10. Kinetic energy loss over time of the robot when tracking the
square path.

FIGURE 8. The square path tracking of the robot.

of the velocity change on the motion quality and energy loss

of the robot.
This simulation creates some experiments of these are three
main experiments presented below:
The first experiment proves that using the input parame-
ters given in Tables 1-4, the linear reference velocity vri =
1.0 ms−1 , the angular reference velocity ωri = 0 rads−1 ,
the linear and the angular velocity of the controllers are lim-
ited (v is limited in the range [0.2 ms−1 to 1.0 ms−1 ], ω is also
limited in the range [−0.75 rads−1 to 0.75 rads−1 ]). Running
the simulation, we will receive some important results shown FIGURE 11. Linear velocity over time of the robot when tracking the
square path.
in Figures 8 to 17 and in Tables 5 and 6 below:
The second experiment testifies that using the Lyapunov-
based controller with the input parameters given in
Tables 1–4, the linear reference velocity vri = 1.0 ms−1 , ms−1 , v is limited in the range [0.2 ms−1 to 1.15 ms−1 ]. ω is
1.1 ms−1 , 1.15 ms−1 , the angular reference velocity ωri = 0 limited in the range [-0.75 rads−1 to 0.75 rads−1 ]). Running
rads−1 , the linear and angular velocity of the controllers are the simulation, we will receive some important results, shown
limited too. When vri = 1.0 ms−1 , v is limited in the range in Figures 18 and 19 and Tables 7 and 8 below:
[0.2 ms−1 to 1.0 ms−1 ]. When vri = 1.1 ms−1 , v is limited Some important results of the third experiment are shown
in the range [0.2 ms−1 to 1.1 ms−1 ], and when vri = 1.15 in Figure 20 and Table 9 below:

VOLUME 8, 2020 170921

 A. Stefek et al.: Energy Comparison of Controllers Used for DDWMR

FIGURE 12. Angular velocity over time of the robot when tracking the FIGURE 15. Kinetic energy loss over time of the robot when tracking the
square path. straightforward path.

FIGURE 13. The straightforward path tracking of the robot. FIGURE 16. Linear velocity over time of the robot when tracking the
straightforward path.

FIGURE 14. Travel distance over time of the robot when tracking the
straightforward path. FIGURE 17. Angular velocity over time of the robot when tracking the
straightforward path.

The third experiment above proves that using input param-

eters given in Tables 1, 3, 4, and adding some more waypoints The extra experiment 1: A diamond-shape path tracking
in each segment of the straightforward path. The linear refer- of the robot. The results of the experiment are shown in
ence velocity vri = 1.0 ms−1 , the angular reference velocity Figure 21 and Table 10.
ωri = 0 rads−1 , linear and angular velocity of the controllers The extra experiment 2: A sharp turn path tracking of
are limited (v is limited in range [0.2 ms−1 to 1.0 ms−1 ], ω is the robot. The results of the experiment are shown in
limited in range [−0.75 rads−1 to 0.75 rads−1 ]. Figure 22 and Table 11.
We can also see the results in some extra experiments as In order to compare the energy loss of other controllers,
follows: we add some more different controllers and paths to the

170922 VOLUME 8, 2020

A. Stefek et al.: Energy Comparison of Controllers Used for a DDWMR

TABLE 5. The travel distance and energy consumption of the robot when
tracking the square path.

TABLE 6. The travel distance and energy consumption of the robot when FIGURE 19. Kinetic energy loss over time of the robot controlled by
tracking the straightforward path. Lyapunov-based control to track the square path.

TABLE 7. The travel distance and energy consumption of the robot when
tracking the square path by ‘‘Lyapunov-based controller’’.

TABLE 8. The travel distance and energy consumption of the robot when
tracking the straightforward path by ‘‘Lyapunov-based controller’’.

FIGURE 18. Travel distance over time of the robot controlled by

Lyapunov-based control to track the square path.

algorithms to move in a full path, as shown in Tables 5-8 and

open-source code programmed in Python. The program will in Fig.8-10, Fig.13-15 above.
show us all of the results. In more detail, Table 5 indicates that when the robot is
controlled to track the square path based on the ‘‘Lyapunov-
B. DISCUSSION based controller’’, and the straightforward path based on the
1) THE USED CONTROL METHODS COMPARING AND ‘‘geometry-based control’’ algorithm, its travel distance is the
EVALUATING smallest of all.
From these simulation results, it can be seen that when the When it is controlled based on the ‘‘feedback-based con-
robot tracks the square path or the straightforward path, its troller for the circular path G = 4’’, the travel distance is the
travel distance and energy loss are different when the different smallest in comparison with the travel distance based on the
control algorithms are used in each segment of the given path. other control algorithms. When it is controlled by the rest
Therefore, the total travel distance and total energy loss are of the control algorithms, the travel distance is variable and
different when the robot is controlled by the different control depends on each kind of the given path.

