Sei sulla pagina 1di 9

Research Article

Research Article

Adaptive control of two-wheeled mobile balance robot capable to adapt different surfaces using a novel artificial neural network–based real-time switching dynamic controller

Ali Unluturk and Omer Aydogdu

International Journal of Advanced Robotic Systems March-April 2017: 1–9 ª The Author(s) 2017 DOI: 10.1177/1729881417700893 journals.sagepub.com/home/arx

10.1177/1729881417700893 journals.sagepub.com/home/arx Abstract In this article, a novel real-time artificial

Abstract In this article, a novel real-time artificial neural network–based adaptable switching dynamic controller is developed and practically implemented. It will be used for real-time control of two-wheeled balance robot which can balance itself upright position on different surfaces. In order to examine the efficiency of the proposed controller, a two-wheeled mobile balance robot is designed and a test platform for experimental setup is made for balance problem on different surfaces. In a developed adaptive controller algorithm which is capable to adapt different surfaces, mean absolute target angle deviation error, mean absolute target displacement deviation error and mean absolute controller output data are employed for surface estimation by using artificial neural network. In a designed two-wheeled mobile balance robot system, robot tilt angle is estimated via Kalman filter from accelerometer and gyroscope sensor signals. Furthermore, a visual robot control interface is developed in Cþþ software development environment so that robot controller parameters can be changed as desired. In addition, robot balance angle, linear displacement and controller output can be observed online on personal computer. According to the real-time experimental results, the proposed novel type controller gives more effective results than the classic ones.

Keywords Adaptive switching controller, ANN-based surface estimation, sensor fusion, two-wheeled mobile balance robot

Date received: 29 November 2016; accepted: 14 February 2017

Topic: Mobile Robots and Multi-Robot Systems Topic Editor: Nak-Young Chong Associate Editor: Euntai Kim

Introduction

The studies in the field of robotics have enormously advanced and improved nowadays. In parallel with this improvement, robots make our lives easy in many fields like industry, health, society, and so on. 1 The capabilities of many robots, used actively now, have advanced consider- ably. The more they advance, the more they need for energy. For that reason, in the last decade, many research- ers have studied two-wheeled balance robots which need less energy and have a very good manoeuvrability instead of four-wheeled robots. 2,3 Many researchers worldwide

have made research about various types of explorative two-wheeled mobile balance (TWMBR) which balances itself, moves backward and forward turns right and left. 4,5

Department of Electrical & Electronics Engineering, Faculty of Engineering, Selcuk University, Konya, Turkey

Corresponding author:

Ali Unluturk, Department of Electrical & Electronics Engineering, Faculty of Engineering, Selcuk University, Konya 42050, Turkey. Email: aunluturk@selcuk.edu.tr

2

International Journal of Advanced Robotic Systems

2 International Journal of Advanced Robotic Systems Figure 1. The structure of TWMBR: (a) front view,

Figure 1. The structure of TWMBR: (a) front view, (b) top view, (c) side view and top view of TWMBR on the coordinate system. TWMBR: two-wheeled mobile balance robot.

