developing a real time collision avoid-ant low cost manipulator using ROS

Developing a Real Time Collison Avoidant Low Cost Mobile Manipulator Using ROS

A MASTER’S THESIS

submitted in the partial fulfillment

of the requirements for the award of the degree

of

MASTER OF TECHNOLOGY

in

INFORMATION TECHNOLOGY

(M.Tech in IT specialization Robotics)

Submitted by

Rajan Kumar Singh

Iro2011014

Under the Guidance of:

Prof. G.C. Nandi

Dean Academics IIIT-Allahabad

INDIAN INSTITUTE OF INFORMATION

TECHNOLOGY ALLAHABAD – 211 012 (INDIA)

July, 2013

CANDIDATE’S DECLARATION

I do hereby declare that the work presented in this thesis entitled

“Developing a Real Time Collision Avoidant Low Cost Mobile Manipulator

Using ROS”, submitted in the partial fulfillment of the degree of Masters of

Technology (M.Tech), in Information Technology at Indian Institute of Information

Technology, Allahabad, is an authentic record of my original work carried out under

the guidance of Prof. G.C. Nandi due acknowledgements have been made in the

text of the thesis to all other material used. This thesis work was done in full

compliance with the requirements and constraints of the prescribed curriculum.

Place: Allahabad Rajan Kumar Singh

Date: R. No. IRO2011014

CERTIFICATE FROM SUPERVISOR

Date:

I do hereby recommend that the thesis work prepared under my supervision by Mr.

Rajan Kumar Singh titled “Developing a Real Time Collision Avoidant Low

Cost Mobile Manipulator Using ROS” be accepted in the partial fulfillment of the

requirements of the degree of Master of Technology in Information Technology for

Examination

Date: Pro. G.C. Nandi

Place: Allahabad DEAN(Academics), IIITA

P a g e | 1

Indian Institute of Information Technology | Allahabad

CERTIFICATE OF APPROVAL

The forgoing thesis is hereby approved as a credible study in the area of Information

Technology Engineering and its allied areas carried out and presented in a manner

satisfactory to warrant its acceptance as a prerequisite to the degree for which it has been

submitted. It is understood that by this approval the undersigned do not

necessarily endorse or approve any statement made, opinion expressed or conclusion

drawn therein but approve the thesis only for the purpose for which it is submitted.

Signature & Name of the Committee members On final examination and approval of the

thesis

P a g e | 2


ACKNOWLEDGEMENTS

I would like to thanks my professors Prof. GC Nandi, Dr. Pavan Chokroborti who guided me

in right direction to complete my thesis. I would like special thanks to Mr. Abhijit Makhal who

really helped me a lot in my thesis. I would also like to thank Mr. Avinash Singh for his

support. I would like to thanks My Parents for their blessings, financial support, motivation.

And finally thanks to all My Friends for their co-operation and motivation. And finally thanks to

all my juniors for their support and love.

Place: Allahabad Rajan Kumar Singh

Date: M Tech Final Year, IIITA

P a g e | 3


ABSTRACT

Many researchers and robotic scientist want to launch the robots in our daily life. As we also know that

many manipulators are used in so many places like industries but their use are restricted due to their

complex environmental condition and their application. Those manipulators have no intelligence to work

in different environment and for various application. So our aim is to develop a such autonomous mobile

manipulator which can work in every complex environmental condition and can be used for various

application like table top operation, in kitchen environment, in office environment, in house environment,

in industries, in hazardous place also.

In our thesis we have tried to build an autonomous robotic arm which can work in those given complex

environmental condition and can be used for various application. Basically the whole thesis can be

categories in three parts 1st one is perception, 2

nd one is planning, 3

rd one is actuation.

In perception we use Microsoft kinect which uses the IR to capture the depth image which can be further

converted to laser scan by using pointcloud_to_laserscan package. By perception ros has full information

about environment and initial position and goal position.