VOLUME 8, 2020 170923

 A. Stefek et al.: Energy Comparison of Controllers Used for DDWMR

FIGURE 21. A diamond-shape path tracking of the robot.

FIGURE 20. The many waypoints tracking of the robot.

TABLE 9. The travel distance and energy consumption of the robot when
tracking the straightforward path with many Waypoints.

FIGURE 22. A sharp turn path tracking of the robot.

TABLE 10. The travel distance and energy consumption of the robot
when tracking the diamond-shape path. TABLE 11. The travel distance and energy consumption of the robot
when tracking the sharp turn path.

Besides, from Table 5, Table 6, and Fig. 8, Fig. 13 it is

obvious that the trajectory of the robot is always the furthest when it is controlled by the respective controllers shown
from all the segments of the given path when control is in Table 5, Table 6, and Fig.8, Fig.13. It is also evident that
based on the ‘‘feedback-based controller for the circular path the cross-track error in each segment is dependent on not only
G = 4’’. In other words, the average of the error-track is the the control algorithms used but also the angle created by two
biggest when the robot is controlled based on the ‘‘feedback- contiguous segments. The smaller that angle is, the bigger the
based controller for the circular path G = 4’’. The cross-track cross-track error is.
average of the robot is the smallest when control is based Therefore, if we evaluate the robot control quality based
on the ‘‘Lyapunov-based controller’’ to track the square path, on the criterion of the travel distance, the control algorithm
and the cross-track average of the robot is the smallest when of the ‘‘Lyapunov-based controller’’ is the best when the
it is controlled based on the ‘‘geometry-based controller’’ robot tracks the square path, and the ‘‘geometry-based con-
to track the straightforward path. Thus, it is very intuitive troller’’ is the best when the robot tracks the straightforward
to recognize and compare the travel distances by the robot path because the travel distance is the smallest. The control