Many of these studies published are based upon simulation results and theoretical analysis. Because of their non-linear structure, that kind of system dynamics are rather complex and their motion planning is very hard. 6 However, many researchers have built real-time TWMBR systems and implemented their own control algorithms on those robot systems. For example, they have developed a robot called JOE in Industrial Electronic Laboratory in Sweden Federal Technology Institute. 7 Kuindersma et al. working for Per- ceptual Robotic Laboratories have studied a humanoid U Bot-5. 8 National Aeronautics and Space Administration’s (NASA’s) Robonaut, another two-wheeled robot, has been made by Defense Advanced Research Projects Agency (DARPA) with the cooperation of NASA. 9 Cardea, devel- oped by Massachusetts Institute of Technology, is another sample of TWMBR. 10 The two-wheeled vehicle called B2 has been designed as an alternative automatic commercial transportation system. 11 Numerous controller structures are created for the control and balance of TWMBR systems such as classical control, 12 proportional integral-velocity control, 13 Fuzzy logic-based control, 14 neural network con- trol, 3 sliding mode control 4 and adaptive control. 15 As seen in the literature studies above, most of the studies about TWMBR have been made on flat and firm grounds. In addition, Kim and Lee have developed a magnet-wheeled mobile robot moving on curved surfaces. 16 Qiao et al. have proposed wheel-legged robot with a front module, a rear module and an active waist joint in order to make the robot pass through the curved narrow channel. 17 In this article, balance performance of the robot is observed on loose surface such as sand, pebble and soil. Besides, artificial neural network (ANN)-based real-time switching dynamic controller, which is applied on the robot to solve balance problem on such surfaces, is a novel structure. This article is organized as follows: The second sec- tion describes structure of the TWMBR, the mobile robot hardware components and the robot control inter- face on personal computer (PC). The third section gives

the mathematical analysis for developed TWMBR sys- tem. The fourth section describes dynamic surface esti- mator on the control system and controller mechanism of the robot system. The fifth section presents experi- mental results of the balance robot, and the last section gives the study results.

Structure of the TWMBR and the robot control hardware

For TWMBR’s mechanical design, an appropriate equili- brium point must be tuned in order to control the robot because of its non-linear structure. The hardware of the robot was built regarding all the requirements. As shown in Figure 1(a) and (b), the developed TWMBR has three layers which make it lighter and has a good centre of gravity. At the bottom layer, there are two identical per- manent magnet DC motors coupled to a gearbox for each wheel and with an integrated encoder. There is also micro- electro-mechanical systems (MEMS)-based inertia mea- surement unit sensor which contains gyroscope and accelerometer together. Cortex M3 microprocessor–based development board which controls all units of the robot system and provides data communication by using Blue- tooth wireless unit. In the middle layer, there is a DC motor driver circuit which can control two identical DC motors moving independently at the robot system. At the top layer, there is a lithium polymer battery supplying energy for the whole system. Figure 1(c) shows the side view and top view of TWMBR on coordinate system for mathematical analysis. The general block diagram given in Figure 2(a) shows the detailed hardware component of the developed TWMBR control system. As shown in Figure 2(a), sensor data are taken for the robot control on I2C data line by Cortex M3 microprocessor. Furthermore, left and right encoder data are taken on timer 1 and timer 2 input ports. The robot control interface was designed in Visual Cþþ software development environment for online control of TWMBR remotely, for online change of all robot control

Unluturk and Aydogdu

3

Unluturk and Aydogdu 3 Figure 2. (a) The general block diagram of TWMBR control hardware; (b)

Figure 2. (a) The general block diagram of TWMBR control hardware; (b) robot balance test platform. TWMBR: two-wheeled mobile balance robot.

parameters and to observe and make sense of robot’s reac- tion to different surfaces diagrammatically on PC by using the wireless communication unit. Mathematical basis of parameter adaptation algorithm was built with these online data and performed real-time through the codes of Cortex M3 microprocessor. Many real-time experiments were done to enable robot’s stable balance on various surfaces by means of the designed user interface. On the designed TWMBR control system, MEMS-based sensor groups were used, which include three-axis gyro- scope sensors and three-axis accelerometer sensors. The information how we achieved sensor fusion and sensor calibration for TWMBR is given in detail in the work by Guner et al. 18 Additionally, designed robot balance test platform is 70 cm 210 cm 4 cm of sizes. It was divided into three parts and each part was filled with soil, pebble and sand as shown in Figure 2(b).

