Omniwheel mobile robot learning with sensor network based Fuzzy Logic

Anindya Devi R Jurusan Teknik Mekatronika, PENS-ITS Email : [email protected] ABSTRACT In this final project has been made a robot that is included in a category holonomic robot, that is omni wheel mobile robot. This robot include in holonomic robot category because this robot can move in any direction. The problem of using any kind of robot is how to control that robot so that they can obey any instruction from operator. To overcome that problem, used Fuzzy Logic Control in this final project, to control the robot so that the robot can move to any direction according operator instruction. Sensor network also used in robot to help robot know the obstacle around the robot. From the experiment that has been done we can get the result that environmet around the robot affected robot movement, for example this robot can not move in slippery place. The metode has been used can help navigation problem for mobile robot. Keyword : mobile robot, reinforcement learning, sensor network, fuzzy logic 2. Omni Robot with three wheel 1.Introduction The robot than can move to any direction is a kind of Autonomous mobile robot is a kind of intelligent robots that have the ability to make their own decisions, have control and power supply systems are integrated and have the ability to navigate the number of operations that allow the robot to achieve the expected goals. There are two types of mobile robots are robots that walk using the legs and a walking robot with wheels. The use of fuzzy logic technology has been widely applied in various fields, ranging from industrial process control , electronics, household and others. Likewise, the application of fuzzy logic to control the robot. With the implementation of a concept on a robot, it will provide convenience in determining the control rules that are used to bring the process into action is expected. Mobile robots are made are expected to have capabilities such as humans, adapt and learn better from environmental changes and explain the basis for decision making. One evolving concept is fuzzy, the fuzzy control system is a knowledgebased systems that are formulated in the form of rules (rules base)as an accumulationof experience in this project will discuss the use of fuzzy logic algorithms for planting in an omniwheel mobile robot

holonomic robot, this type of robot is very intersting because it has better manuver ability and more efficient. The robot with three wheel or more can have independent tangential, normal and angular velocity. Figure 1 is the mechanical configuration of omni mobile robot with three wheel. V2, ω2 f2, T2 v,Fv

x

θ ω,T,

V0, ω0

y

,

d Vn,Fv

f0, T0

n

V1, ω1 f1, T1

Figure 1. mechanical configuration omni mobile robot (Source: Helder P. Oliveira et al,. “Modelling and assesing of omni-directional robots with three wheels and four wheels”, Universidade do Porto, Faculdade de Engenharia-Portugal)

Description :  x,y,θ – robot position (x,y) and θ is robot direction toward

        

d [m] – distance of wheel with the center of robot v0, v1, v2 [m/s]– wheel linear velocity ω0, ω1, ω2 [rad/s] – wheel angular velocity f0, f1, f2 [N] – fraction of wheel T0, T1, T2 [Nm] – wheel torque V, vn [m/s] – robot linear velocity ω [rad/s] – robot angular velocity Fv, Fvn [N] – robot force for v and vn T [Nm] – robot torque(the value is depend with value of ω)

In this robot, be determined where the direction toward the robot (front). Direction V on the front ofthe picture is the robot, while the direction v is orthogonal. To find the model of the robot kinematic movement omniwheel three wheels, which must be identified or known in advance are x, y, θ, where they are closely related to velocity 𝑣𝑥 𝑡 = 𝑑𝑦 (𝑡) 𝑑𝑡

,𝜔 𝑡 =

𝑑𝜃 (𝑡) 𝑑𝑡

𝑑𝑥 (𝑡) 𝑑𝑡

2.1 Mechanical system Figure 1 is the figure of mechanical system of omni wheel mobile robot that have been made. The diameters of robot is 40 cm, with a circular platform from stainless steal. Around it there are 6 Sharp GP2D12 infrared sensor used to detect obstacles near robot. Robot uses 3 DC motors. Robot used omni wheel with diameter 8 cm. This robot can move to the 8 directions that have been determined. This robot is include to a holonomic mobile robot, because it can move to any direction. This robot will move according to the operator instruction. Robots will also get learning from the operator. If it move appropriate to the rule it will get reward but if it does not move appropriate to the rule it will get punishment. This learning called reinforcement learning.

, 𝑣𝑦 𝑡 =

.

To obtain a kinematic model of mobile robot omni wheel, the necessary transformation of the linear velocity Vx and Vy (velocity at the static axis) to the robot linear velocity v and vn (robot axis).Here is an equation (Helder P. Oliveira et al): Figure 1 Omniwheel robot design

𝑣 𝑡 𝑣𝑛 𝑡 𝜔 𝑡

cos 𝜃 𝑡 = − sin 𝜃 𝑡 0

sin 𝜃 𝑡 cos 𝜃 𝑡 0

0 𝑣𝑥 (𝑡) 0 . 𝑣𝑦 𝑡 𝜔(𝑡) 1

Since the kinematic model sought is x, y and θ then the equation can be searched what value 𝑣𝑥 𝑡 , 𝑣𝑦 𝑡 , 𝜔 𝑡 of the matrix equation above. Having discovered how much value of 𝑣𝑥 𝑡 , 𝑣𝑦 𝑡 , 𝜔 𝑡 can be calculated how much the value of x, y and θ it.