In planning part ros uses OMPL (open motion planning library) to plan the trajectory. Kinematics is

solved for our arm by using atom_arm_kinematics package which is generated by using the planning

description configuration wizard for our urdf (unified robot description format) file of our robot.

After planning the trajectory by planner robotic arm actuates after receiving the joint_trajectory_msgs.

Our arm is 4 DOF arm which uses 4 AX-12 dynamixel servo motors which have been interfaced by using

usb2dynamixel device which continuously publishes its states and receive joint_trajectory_msgs.

Transform of the arm is maintained by using package named tf.

So that’s the whole plan of this thesis to create this robotic arm which can be further more helpful to

many young researchers and scientist to do their research.

P a g e | 4


Table of Contents :

1. Introduction……………………………………………………………………………7

1.1 Currently existing technologies………………………………………………..7

1.2 Analysis of previous research in this area……………………………………..9

1.3 Problem definition and scope.…………………………………………………15

1.4 Formulation of the present problem……………………………………………16

1.5 Organization of the thesis……………………………………………………..17

2. Description of Hardware and Software Used………………………………………17

2.1 Hardware………………………………………………………………………17

2.1.1 AX-12 dynamixel servo motor………………………………...17

2.1.2 Controller……………………………………………………….20

2.1.3 Perception and sensor…………………………………………..21

2.1.4 Phidget I/O board………………………………………………24

2.1.5 Power supply……………………………………………….…..24

2.2 Software………………………………………………………………….…….26

2.2.1 PCL (point cloud library)………………………………………26

2.2.2 Arbotrix controller………………………………………….….27

2.2.3 TF (transform)……………………………………. …………..28

2.2.4 RVIZ…………………………………………………..………29

3. Theoretical Tools – Analysis and Development……………………………………30

3.1 Forward kinematics ………………………………………………..……...…..30

3.2 Inverse kinematics…..………............................................................................33

3.3 Jacobean……………………………………………………………….…...…..36

4. System architecture….. ………….………………………………………...................37

4.1 Perception pipeline…….…………………………………………………….…38

4.1.1 Sensor I/P…………………………………………………...….38

4.1.2 Noisy point cloud dealing………………………………….…..39

4.1.3 Construction of collision environment……………………..…..41

4.2 Motion planning algorithm………………………………………………..……42

4.2.1 Sample based motion planning……………………………..…..42

4.2.2 CHOMP…………………………………………………….…..43

P a g e | 5


4.3 Motion execution monitor……….…………………………………………..…43

5. Software hardware development………………………………………………….....47

5.1 Hardware development…………………………………………………...….48

5.1.1 base development…………………………………………..…..48

5.1.2 head development……………………………………………...50

5.1.3 arm development…………………………………………….…51

5.1.4 Gripper development………………………………………...…54

5.2 Software development………………………………………………………....55

5.2.1 URDF development………………………………………..…..56

5.2.2 Generation of packages for our robot……………………….….59

5.2.2.1 joint state publisher/robot state publisher……….….61

5.2.2.2 dynamixel follow controller……………………..…62

6. Testing and Analysis…………………………………………………………………..62

7. Conclusion ……………………………………………………………………………..67

8. Recommendations and Future Work…………………………………………..…….68

References…………………………………………………………………………..…..69

P a g e | 6


List of Figures:

Figure

no.

Name of figure Page

no.