Mathematical model of TWMBR system

Mathematical dynamic modelling of TWMBR is rather important in terms of stability analysis and system simula- tion. It is also very important that control algorithms are created according to these model parameters. Many researchers generally studying in the field of TWMBR obtain the robot’s non-linear dynamic model by utilizing the equations such as Lagrange-Euler or Newton-Euler. In this article, TWMBR system is modelled based on Lagrange-Euler equations referred as Yamamoto. 19 Figure 1(c) shows coordinate system of side and top views of TWMBR. In Figure 1(c), x b , y b , z b denote the system’s centre of gravity, x m , y m , z m denote the robot wheel’s centre of gravity and θ l ; r defines left and right wheel angle on the coordinate system. It is necessary to specify the robot’s physical parameters very well to get the TWMBR’s equa- tion of motion properly. Physical parameters of the applied balance robot are given in Table 1. Considering the coordinate system in Figure 1(c) and physical parameters of balance robot in Table 1, TWMBR’s equations of motion were derived by using

Table 1. Physical parameters of TWMBR.

Symbol

Description

Unit

g

Gravitational acceleration

m/s 2

m

Wheel weight (for each of wheel)

kg

R

Wheel radius

m kg m 2

J

w ¼ mR 2 = 2

Wheel inertia moment (one wheel)

M

Body weight

kg

w

Body width

m

D

Body depth

m

H

Body height

m

L

¼ H= 2

Distance of the centre of mass m

J Ψ

¼ ML 2 = 3

Body pitch inertia moment

kg m 2 kg m 2 kg m 2 O

J ;

¼ Mð w 2 þ D 2 Þ = 12 Body

yaw inertia moment

J m

DC motor inertia moment

R m

DC motor resistance

K b

K t

B m

L m

n ¼ N 2 : N 1

f m

f w

DC Motor back EMF constant Vs/rad

DC motor torque constant

Nm/A

Rotating damper

Nms/rad

DC motor inductance

H

Gear ratio

Friction between body and DC motor

Friction between wheel and floor

TWMBR: two-wheeled mobile balance robot; Electromotive Force: EMF.

Lagrange method. If TWMBR’s front becomes positive x-axis and t ¼ 0, balance equation for each coordinate will be as given below

½ θ; ; ¼ 2 ð θ l þ θ r Þ; w R ð θ r θ l Þ

1

(1)

where the robot’s right and left wheels mean angle is defined

by θ and body yaw angle by ; . As shown in Figure 1(c), shaft

centre to which wheels are integrated (x m ; y m ; z m ), the

posi-

tion of left wheel (x l ; y l ; z l ) and right wheel (x r ; y r ; z r ) on the coordinate system and robot’s bodyweight centre

(x b ; y b ; z b ) are expressed in equations (2) to (5), respectively

4

International Journal of Advanced Robotic Systems

ð x m ;

y m ;

z m Þ ¼

Z

x_ m dt ;

Z

y_ m dt ;

ð x_ m ; y_ m Þ¼ð R θ cos; ; R θ sin ; Þ

_

_

R ;

ð x l ;

y l ;

ð x r ;

y r ;

ð x b ;

y b ;

z l Þ ¼

z r Þ ¼

z b Þ

x m þ sin ; ; y m þ w cos; ; z m

2

2

x

m

þ w sin ; ; y m w cos ; ; z m

2

2

(2)

(3)

(4)

¼ ð x m þ L sin Ψ cos ; ; y m þ L sin Ψ sin ; ; z m þ Lcos Ψ Þ

(5)

where Ψ denotes body tilt angle. The robot’s energy equa- tions can be defines as

T 1 ¼

1

2

þ

m ðx_ l þ y_ l

2

2

1

þ z_ l Þ þ 2 mð x_ r þ y_ r þ z_

2

2

2

1

2 M ðx_ þ y_ þ z_ Þ

2

b

2

b

2

b

2

r

Þ

(6)

T 2 ¼

1

2 J w

þ

_

θ

2

l þ

1

2 J w

_

θ

2

r þ

_

1

2 n 2 J m ð θ r Ψ Þ 2

_

1

2 J Ψ

Ψ 2 þ

_

1

2 J ; ; 2 þ

_

1

2

n 2 J m ð θ l Ψ Þ 2

_

_

(7)