170924 VOLUME 8, 2020

A. Stefek et al.: Energy Comparison of Controllers Used for a DDWMR

algorithm ‘‘feedback-based controller for the circular path the energy loss is in both types of the given paths. It means
G = 4’’ is the least qualified, because the travel distance is the robot will move and reach the destination faster, but its
the largest. total energy loss becomes greater.
The [23] mentioned mobile robots rely on a battery as In this case, it is clear that when the linear velocity
their power source for most of the time. However, batteries is set from 1.0 ms-1 to 1.1 ms-1 and then goes up to
have a very limited energy capacity. This finite amount of 1.15 ms-1, the travel time is reduced significantly (about 5s)
energy leads to a very short operational time of the robot, and the energy loss greatly increases (more than 20%) while
which may not be enough for missions or tasks that require the increase in the travel distance is negligible (less than
more time and energy to be completed. Therefore, despite the 0.4154%). Thus, to shorten the travel time, the robot must
intelligence and capabilities, their use in such tasks is not increase energy consumption significantly. This result has
feasible. Although the time of operation can be prolonged raised a question if the use of high velocity is so important
by increasing the number of batteries or by diverting the in case that higher energy must be provided. So, based on
robot back to the charging station, both of these techniques the specific tasks of the robot, users can decide to increase
come at the expense of the increased cost and size which can energy consumption significantly, or not, so that the robot can
cause control problems. Another way to increase the time of quickly move to its destination.
operation in robots is by reducing the energy consumption
in the robot system itself and thus increasing their energy 3) MORE WAYPOINTS ADDED INTO THE SEGMENTS
efficiency [23]. So, using less energy is considered as a OF THE GIVEN PATHS
very important criterion to evaluate the quality of control When some more waypoints are added into the segments of
algorithms for the robot. It means that the less energy the the given paths, the robot will track that path tightly. So the
robot uses, the better control algorithm is. cross-track error of the robot will be reduced significantly.
According to the energy loss criterion and based on Table 5, Figure 20 and Table 9 show that the travel distance will
6, Fig.10, Fig. 15 it can be seen that the ‘‘feedback-based con- be increased, or reduced, while the cross-track error will be
troller for circular path G = 4’’ algorithm is the best because reduced and most of the total energy loss will increase a lot.
the total energy loss is the smallest of all. The ‘‘Kanayama- This result also has raised a question, if the decrease of the
based controller’’ is the worst when the robot tracks the cross-track error is so important that higher energy must be
square path, and the ‘‘geometry-based controller’’ is the worst provided. The answer to that question depends on the specific
when the robot tracks a straightforward path because the total tasks of the robot. If the robot has to track the given path
energy loss is the most significant of all. tightly, it must increase energy consumption significantly.
In this research, the Lyapunov, Robins Mathew, geometry, In this way, the robot can reduce its cross-track error. But
and Kanayama-based controllers are called the high energy in some tasks where the robot does not have to track the
controllers. The feedback-based controller for a circular path given path tightly, it is unnecessary to increase the energy
and the Dubins path-based controller are called the low consumption to reduce its cross-track error.
energy controllers. It is also evident that when the robot is
controlled by the high energy controllers, the energy con- C. OPEN-SOURCE CODE TO IMPLEMENT EXPERIMENTS
sumption is significantly more excessive (more than 270%) This paper proposes an open-source code as a
than the energy consumption of the robot controlled by the tool from https://colab.research.google.com/drive/
low energy controllers. Meanwhile, when the robot is con- 1UP2ru4v51peFy9jO650vh8tFQTtlJCMd which was pro-
trolled by the high energy controllers the length of the travel grammed with Python language in Jupyter Notebook to
path is only slightly shorter than when it is controlled by implement and compare the different control algorithms
low energy controllers (less than 5.5207%). So it can be introduced in section 3. Users must click on this link to open
said that when the robot is controlled to track a given path, the simulation program. However, to edit it, users have to
the high energy consumption has to increase (greater than copy it to their google drive by clicking the ‘‘Copy to Drive’’
270%) to shorten only a little (less than 5.5207%), which is or ‘‘File > Save a copy in Drive’’. After that, they must
unnecessary in most robot missions. When it is necessary to get access to https://drive.google.com/drive/u/0/my-drive to
reduce the length of the travel path, we need to improve the open and edit that copy of the simulation program by using
control algorithms to shorten the length of the travel path with Google Colaboratory.
no increase of energy loss or negligible increase of energy This program consists of an inevitable program initial-
loss. ization section, a robot model, simulation tools, controllers,
a simulation section, an interactive simulation and a simula-
2) USING ONLY ONE CONTROLLER WITH DIFFERENT tion results section.
REFERENCE LINEAR VELOCITY
When the robot is controlled by only the ‘‘Lyapunov-based 1) THE INEVITABLE PROGRAM INITIALIZATION SECTION
controller’’, it is set on different values of the reference linear This section declares some libraries (NumPy, Math, SciPy.
velocity. Based on Table 7, Table 8, and Fig.18, Fig.19 it can Integrate, matplotlib. pyplot, etc.) used for programming in
be found that the faster linear velocity increases, the bigger the next sections.

VOLUME 8, 2020 170925

 A. Stefek et al.: Energy Comparison of Controllers Used for DDWMR

2) THE ROBOT MODEL SECTION the carried experiments, the controllers from a group of the
The Robot model section introduces some different shapes of low energy controllers have been offering significantly less
the DDWMR, robot parameters, basic equations used for the energy consumption and still considerable precision in the
DDWMR, the kinematic model, kinematic model implemen- tracking of the given path. The experiments have shown that
tation, the dynamic model, and dynamic model implementa- the shortage of the travel path is less than 5 percent while
tion. Users can change the robot parameters in this section for energy consumption is almost more than 200 percent. The
their own robot parameters. comparison of low energy controllers and high energy con-
trollers has raised a question if the advantages of high energy
3) THE SIMULATION TOOLS SECTION controllers are so crucial their real usage. As the comparison
Simulation tools are used for numerical computation always matters, standard and open tools should be used. Such
RK45 function from SciPy library. They consist of computa- tools allow us to build a rich database of experiments and
tion, given path, robot model and controller connection, clo- transparent comparison of any given controller with the old
sure function for simulation, simulation runner, path plotter, and already proven one. To support this idea the software
value plotter, all in one plotter, and data extractor. has been developed with the use of the Google Colab R and
Jupyter platforms and opened for all users.
4) THE CONTROLLERS SECTION In our future research, we will focus on improving a motion
The controllers section allows users to add some controllers control algorithm for the DDWMR that tracks a given path
or define their own control algorithm at ‘‘Define your con- with higher accuracy, less energy loss but smooth movement
troller here’’. That function will return the velocity and omega without or less vibration and a shorter travel time. The result
value of the robot. will be tested by using this open-source code program.