Figure 2 is a picture of the location of the sensors and the possible directions that will be generated by the robot. forwar maj du

Depa left Front n kiri

2

Front right Dep an kan 3 an

left

4

1

3. Description of the mobile robot system System had been made consist of mechanical system, hardware and control system. The mechanical system consist of base part that used for base of robot, used for wheel place and motor, there are place for sensor around the robot. Hardware consist of electricity circuit for sensor and actuator. Sensor that had been used is infrared sensor SHARP GP2D12, while the actuator are the motor of robot.

6

right Ges er kan an

5

Belaka Back left ng kiri

Back Belaka right ng backward mund kanan Figure 2 sensor placement ur and robot move direction

Description: the number is showing the location of the sensor

The possible directions of robot motion according to the active sensor in robot: 1. If the sensor is active (there is attention to the hands range) number 5 and 6 then the robot will move forward. 2. If the sensor is active (there range attention with their hands) No. 1 and 6 then move the robot will move toward the front right 3. If the sensor is active (there is attention to the hands range) number 1,2 and 6 then the robot will move to right. 4. If the sensor is active (there range attention with their hands) No. 1 and 2 then the robot will move toward the right rear 5. If the sensor is active (there is attention to the hands range) number 2 and 3 then the robot will move backward. 6. If the sensor is active (there is attention to the hands range) number 3 and 4 then the robot will move toward the left rear 7. If the sensor is active (there is attention to the hands range) number 3.4 and 5, the robot will shift to the left. 8. If the sensor is active (there is attention to the hands range) number 4 and 5, the robot will move toward the left front 2.2 Block diagram of the system Sensor input

Sensor network

Internal model

Motor output

Fuzzy rule

Human evaluation

Sensor input human Human

Figure 3 block diagram system

Block diagram of the system can be seen in Figure 3 can be seen clearly how the system works. Input gave from the sensor distance and obstacle which will enter the network to enable the sensors, sensor networks will be processed by the fuzzy rule to be able to activate the motor. From this system a sensor network has the task to translate the data from the sensors to grade members of the linguistic variables for fuzzy controller. Internal model is

composed of internal criteria for evaluating the condition of the robot. Based on the results of the evaluation, the robot produces reiforcement model of learning with the help of humans. The model of learning here is using the aid of human hands procession on certain sensors that have been determined, so it has been known to have some movement that can be done by robots. Therefore, the robot translate, symbolic movements made by human hands as a learning model. The more rare the robot receives procession tobot the hands of the more clever. 3. Fuzzy Logic System Fuzzy logic has been widely used for navigation problems in mobile robotics because somepowerful features. First, fuzzy logic control can easily represent the environment surrounding the robot with the rules' if-then' her. Second, robot navigation in unstructured environments orenvironments that are not visible, most require non liniear system, quickly and dynamically to map sensor values to the movement of the robot (YukiOno, 2003). Fuzzy logic has been widely applied in many aspects of robot navigation, collision avoidance, avoid the obstacle is the most important part that must be controlled in a mobile robot. Robot capable of moving autonomous in unstructured environments must know how to make him self ableto walk safely. Fuzzy Logic Control is one form of intelligent control . This control is often used to solve navigation problems. In this project, the fuzzy logic system, which is considered as a crisp input is the distance between the robot with the owner's hand. While the crisp output of fuzzy logic is the car speed and direction of motor rotation. Crisp input range with an obstacle the robot hand was designed in the 8 directions of motion moile robot. Eight direction is forward, forward right, slide right, rear right, back, back left, slide left, front left. Membership function with the distance of obstacles by the robot has five labels, ie VD (very dangerous), D (dangerous), N (normal), S (safe) and VS (very safe). Membership function for input can be seen in figure 4.