(8)

where T 1 denotes translational kinetic energy, T 2 denotes rotational kinetic energy and U defines potential energy. Using kinetic energy equations, Lagrange function L has the following expression

L ¼ T 1 þ T 2 U

(9)

Using the Lagrange function, force components of θ , Ψ , ; and for linearizing process in the equilibrium point of TWMBR, we consider the Ψ 0 ð sin Ψ Ψ ; cosΨ 1 Þ

and neglect the second order term like Ψ 2 0. In this case,

motion equations are defined approximately as follows:

F θ ¼ ½ð 2m þ M Þ R 2 þ 2J w þ 2n 2 J m θ þ ð MLR 2n 2 J m Þ Ψ

U ¼ mgz l þ mgz r þ Mgz b

_

(10)

F Ψ ¼ ð

MLR 2n 2 J m Þ θ þ ð ML 2

þ J Ψ þ 2n 2 J m Þ Ψ MgLΨ

(11)

F ; ¼

1

2

mw

2

þ

J ;

þ

w

2

2R

2

ðJ w þ n

2

J

m

Þ

;

(12)

Considering DC motor torque and viscose friction, gen- eralized forces are represented as in the following equations

ð F θ ;

F Ψ ;

F ; Þ ¼

F l þ F r ;

F Ψ ; 2R ð F r F l Þ

w

_

F l ¼ nK t i l þ f m ð Ψ θ l Þ f w θ l

_

_

(13)

(14)

 

_

_

_

F r ¼ nK t i r þ f m ð Ψ θ r Þ f w θ r

 

(15)

_

_

_

_

F Ψ ¼ nK t i l nK t i r f m ð Ψ θ l Þ f m ð Ψ θ r Þ

(16)

where i l ; r denotes left and right motor current and v l ; r

denotes left and right motor voltage. In the study, we will

evaluate DC motor equation utilizing the relation between

i l ; r current and v l ; r voltage. If we neglect the frictions

inside the motor (B m ) and motor inductance (L m ), the DC motor current will be as in equation (17)

i l ; r

¼

V l ; r þ K b ð

_

Ψ

_

θ l ; r Þ

R m

(17)

Using equation (17), the generalized forces are expressed as follows

F θ ¼ ð v l þ v r Þ 2 ð þ f w Þ θ þ 2 Ψ

(18)

(19)

_

_

_

_

F Ψ ¼ ð v l þ v r Þ þ 2 θ 2 Ψ

F ; ¼

w

w 2

2R ð v r v l Þ 2R 2 ð þ f w Þ

_

;

(20)

where

As shown in the equations of motion above, the motion dynamic of the system depends on surface struc- ture. In this study, it is shown with real-time application results that controller parameters change on loose sur- faces. State space expressions of TWMBR, developed according to sytem parameters, are presented in “Experimental results” section.

¼ nK t ; ¼ nK t K b

R

m

R

m

þ

f m .

Design of adaptive dynamic controller for different surfaces

It is rather hard for balance robot system to keep its balance on two wheels and on all surfaces robustly and presents the

desired behaviour. It is necessary to design a dynamic con- troller so that we can overcome the robot’s adaptation prob- lem on different surfaces shown in Figure 2(b), and the

system behaviour can be more stable on all various grounds. In this study, TWMBR is able to sense the surface autonomously by means of the proposed novel algorithm.

Thus, it can update the optimal controller parameters according to surface by virtue of ANN-based adaptive con- troller. The proposed ANN-based adaptive switching con-

troller structure is given in Figure 3.