REFERENCES
5) THE SIMULATION SECTION
[1] R. Mathew and S. S. Hiremath, ‘‘Development of waypoint tracking con-
The simulation section allows users to definite square path, troller for differential drive mobile robot,’’ in Proc. 6th Int. Conf. Control,
straightforward path, or put their own path. In this section, Decis. Inf. Technol. (CoDIT), Paris, France, Apr. 2019, pp. 1121–1126.
[2] X. Zhou, B. Ma, and L. Yan, ‘‘Adaptive output feedback tracking controller
the experiment is described and executed. for wheeled mobile robots with unmeasurable orientation,’’ in Proc. 37th
In the ‘‘Experiment Description’’, users can change the Chin. Control Conf. (CCC), Wuhan, China, Jul. 2018, pp. 412–417.
value of velocity limitation and also can change the value of [3] M. Yallala and S. J. Mija, ‘‘Path tracking of differential drive mobile robot
using two step feedback linearization based on backstepping,’’ in Proc. Int.
the coefficients in the controllers. Conf. Innov. Control, Commun. Inf. Syst. (ICICCI), Greater Noida, India,
Aug. 2017, pp. 1–6.
6) THE INTERACTIVE SIMULATION SECTION [4] N. V. Tinh, N. T. Linh, P. T. Cat, P. M. Tuan, M. N. Anh, and
N. P. T. Anh, ‘‘Modeling and feedback linearization control of a nonholo-
In the Interactive Simulation section, you can choose differ- nomic wheeled mobile robot with longitudinal, lateral slips,’’ in Proc. IEEE
ent experiment sets on different paths by selecting different Int. Conf. Autom. Sci. Eng. (CASE), Fort Worth, TX, USA, Aug. 2016,
‘‘experimTag’’ and ‘‘pathTag’’. pp. 996–1001.
[5] D. Diaz and R. Kelly, ‘‘On modeling and position tracking control of the
generalized differential driven wheeled mobile robot,’’ in Proc. IEEE Int.
7) THE SIMULATION RESULTS SECTION Conf. Automatica (ICA-ACCA), Curico, Chile, Oct. 2016, pp. 1–6.
[6] J. Meng, A. Liu, Y. Yang, Z. Wu, and Q. Xu, ‘‘Two-wheeled robot platform
In this section, users can get the simulation text results, based on PID control,’’ in Proc. 5th Int. Conf. Inf. Sci. Control Eng.
change the simulation image output size, simulate all graphs (ICISCE), Zhengzhou, China, Jul. 2018, pp. 1011–1014.
in one figure, simulate the path time, simulate the travel [7] S. Peng and W. Shi, ‘‘Adaptive fuzzy output feedback control of a non-
holonomic wheeled mobile robot,’’ IEEE Access, vol. 6, pp. 43414–43424,
distance over time, and simulate the robot kinetic energy loss 2018.
over time. [8] Y. Jinhua, Y. Suzhen, and J. Xiao, ‘‘Trajectory tracking control of WMR
Thus, it is clear that this open-source code allows users to based on sliding mode disturbance observer with unknown skidding
and slipping,’’ in Proc. 2nd Int. Conf. Cybern., Robot. Control (CRC),
use different types of paths for the robot by defining them in Chengdu, China, Jul. 2017, pp. 18–22.
the section ‘‘Simulation’’ and subsection ‘‘Paths Definition’’. [9] F.-G. Rojas-Contreras, A.-I. Castillo-Lopez, L. Fridman, and
At the same time, it also allows users to use other control V.-J. Gonzalez-Villela, ‘‘Trajectory tracking using continuous sliding
mode algorithms for differential drive robots,’’ in Proc. IEEE 56th Annu.
algorithms by adding or modifying the codes in the section Conf. Decis. Control (CDC), Melbourne, VIC, Australia, Dec. 2017,
‘‘Controllers’’ and subsection ‘‘Define your controller here’’. pp. 6027–6032.
[10] B. B. Mevo, M. R. Saad, and R. Fareh, ‘‘Adaptive sliding mode control
of wheeled mobile robot with nonlinear model and uncertainties,’’ in
V. SUMMARY AND CONCLUSION Proc. IEEE Can. Conf. Electr. Comput. Eng. (CCECE), Quebec City, QC,
In this paper, the set of various well-known controllers for Canada, May 2018, pp. 1–5.
the differential driven robot has been discussed. As a main [11] P. Petrov and V. Georgieva, ‘‘Adaptive velocity control for a differential
drive mobile robot,’’ in Proc. 20th Int. Symp. Electr. App. Technol. (SIELA),
attribute of those controllers, the model of energy consump- Bourgas, Bulgaria, Jun. 2018, pp. 1–4.
tion has been taken into account. Visually, the ability to [12] A. Štefek, V. Křivánek, Y. T. Bergeon, and J. Motsch, ‘‘Differential drive
track the given path is important and thus the differences of robot: Spline-based design of circular path,’’ Dynamical Systems: Theo-
retical and Experimental Analysis (Springer Proceedings in Mathematics
the energy consumptions have been evaluated in relation to & Statistics), vol. 182, J. Awrejcewicz, Ed. Cham, Switzerland: Springer,
differences in the total lengths of travel paths. According to 2016, pp. 331–342.