1 Asimo robot 7

2 Pr2 robot 9

3 Hardware of pr2 robot 10

4 Pr2 7 DOF arm 11

5 Pr2 mobility 12

6 Stereo with texture projector and 5 mp camera 14

7 AX-12 A dynamixel motor 17

8 Usb2dynamixel controller 19

9 Top view of the CM-5 20

10 Hokuyo UTM-30 LX laser range finder 21

11 Microsoft kinect 22

12 Phidget i/o board 23

13 Li-ion battery 24

14 Usb power phidget i/o board 25

15 Point cloud of the objects and environment 26

16 Transform of pr2 28

17 Interactive marker control of our robot 29

18 Co-ordinate frameattach to each link 32

19 Multiple solutions in inverse kinematics 34

20 Figure of Atan2 function 34

21 Block diagram of our system architecture 36

22 Point cloud of the environment 38

23 Filtering in PCL 40

24 Segmentation of the objects on table 41

25 Collision map of the environment 41

26 Motion planning and execution 44

27 Description of all parts of our robot (ATOM) 49

28 Head joints 50

29 Initial development of robotic arm 51

30 Dual motor robotic arm 53

31 Two sided gripper control 54

32 Single sided gripper control 55

33 Urdf tree structure 56

34 1st developed URDF 57

35 Dual motor URDF using meshes 58

36 Our latest URDF developed 59

37 Package generation of our robot model using planning description

configuration wizard 60

38 Demonstration of synchronization of URDF and real robot 62

39 Block manipulation by our robot 63

40 Collision free trajectory generation using planning scene warehouse viewer 65

41 Torque analysis of all joints during block manipulation 66

P a g e | 7


Chapter: 1

1. Introduction:

1.1. Current existing technology:

As we know the importance of robot in our daily life. Every roboticist has desire to make such a

social robot which used in our daily life work. Some robots are created to use in home work like

collecting and cleaning plates in kitchen environments and some are created to use in industries

like object pick and place that is manipulation and some are created to use in hazardous place

like work in atomic reactor where very high radiation level exist. So overall it is very vital role of

robots in our daily life. I guess in coming few year we are much dependent on robots (social and

industrial both). Some robots are already created which have capabilities to do work like if I say

“make an omelet for me” if it know the method to make the omelet then ok it will cook for but it

do not know the method to cook the omelet the it will go to internet search the method to cook

the omelet and download the process and then follow them to finish his work. So we have

already developed such type of robot which can be used in our daily life.

In last few years many roboticist has spent their life to create such type of great robot which

help us in our daily life under some circumstances. A few year back Japanese scientist filled with

astonishment everyone by creating the beautiful biped robot series named ASIMO. This robot is

biped robot which uses two legs to walk and great body structure. Due to biped this robot is more

like human and different from other robots which use wheel to move. ASIMO is a biped robot

designed by HONDA in 2000.

Fig1: ASIMO ROBOT

P a g e | 8


As we know that biped locomotion is a very researchful topic for robotics engineers. A lot of

robotics scientists are doing their research on the biped locomotion. In the biped locomotion

push recovery is the latest challenging problem on which all over the world every university and

research organization are focusing. Some of my friends are also doing research on the push

recovery topic in my institute.

Here we are trying to create a robot (mobile manipulator robot) and I am doing my research on

the robotic arm (autonomous robotic manipulator) for this robot platform named ATOM

(Autonomous Testbed of Mobile Manipulation). This robot platform ATOM is ROS (Robot

Operating System) based robot.

ROS (robot operating system) is a software framework initially designed by Stanford University

in 2007. After that development continues by the Willow Garage in 2008. ROS provides the

platforms, libraries, drivers, visualize etc. We can make our robot using ROS very easily. It is

ubuntu compatible software, currently it is being developed for the windows by the Willow

Garage.

There is so many robot which are using the ROS (robot operating system). Some of them are

very costly common people can’t afford them. Some of them having little bit less cost, but here

we made our robot ATOM in very low price so that common people can afford this robot for

their different area( for home work, office work, and for education work). Previous ROS based

robots are given below.

1.2. Analysis of previous research in this area:

Every robotics engineer’s dream to build such type of autonomous robot which can interact

naturally to the human being, which can perform object manipulation and help in daily life of

human being, which can wander in any cluttered environment without getting collision

autonomously.