As shown in Figure 3, the system consists of multi- controller, ANN-based adaptation mechanism and sensor fusion blocks. In this figure, Ψ Target expresses target tilt angle and X Target is target linear displacement. And also,

^

X Wheel is wheel displacement calculated as R θ . Ψ is system

output, y^ 1 ð t Þ is estimator output and σ is surface switching

function which can take the values 1, 2, 3. The general

Proportional Integral Derivative (PID) controller structure used for the robot control is given in equation (21)

Unluturk and Aydogdu

5

Unluturk and Aydogdu 5 Figure 3. TWMBR’s block diagram of adaptive control with switching. TWMBR: two-wheeled

Figure 3. TWMBR’s block diagram of adaptive control with switching. TWMBR: two-wheeled mobile balance robot.

with switching. TWMBR: two-wheeled mobile balance robot. Figure 4. Change in (a) tilt angle, (b) linear

Figure 4. Change in (a) tilt angle, (b) linear displacement, (c) controller output on three floors.

u σ ð t Þ ¼ K p σ e ðt Þ þ K i σ Z eð τ Þd τ þ K d σ

deð t Þ

dt

;

ð σ ¼ 1; 2; 3 Þ

(21)

where u σ ð t Þ denotes control signal for different surfaces

and eð t Þ defines control error ( Ψ Target Ψ Þ . Fine tuning

of the controller parameters is found by trial-and-error via robot control interface. The most sensitive PID parameters

6

International Journal of Advanced Robotic Systems

6 International Journal of Advanced Robotic Systems Figure 5. (a) MATADE-MATDDE-MACO data on three floor; (b)

Figure 5. (a) MATADE-MATDDE-MACO data on three floor; (b) ANN model of TWMBR control system. MATADE: mean absolute target angle deviation error ; -MATDDE: mean absolute target displacement deviation error; MACO: mean absolute controller output; ANN: artificial neural network; TWMBR: two-wheeled mobile balance robot.

Table 2. ANN system output performance for HNN is 8.

Act. Func.

HNN

FN

Performance

LR ¼ 0.2 Test

LR ¼ 0.4 Test

LR ¼ 0.6 Test

LR ¼ 0.8 Test

LR ¼ 1 Test

LR ¼ 1.2 Test

Tangent sigmoid

8

3

% Accuracy Test period (sn) Iteration % Accuracy Test period (sn) Iteration % Accuracy Test period (sn) Iteration

94.87

95.40

95.13

94.67

94.40

94.20

 

0.0053

0.0055

0.0056

0.0072

0.0052

0.0105

109

175

350

148

118

240

Logarithmic sigmoid

8

3

95.93

95.80

96.07

95.73

95.80

95.73

 

0.0056

0.0062

0.0055

0.0059

0.0065

0.0059

208

193

190

987

271

180

Purelin

8

3

99.27

98.67

98.00

97.87

97.67

98.73

 

0.0055

0.0144

0.0062

0.0055

0.0058

0.0054

106

228

180

331

233

175

ANN: artificial neural network; HNN: hidden neuron number; FN: feature number; LR: learning rate.

Table 3. Physical parameters of TWMBR.

Robot parameters

M ¼ 1: 372

g

¼ 9: 81

m ¼ 0: 146

R ¼ 0:0625

J w ¼ 0: 00057

w

¼ 0: 28

D ¼ 0: 16

H ¼ 0:09

L ¼ 0: 045

J ; ¼ 0:01189

J m ¼ 1 10 5

R m ¼ 2.4

K b ¼ 0:308

B m ¼ 0

L m ¼ 0

n ¼ 30 : 1

f m ¼ 0 :0022

J Ψ ¼ 0: 00093

K t ¼ 0: 155

f w ¼ 0

TWMBR: two-wheeled mobile balance robot.

for soil, pebble and sand floor are specified for this goal as given in Figure 3.