170926 VOLUME 8, 2020

A. Stefek et al.: Energy Comparison of Controllers Used for a DDWMR

[13] S. G. Tzafestas, Introduction to Mobile Robot Control, 1st ed. Amsterdam, ALEXANDR STEFEK received the M.Sc. degree
The Netherlands: Elsevier, 2014, sec. 5, pp. 159–169. (summa cum laude) in command automation,
[14] A. L. Nelson, G. J. Barlow, and L. Doitsidis, ‘‘Fitness functions in evolu- and electronic computers, and the Ph.D. degree
tionary robotics: A survey and analysis,’’ Robot. Auto. Syst., vol. 57, no. 4, from the Military Academy, Czech Republic,
pp. 345–370, Apr. 2009, doi: 10.1016/j.robot.2008.09.009. in 1994 and 1998, respectively. His research inter-
[15] R. L. S. Sousa, M. D. do Nascimento Forte, F. G. Nogueira, and est includes stochastic optimization.
B. C. Torrico, ‘‘Trajectory tracking control of a nonholonomic mobile
robot with differential drive,’’ in Proc. IEEE Biennial Congr. Argentina
(ARGENCON), Buenos Aires, Argentina, Jun. 2016, pp. 1–6.
[16] B. Shi, Y. Su, C. Wang, L. Wan, and Y. Qi, ‘‘Recovery path planning
algorithm based on dubins curve for autonomous underwater vehicle,’’ in
Proc. IEEE 8th Int. Conf. Underwater Syst. Technol., Theory Appl. (USYS),
Wuhan, China, Dec. 2018, pp. 1–5.
[17] M. Begnini, D. W. Bertol, and N. A. Martins, ‘‘A robust adaptive fuzzy
variable structure tracking control for the wheeled mobile robot: Simula-
tion and experimental results,’’ Control Eng. Pract., vol. 64, pp. 27–43,
Jul. 2017, doi: 10.1016/j.conengprac.2017.04.006.
[18] S. Peng and W. Shi, ‘‘Adaptive fuzzy output feedback control of a non- THUAN VAN PHAM received the engineer’s
holonomic wheeled mobile robot,’’ IEEE J. Control Eng. Pract., vol. 64,
degree and the master’s degree in cybernetics and
pp. 27–43, 2018, doi: 10.1109/ACCESS.2018.2862163.
automation engineering from Le Quy Don Univer-
[19] M. Yue, L. Wang, and T. Ma, ‘‘Neural network based terminal sliding mode
control for WMRs affected by an augmented ground friction with slip- sity, Hanoi, Vietnam, in 2007 and 2015, respec-
page effect,’’ IEEE/CAA J. Automatica Sinica, vol. 4, no. 3, pp. 498–506, tively. He is currently pursuing the Ph.D. degree
Jul. 2017. in technical cybernetics and mechatronics with
[20] H. Yang, X. Fan, P. Shi, and C. Hua, ‘‘Nonlinear control for tracking and the University of Defence, Brno, Czech Republic.
obstacle avoidance of a wheeled mobile robot with nonholonomic con- His research interests include intelligent sensing
straint,’’ IEEE Trans. Control Syst. Technol., vol. 24, no. 2, pp. 741–746, and control, embedded computers, navigation for
Mar. 2016, doi: 10.1109/TCST.2015.2457877. mobile robot, motion control of wheeled mobile
[21] L. Xin, Q. Wang, J. She, and Y. Li, ‘‘Robust adaptive tracking control of robots, and intelligent control systems.
wheeled mobile robot,’’ Robot. Auto. Syst., vol. 78, pp. 36–48, Apr. 2016,
doi: 10.1016/j.robot.2016.01.002.