These all dreams became true when Willow Garage launched the ROS based robot named PR2

(personal robot 2). But cost of this robot (PR2) is very high approximately $400,000. Common

people even most of the robotics engineers cannot afford this robot. Only 50 pr2 robots are in the

P a g e | 9


world and Willow Garage. Willow Garage was initially founded by two people by recruiting

scientist from Stanford university to accomplish the project solar powered boat in open ocean in

fall of 2006. After this in 2007 Willow Garage released the hardware and open source software

called ROS (robot operating system). They 1st release the pr2 robot which can manipulate objects

autonomously and can wander in any unstructured and cluttered environment autonomously.

PR2 robot:

Many robotics engineers’ dream to make an autonomous robot became true when Willow

Garage launched pr2 (personal robot 2) into market. Initially its cost was very high, common

people even many researchers can’t afford this robot but after some time its cost reduces to

$400,000 now many research institute have this robot for the research work.

Fig2. pr2 robot

This pr2 robot project was funded by NSF: Major Research Instrumentation Program

(MRI), Small Business Innovation Research (SBIR), NSF: Grant Opportunities for

Academic Liaison with Industry (GOALI), Defense University Research Instrumentation

Program (DURIP), DOD Multidisciplinary University Research Initiative (MURI).

Pr2 robot can be categorized into many different modules. All these modules are described

below.

P a g e | 10


1 Hardware and Software:

Willow Garage 1st success was to build the PR2 robot. This is two armed and wheeled

autonomous robot which can be used for so many purposes like education (Test bed for

research), in the home environment, in office environment, in kitchen environment etc. It has two

autonomous arm which will be described in next module. It physical size is about the size of the

human.

ROS (robot operating system) is the Software part of the pr2 robot. ROS is a open source

software developed by Willow Garage. In ROS you will get so many libraries which can be used

in creation of the robot. If you want to work on pr2 robot and you don’t have the pr2 robot then

don’t worry about that because they have created the one open source simulator named gazebo

and open source visualizing software named rviz where you can do your research, where you can

indulge yourself to develop the field of robotics. ROS gives the platform and the libraries and

tools to create the advance robotics application. You can find the research and core libraries on

which they are working currently, from there you will always updated what is going on, what is

the current new challenging problem we are facing, on which you can work you can do your

research and submit your research by making the reporjitary of your package. From where

anyone can download your package and test it and further can research on it. So Willow Garage

is trying to create the community from all over the world of researchers.

Fig3. hardware of the pr2 robot

P a g e | 11


2 Manipulation:

It is the 2nd

module of the pr2 robot. Pr2 arm is the 7 DOF (degree of freedom) autonomous

robotic arm. Pr2 arm is backdrivable and current control also hence it can manipulate in any

cluttered and unstructured environment.

It has total 7 DOF arm, in which 2 DOF in the shoulder (shoulder_lift and shoulder_pan), 2 DOF

in the elbow (elbow_lift and elbow_roll) and finally 3 DOF in the wrist (yaw, roll, pitch). It has

two sided gripper hence they have used total 9 servo motors to create one such robotic arm.

When we give the goal position to place the object to the pr2, 1st it do the perception by using the

tilted laser scan it creates the collision map using the library pcl (point cloud library). By using

pcl it will get the evry information about the environment about the obstacles, about the object.

Such a way it will get the initial and goal position by perception and then its planner starts to

plan the trajectory without getting collision with any obstacles in the environment.

Fig.4 pr2 7 DOF arm

P a g e | 12


3 Mobility:

This is the 3rd

module of the pr2 robot. Movement of the robot in any cluttered and unstructured

environment is also a challenging problem. This robot (pr2) is a wheeled robot with navigation

capability. It can navigate the optimal path in the environment. After the 3d scan of the

environment by using tilted laser scan it will create the collision map of the environment such a

way it will get the information (such as it recognizes the object and get the co-ordinate the

obstacles) about all the obstacles present in the environment, then its planning section plans the

trajectory without getting collision. It use the SLAM (simultaneous localization and mapping)

for navigation.