ANN-based adaptation mechanism

In the study, the robot’s real-time balance experiments are done on three different floors which have soil, pebble and sand on it. During these experiments, actions of the robot are analysed considering data such as tilt angle Ψ ð t Þ , linear displacement of robot X ðt Þ and plant input u σ ð t Þ for three different surfaces. The action of the robot on each floor (soil, pebble and sand) is analysed one by one referring

to the PID controller test parameter such as K p t ¼ 29; K i t ¼ 12; K d t ¼ 9 obtained for soil and data about robot are transferred to computer via bluetooth unit on the robot. Firstly, as shown in Figure 4(a), the robot’s tilt angle; Figure 4(b), the distance it takes; and Figure 4(c), controller output u σ ð t Þ ; ðσ ¼ 1 ; 2; 3Þ change in terms of time are observed on three surfaces for 20 s without any external factor. As Figure 4 shows, the reaction of the robot to different floors is not the same. The robot’s centre of gravity is 0 as a result of experi- mental tests. Therefore, it was chosen as Ψ Target ¼ 0 . In this article, to overcome this problem, an ANN-based

Unluturk and Aydogdu

7

Unluturk and Aydogdu 7 Figure 6. Change in (a) tilt angle, (b) linear displacement and (c)

Figure 6. Change in (a) tilt angle, (b) linear displacement and (c) controller output on three floors.

adaptation mechanism was designed. For the designed parameter adaptation mechanism, the balance state of the robot on each floor was observed with the same controller parameter for 2 min without having any external factor, and for each floor, 12,000 data were obtained totally. Because that the robot system actions are time-variable and non-linear, the data are re-evaluated with the method of systematic sampling (N ¼ 600). The equations about mean absolute target angle devia- tion error (MATADE), mean absolute target displace- ment deviation error (MATDDE) and mean absolute controller output (MACO) are given in equations (22–24), respectively

MATADE ¼ 1

N

MATDDE ¼ 1

N

MACO ¼ 1

N

j ¼ 1

N

N

X jΨ Target Ψ Output j j

j ¼ 1

N

X j X Target X Output j j

X j U j j

j ¼ 1

(22)

(23)

(24)

Hence, we have 20 meaningful data about the robot for each floor in a controller test parameter. In Figure 5(a), the 20 data (MATADE, MATDDE and MACO) used in the structure of ANN surface classifier are shown.

These data are classified with ANN in order to define the floor on which the robot stands. After surface classification, the prediction of the floor is made by parameter adaptation mechanism. The model of ANN structure used for this study is shown in Figure 5(b). That ANN model is capable of evaluating input data of the robot and defining its ground. Also, three output signals ð σ ¼ 1 ; 2 ; 3 Þ making controller choice for three surfaces (pebble is 1, sand is 2 and soil is 3) are obtained. In the study, it is chosen that hidden neuron number (HNN) as 4, 5, 6, 7, 8, 9, 10, learning rate (LR) as 0.2, 0.4, 0.6, 0.8, 1.0, 1.2 values, feature number as 3, target stopping criteria as 0.001 accuracy value or 1000 iteration number and the

momentum constant as 0.9 to determine the appropriate

ANN structure. Also, three different activation functions

(logarithmic sigmoid, tangent sigmoid, purelin) are used for neurons on hidden layer and output layer in each experi-

ment. That ANN structure has a feed-forward and multi-

layer network trained with back propagation algorithm. The best performance value for ANN model has been

obtained using LR ¼ 0.2, HNN ¼ 8 in purelin activation function. The table of some of ANN structure used for this

study is shown in Table 2. A real-time ANN algorithm is made and applied on the robot real-time considering the weight bias and activation function of this network structure and by applying the opti- mum controller parameter on the robot it is achieved that the robot can keep its balance more stably.

8

International Journal of Advanced Robotic Systems

Table 4. TWMBR performance evalution.