Table 1 Fuzzy Rule that had been made `

sensor

direction

1

2

3

4

5

6

m1

m2

m3

vs

vs

vs

vs

vd

vd

z

pf

nf

a1

b1

c1

d1

e5

f5

vs

vs

vs

vs

n

n

z

ps

ns

a1

b1

c1

d1

e3

f3

vd

vs

vs

vs

vs

vd

nf

pf

z

a5

b1

c1

d1

e1

f5

n

vs

vs

vs

vs

n

ns

ps

z

a3

b1

c1

d1

e1

f3

vd

vd

vs

vs

vs

vd

nf

pf

pf

a5

b5

c1

d1

e1

f5

n

n

vs

vs

vs

n

nf

pf

pf

a3

b3

c1

d1

e1

f3

vs

vs

vs

vd

vd

vs

pf

z

nf

a1

b1

c1

d5

e5

f1

vs

vs

vs

n

n

vs

ps

z

pf

a1

b1

c1

d3

e3

f1

vs

vs

vd

vd

vd

vs

pf

ns

ns

a1

b1

c5

d5

e5

f1

vs

vs

n

n

n

vs

pf

ns

ns

a1

b1

c3

d3

e3

f1

vd

vd

vs

vs

vs

vs

nf

z

pf

a5

b5

c1

d1

e1

f1

n

n

vs

vs

vs

vs

ns

z

ps

a3

b3

c1

d1

e1

f1

vs

vs

vd

vd

vs

vs

pf

nf

z

a1

b1

c5

d5

e1

f1

vs

vs

n

n

vs

vs

ps

ns

z

a1

b1

c3

d3

e1

f1

vs

vd

vd

vs

vs

vs

z

nf

pf

a1

b5

c5

d1

e1

f1

vs

n

n

vs

vs

vs

z

ns

ps

a1

b3

c3

d1

e1

f1

forward

motor

Figure 4 Membership function of intput

While the membership function for output also consists of five labels, ie NF (negative fast), NS (negative slow), Z (zero), PS (positive slow) and PF (positive fast). Membership function with a negative label member opposed to the sense of motor rotation clockwise. Membership function for the output can be seen in Figure 5.

Front right

right

Front left

Figure 5 Membership function of output

Membership function of inputs and outputs that have been made it will be rules which are used as rule by the robot in taking action in accordance with the insert provided. Table 1 shows the design rule that has been made. From that table we can see linguistic rules that used to control the movement of motor 1, 2 and 3. The rule can not be directly used, because it must be calculated the value of degrees of membership function of each input, then compared and selected values of the smallest degree of membership function.

left

Back right

Back left

backward

4. Sensor Network

6. Experimental result

Sensor network is a set of sensors or actuators can also be a resulting feedback on the overall operation in accordance with information obtained from sensors or actuators are mounted. Sensor networks are commonly used to aid navigation on a mobile robot so the robot can walk with a "safe" according to information obtained from the sensor network. In this project used the Fuzzy control method. If there are many obstacles in an environment, the robot should move slowly towards the target to avoid a collision. Another case, if there are few obstacles, the robot can move more easily toward the target without having to move slowly. 5. Reinforcement

learning

Reinforcement learning can generally be categorized as one of supervised learning methods (Nayouki Kubota et al, .1999). With this learning, the robot must be able to be prosecuted either in the sense of obeying rules which have been previously applied, if the robot violates, then the robot will get the punishment/penalty. Reinforcement learning is built with the help of teaching human models (human evaluation). Robots that the algorithm has been built, trained or given the procession toward learning by hands on this final task. Learning is used robots to be able to reach the point of destination in accordance with the expected trajectory without the assistance of the procession hands. With continuously trained robot is expected to be by itself can reach the destination point by taking the trajectory as expected. The more rare the robot receives lessons from humans to determine the direction to be addressed, then the robot will be getting smarter. But if otherwise, the robot must be trained in order to have more frequent according to the desired results. The desired outcome is that the robot can walk reaches the destination point to reach the destination point with the expected trajectory. The results are important, but the learning process and how many times the robot should be learning in order to get the desired result is the most important thing.

6.1Comparison of robot movement on different terrains Tests conducted on the robot in the two conditions. The first condition, the robot walks on a carpet. The second condition, the robot walking on sponge. Apparently different results obtained. When the robot walking on carpet, spinning wheel of the robot is too fast that sometimes causes slip. When the robot walking on a sponge, the spinning wheel of the robot slower when the robot walking in carpet. When walking on a sponge, slip does not happend on the robot. A carpet for robots, maybe it is still felt slippery compared with a sponge base. This is influenced by the omni-wheel which is not coated with rubber, so that the robot can not walk on slippery places. Figure 6 are the snapshot of robot movement in carpet and in a sponge. Movement comparison of the robot while walking on a carpet and sponge can be seen in Table 2.

Figure 6 snapshot of robot movement Table 2 Comparison of robot movement

spinning wheel Response movement Slip occurs or not? Reason slip?

Carpet faster

Sponge slower

Quick

slower

sometimes yes It was too slippery

not -

6.2 The relationship of sensor data with direction and motor speed This section will explain the data of the sensor affects the directionand speed of motor rotation. Table 3 Relationship of sensor data with the direction and motor speed Sensor data Directi on Forwa rd Front right Right Back right Backw ard Back left Left Front left

velocity

S1

S2

S3

S4

S5

S6

M1

M2