Fig.5 pr2 mobility

4 Sensors used in pr2

Sensors have the important role in the robotics field, pr2 uses many sensors for taking inputs

from the environment, for grasping, for perception. Manipulation stereo with Texture Projector

and environment stereo and 5MP camera are used in the pr2 for visualization of the target object

and the obstacles present in the environment. One base planner laser scanner for the localization

P a g e | 13


and mapping in the environment. When we give command to pick the object put it to the goal

position sometimes when pr2 moves it arm to grasp the object some area occluded by the arm it

can’t see the object anymore it uses one fore arm camera for avoid the occlusion. Its gripper uses

the pressure sensor array, this sensor array sends feedback continuously and increases the

pressure so that it can hold the object without damaging, means by using pressure sensor array it

try to hold the object without slipping and damaging the object by applying the exact pressure to

hold. And finally one tilting planner laser scanner to create the collision map using point cloud.

P a g e | 14


Fig. 6 (a) Stereo with texture projector and 5MP camera (b) base laser scanner (c) fore arm camera (d)

gripper pressure sensor array (e) tilting laser scanner

1.3 Problem Definition and scope:

As we know many manipulators or robotic arm are available in the market which can be bought

and use but main problem with that their cost is too high, a common people can’t afford to buy

that manipulator. Their cost are approximately million dollar so its impossible to buy those high

cost manipulator for their respective work or research.

My main aim to build a such manipulator or robotic arm which has similar functioning like those

robotic arm and cost is in range of a common researchers who can buy this robotic arm for their

various application or their further research.

Cost of this robotic arm is approximately INR 30000/- nearly $ 450, so cost is so less.

Researchers can build their own robotics arm by using our method and can do their research on it

for further enhancement.

We are using dynamixel servo motors for building this autonomous robotic arm. The cost of the

dynamixel motor is approximately INR 2500/ per motor, and one usb to dynamixel device to

connect the dynamixel motor with system. Very high configuration system is needed which cost

is not added in the cost of the robotic arm. For gripper we use pressure sensor array which are

P a g e | 15


used to calculate the total pressure implied on the object. And for perception we are using

microsoft kinect which cost is approximately INR 11000/. Other manipulators are using high

cost perception device named LASER RANGE FINDER which has the same function like

Microsoft kinect. The cost of the laser range finder is very high which is approximately $ 3800

(INR 200,000) . One highly regulated 12 volt DC power supply to run the servo motors.

Software used in this project is totally free which are open source software named ROS (robot

operating system). This open source software is developed by Willow Garage.

So overall my approach to keep the minimum cost for building a such autonomous robotic arm.

We are using this method for creating the robotic arm in minimum cost and having same

function to that of high cost manipulator have. It is also as accurate as high cost manipulator.

Scope of this project is anyone can build their own autonomous robotic arm for various

application by following this project’s method. Here anyone can do their research on this

platform related to table top operation to pick and place the small objects on the table. Any

researcher can work on this platform to serve the water glass or wine glass. It can be used in fine

manipulation also if the hardware are of the high cost which could give very précis zed result.

Since it has searching and path planning algorithm by which it can play the chess.

So overall the cost is as minimum as possible and it is as accurate as the high cost manipulator.

Its cost is very minimum as compared to the other high cost autonomous robotic arm and it has

the similar function like those high cost manipulator. I have published one journal paper on basis

of the literature survey , in this paper description about all other high cost manipulators are given

also compared with my robot . It has various scope in the field of manipulation.

One more scope of this robot in the field of surgery of a patient. When doctor is not available in

hospital this robot can follow the doctor and can do the surgery. It’s a very fine manipulation so

for this purpose we need more advance servo motors with great feedback for precise actuation.