[22] M. Wahab, F. Rios-Gutierrez, and A. El Shahat, ‘‘Energy modeling of
differential drive robots,’’ in Proc. SoutheastCon, Fort Lauderdale, FL,
USA, Apr. 2015, pp. 1–6. VACLAV KRIVANEK (Member, IEEE) received
[23] S. Armah, S. Yi, and T. A. Lebdeh, ‘‘Implementation of autonomous
the M.Sc. degree in control and guidance sys-
navigation algorithms on two-wheeled ground mobile robot,’’ Amer.
tems of missiles from the Military Academy Brno,
J. Eng. Appl. Sci., vol. 78, pp. 36–48, Apr. 2016, doi: 10.3844/aje-
assp.2014.149.164. Czech Republic, in 2002, the Mastèrè Spécialisé
[24] L. A. Yekinni and A. Dan-Isa, ‘‘Fuzzy logic control of goal-seeking 2- degree in techniques for aeronautics and space
Wheel differential mobile robot using unicycle approach,’’ in Proc. IEEE from SUPAERO, Toulouse, France, in 2006, and
Int. Conf. Autom. Control Intell. Syst. (I2CACIS), Selangor, Malaysia, the Ph.D. degree in diagnostic methods from the
Jun. 2019, pp. 300–304. University of Defence (UoD), Brno, Czech Repub-
[25] P. Xinzhe, L. Zhiyuan, P. Run, and C. Hong, ‘‘Practical stabilization of lic, in 2010. From 2011 to 2019, he was a Member
wheeled mobile robots based on control Lyapunov function,’’ in Proc. Int. with the Department of Air Defence, UoD. Since
Conf. Control Appl., Glasgow, U.K., Sep. 2002, pp. 345–349. 2020, he has been a Lecturer with the Department of Military Robotics, UoD.
[26] H. Chitsaz, S. M. LaValle, D. J. Balkcom, and M. T. Mason, ‘‘Mini- His research interests include feed-back control systems applied to the air
mum wheel-rotation paths for differential-drive mobile robots,’’ in Proc. defence systems and mobile robot design.
IEEE Int. Conf. Robot. Autom. (ICRA), Orlando, FL, USA, May 2006,
pp. 1616–1623.
[27] X. Song and S. Hu, ‘‘2D path planning with dubins-path-based A∗ algo-
rithm for a fixed-wing UAV,’’ in Proc. 3rd IEEE Int. Conf. Control Sci.
Syst. Eng. (ICCSSE), Beijing, China, Aug. 2017, pp. 69–73.
KHAC LAM PHAM received the engineer’s
[28] M. Elbanhawi, M. Simic, and R. N. Jazar, ‘‘Continuous path smoothing
for car-like robots using B-Spline curves,’’ J. Intell. Robotic Syst., vol. 80, degree in informatics and control systems from
no. S1, pp. 23–56, Jan. 2015, doi: 10.1007/s10846-014-0172-0. Bauman Moscow State Technical University,
[29] S. Liu and D. Sun, ‘‘Minimizing energy consumption of wheeled Moscow, Russia, in 2010. He is currently pursuing
mobile robots via optimal motion planning,’’ IEEE/ASME the Ph.D. degree in electrical engineering with the
Trans. Mechatronics, vol. 19, no. 2, pp. 401–411, Apr. 2014, University of Defence, Brno, Czech Republic. His
doi: 10.1109/TMECH.2013.2241777. research interests include design and control of
[30] P. B. Deb, O. Saha, S. Saha, and S. Paul, ‘‘Dynamic model analysis of a DC electric drivers for UAV using battery.
motor in MATLAB,’’ IJSER J. Int. J. Sci. Eng. Res., vol. 8, no. 2, pp. 57–60,
Mar. 2017.

VOLUME 8, 2020 170927