 

Pebble

Sand

Tilt angle change

Classic controller ANN-based switching dynamic controller

8 Ψ

7

6 Ψ

5

5 Ψ 6

4 Ψ 4

Displacement change

Controller output change

Classic controller ANN-based switching dynamic controller Classic controller

5 V

7

3

cm X 5 cm X 3

V 4 V

cm

cm

3 cm X 2: 5 cm 2: 5 cm X 2:5 cm 3 V V 4 V

ANN-based switching dynamic controller

2 : 5 V V 2: 5 V

2: 5 V X 2: 5 V

TWMBR: two-wheeled mobile balance robot; ANN: artificial neural network.

mobile balance robot; ANN: artificial neural network. Figure 7. Surface deformation of (a) soil, (b) pebble

Figure 7. Surface deformation of (a) soil, (b) pebble and (c) sand floor.

Experimental results

In this section, we will implement the TWMBR by using the effective controller based on the proposed ANN-based adaptive switching controller. Moreover, in the experimen- tal test, we assume the robot physical parameters as given in Table 3. Equation (25) shows the state space expression obtained from TWMBR physical parameters in Table 3 using math- ematical model equations (10) to (20)

2

6

6

6

6

6

4

_

θ

_

Ψ

θ

Ψ

3

5

7

7

7

7 ¼

7

2

6

6

6

6

6

4

00

00

0

0

32 : 1287

33 : 6544

2

4

6

6

þ 6

6

6

0

0

3 : 3776

4 : 8808

1

0

2: 0883

3 :0176

0

0

3 :3776

4: 8808

3

7

7

7

7

7

5

0

1

2 : 0883

3 :0176

"

v

v

l

r

#

3 2

6 6 6 6 5 7 4 6

7 7 7 7

θ

Ψ

_

θ

_

Ψ

3

7

7

7

7

5 7

(25)

The controller parameters on different surfaces are found roughly with simulation results by utilizing the math- ematical model of the robot system. Then, controller para- meters are obtained by fine tuning via real-time robot control interface. The controller parameters given in Figure 3 were obtained as a result of real-time tests of the

robot for three surfaces as soil (σ ¼ 1), pebble ( σ ¼ 2) and sand (σ ¼ 3). For this controller, the robot’s tilt angle, the distance it takes, and plant input u σ ð t Þ change in terms of time are observed as shown in Figure 6, three floors for 20 s without any external factor. When Figure 6 is compared with the ones obtained with just one controller Figure 4, it is seen that the system keeps balance much better on different surfaces. Obtained bal- ance performance values are shown in Table 4. However, parameters in Figures 4 and 6 change rapidly

because the surfaces are loose. As seen in Figure 7, surface

deformation affects desired balance of the robot very much.

Conclusions

Many researchers have studied about two-wheeled balance robots. However, almost all of the researches have focused

on the solution of the balance problem on just one kind of

surface. In this study, a new controller structure was created

by analysing the balance problem of the robot on different surfaces. First of all, the angle oscillation, the distance it takes forward and backward and controller output changes at the same time when the robot tries to keep balance were classified with ANN and surface estimation was achieved. At the end of ANN-based classification process, the con- troller parameter of the robot was changed on the robot with respect to surface by using the adaptation mechanism. The proposed adaptive control structure with switching was applied real-time on TWMBR for different surfaces. These

Unluturk and Aydogdu

9

real-time experimental results show that the proposed con- troller structure had a positive contribution to the robot behaviour on different grounds, because the angle oscilla- tion and distance it takes forward and backward were decreased. At the same time, the controller output change was considerably limited, thus, the energy efficiency was increased. It is seen that TWMBR system becomes more robust and stable.

Declaration of conflicting interests

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

Funding

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

References

1. Ahn JK and Jung S. Development of a two-wheel mobile manipulator: balancing and interaction control. Camb J Robot 2014; 32: 1135–1152.

2. Sukvichai K and Parnichkun M. Double-level ball-riding robot balancing: from system design, modelling, controller synthesis, to performance evaluation. Elsevier Mech 2014; 24: 519–532.

3. Kausar Z, Stol K and Patel N. The effect of terrain inclination on performance and the stability region of two-wheeled mobile robots. Int J Adv Robot Syst 2012; 9: 1–11.