The most important thing in this application is getting highly calibrated perception, for this

purpose we need LASER RANGE FINDER which cost is very high. We have to model our

URDF (unified robot description format), urdf is the just simulating model of our robot which is

interfaced with our real robot and synchronize with our real robot by using node

robot_state_publisher and joint_state_publisher. Both urdf and real robot will synchronize with

these two node urdf continuously publish its state in the ros messages and then real robot will

P a g e | 16


receive the same state of the urdf and follow it. Urdf can be controlled by hand gesture

movement so doctor can see the patient of live camera and can move his hand this movement can

be captured by urdf it publish the robot state and the same time real robot will receive the robot

state and follow the doctor’s hand movement.

1.4 Formulation of the present problem:

This is the very important thing before start to build our robotic arm. We should know about

constraints, about our environment and application. Then we have to formulate our strategy to

build our proper robot which will operate in given constraint and given environment.

Firstly we have to modeled our robot description format means urdf, this is the main thing which

need to build properly, we should build our urdf very precisely. We should keep in our mind

what approach should be applied.

I’m developing a low cost mobile manipulator for pick and place application, for that application

I built the 4 degree of freedom robotic arm using 4 dynamixel servo motor. As in high cost

manipulator high cost servo motors are used. Perception is the main step by which it can percept

the environment by using Microsoft kinect.

1.5 Organization of the thesis:

I got motivation by SMART (social mobile autonomous robot for testbed) which is developed in

ROBOTICS AND AI LAB IIIT ALLAHABAD. Arm of SMART is not autonomous that’s why I

chose to develop an autonomous robotic arm for various application. I developed this robotic

arm in ROBOTICS AND AI LAB Indian Institute of Information Technology Allahabad.

Robotics and AI lab of IIIT Allahabad is a very advance lab for robotics and AI, many work on

autonomous and artificial intelligence is going on. Every facility is available in this lab for doing

our research.

P a g e | 17


Chapter: 2

2. Description of hardware and software used

As we know we are building an autonomous robotic arm. For building this autonomous arm we have to

take care of both hardware and software aspects. First we have to build our hardware part and it

should interfaced with the software ROS (robot operating system). After developing the

hardware we should take care of software part. Before developing the software part we should

know whole idea of our application as well as the environment condition. If you have the

knowledge of everything related to you project then you can only develop your software part.

Our thesis can be categories in two part basically, first one is hardware and second one is

software. Further description about hardware and software are given below.

2.1 Hardware:

For building the hardware part of my thesis, so many hardware are used which are listed below.

2.1.1 AX-12 Dynamixel Servo Motor:

Servo name is basically came from term servant, servo motors have their own controller. It acts

like servant and follow the instruction of its controller like servant that’s why we call it servo

motor. At many places this AX-12 dynamixel servo motors are used to make humanoid robot

also.

It is the one of the main part of hardware. It is actually used as actuator. AX-12 servo motor is

the most advance servo available in the market, which can be used by many researchers and

scientists for building their own robot.

For building of our project by this dynamixel servo motor we have to connect these motors in

series. First motor is connected to computer and next one is connected to previous motor. Every

motor has two port one is used to connect with previous servo and one is used to connect with

P a g e | 18


next motor. And every port has three pins one pin is used for 12 volt, one is used for ground and

3rd

one is used for data.

Fig7 AX-12A dynamixel motor

We know every electronic device operate under some circumstances for proper operation. So

every device have some stats under which that device will give the better result. So AX-12

dynamixel motor’s stats are given below:

Operating voltage for this servo motors are 9-12 volt DC (recommended

11.1volt).

It can offer the torque of 15.3 Kg.cm or 212 oz.inch.

No load speed for this servo motor is 59 RPM or .169 sec/60 degree.

Weight of the AX-12 servo motor is 55Gram.

Dimension of this servo motor is 32*50*40 mm.

Minimum angle it can rotate i.e. resolution of the motor is .29 degree.

Reduction ratio is given 1/254.

Operating angle for AX-12 motor is 300 degree or we can say continuous turn.