4. Yang C, Li Z, Cui R, et al. Neural network-based motion control of an underactuated wheeled inverted pendulum model. IEEE Trans Neural Netw Learn Syst 2014; 25: 2004–2016.

5. Xu JX, Guo ZQ and Lee TH. Design and implementation of integral sliding-mode control on an underactuated two-wheeled mobile robot. IEEE Trans Ind Electron 2014; 61: 3671–3681.

6. Ratajczak A and Tchon K. Motion planning of a balancing robot with threefold sub-tasks: an endogenous configuration space approach. In: 2011 IEEE international conference on robotics and automation (ICRA), Shanghai, China, 9–13 May 2011, pp. 6096–6101. IEEE. http://ieeexplore.ieee.org/docu

7. Grasser F, D’Arrigo A, Colombi S, et al. JOE: a mobile inverted pendulum. IEEE Trans Ind Electron 2002; 49: 107–114.

8. Kuindersma SR, Hannigan E, Ruiken D, et al. Dexterous mobility with the uBot-5 mobile manipulator. In: Advanced

robotics, 2009. ICAR 2009, Munich, Germany, 22–26 June 2009, pp. 1–7. IEEE. 9. Ambrose RO, Savely RT, Michael S, et al. Spain I, Radford

N. Mobile manipulation using NASA’s Robonaut. In: 2004 IEEE international conference on robotics and automation, 2004. Proceedings. ICRA’04. New Orleans, Louisiana, USA, 26 April–1 May, pp. 2104–2109. IEEE.

10. Brooks R, Aryananda L, Edsinger A, et al. Sensing and manipulating built-for-human environments. Int J Humanoid Robot 2004; 01: 1–28.

11. Tirmant H, Baloh M, Vermeiren L, et al. B2, an alternative two wheeled vehicle for an automated urban transportation system. In: Intelligent Vehicle Symposium, 2002. IEEE, 17– 21 June 2002, pp. 594–603. IEEE.

12. Unluturk A, Aydogdu O and Guner U. Design and PID con- trol of two wheeled autonomous balance robot. In: 2013 international conference on electronics, computer and com- putation (ICECCO), Ankara, Turkey, 7–9 November 2013, pp. 260–264. IEEE.

13. Unluturk A, Guner U and Aydogdu O. Yeni bir PI-V tipi denetleyici tasarımı ve denge robotu u¨zerinde uygulanması. Tu¨rkiye Otomatik Kontrol Ulusal Konferansı (TOK), Kocaeli, Tu¨rkiye, 11–13 Eylu¨l 2014, pp. 578–582.

14. Huang CH, Wang WJ and Chiu CH. Design and implemen- tation of fuzzy control on a two-wheel inverted pendulum. IEEE Trans Ind Electron 2011; 58: 2988–3001.

15. Ruan X, Ding M and Gong D et al. On-line adaptive control for inverted pendulum balancing based on feed- back-error-learning. Neurocomputing Elsevier 2007; 70:

770–776.

16. Kim JH and Lee JC. Dead-reckoning scheme for wheeled mobile robots moving on curved surfaces. J Intell Robot Syst Springer 2015; 79: 211–220.

17. Qiao G, Song G, Zhang Y, et al. A wheel-legged robot with active waist joint: design, analysis, and experimental results. J Intell Robot Syst Springer 2016; 83: 485–502.

18. Guner U, Canbolat H, Unluturk A, et al. MEMS tipi sen- so¨ rler kullanılarak genel amac¸lı atalet o¨lc¸u¨ m birimi

tasarımı ve kalibrasyonu. Tu¨rkiye Otomatik Kontrol Ulusal Konferansı (TOK), Denizli, Tu¨rkiye, 10–12 Eylu¨l 2015; pp. 68–73.

19. Yamamoto Y. NXTway-gs (self-balancing two-wheeled robot) Controller Design. Cybernet Systems Co, 2008.