It can have the maximum current of 900mA.

Operating temperature range of this servo motor is -5 to 85 degree Celsius.

P a g e | 19


2.1.2 Controller:

Controllers are used to connect the dyna mixel servo motors with computer directly. All servo

motors are added serially to controllers. Many types of controllers are available in the market

such as CM-2, CM-2+, CM-5, CM-510, usb2dynamixel etc. All are connected to the usb port of

the personal computer. Basically these controllers convert the usb to serial port so easily we can

get serial data from usb port by using this device.

Fig.8 usb2dynamixel controller

Above shown in figure is the usb2dynamixel controller. We have used this controller in our

project. This usb2dynamixel device has Status Display LED, Serial Connector, 3-pin connector,

4-pin connector and Function selection Switch. Status Display LED is used to indicate the power

supply, TXD (transmitting data), RXD (receiving data) status. Function selection switch is used

P a g e | 20


to select the communication method like whether it is TTL or RS-485 or RS-232. AX-series

motor uses 3-pin connector to connect with this usb2dynamixel controller using TTL

communication method. 4-pin connector uses RS-485 communication method to connect the

DX-series and RX-series servo motors. And finally serial connector uses RS-232 communication

method, it change the usb port to serial port.

As below shown in figure CM-5 is another way to control the AX-12 dynamixel servo motor.

CM-5 can store 4 number of program at a time to execute. You can execute those program by

selecting the different modes and can perform different task by your robot.

As shown in figure power switch is used to keep ON/OFF and power jack is used to connect the

power to CM-5. Status LED indicates the status of CM-5 like whether its RXD (receiving data)

or

Fig.9 top view of cm-5

P a g e | 21


TXD (transmitting data) or charging mode. There are 4 direction buttons are in CM-5. Mode

button is used to change the mode of operation of the CM-5 and start button is used to select the

mode of operation. Play button is used to execute the program which is stored in the CM-5.

2.1.3 Perception and sensor:

Perception is the main part of the arm navigation. Many high cost robot like pr2 uses Laser

Range Finder as a perception device. But its cost is so high approximately INR 190,000, Which

can’t be affordable by many young researchers and scientists. We are using another perception

device which cost is very low as compare to Laser Range Finder. Microsoft launched a new

product named Microsoft kinect, which was initially developed for gaming purpose. But we are

using this device as a perception device in our robot.

Hokuyo UTM -30LX laser range finder is latest version of laser range finder. It has some

advance properties over other laser range finder. Its cost is so high approximately $5600. Its

angular resolution is .27 degree and scan range of 30 meter. It will take lesser power

approximately 8.4 watts at 12 volts. Its size is 6.0*6.0*8.5cm.

Fig.10 hokuyo UTM -30LX Laser Range Finder

P a g e | 22


One more perception device exist named PML (poor man’s lidar), which can be constructed in

your own lab. We tried to construct this PML in our lab by using laser projector, receiver and

one servo motor. Laser projector and receiver is mounted on the servo motor by using concept of

the laser range finder. PML’s cost is very low. But it didn’t give the better result that’s why we

didn’t use this device in robot.

As we know that Laser Range Finder gives 650 points in single 240 degree sweep, while PML

(poor man’s lidar) gives only 24 points in single 180 degree sweep.

We are using Microsoft kinect in our robot for perception . kinect is a Microsoft product which

is developed for gaming purpose. In kinect one IR projector, IR receiver, stereo camera and array

of microphones. IR projector projects the IR ray to the environment and provides the depth

image. Such a way we have depth image by this IR projector and receiver. Further depth image

can be converted to laser scan by using package pointcloud_to_lasercsan. This

pointcloud_to_laserscan package provides fake laser scan from point cloud by doing slice

horizontally of the image and taking the nearest distance in each column.

Fig.11 Microsoft kinect

P a g e | 23


As shown in figure Kinect has 3D- depth sensor which are IR transmitter and IR receiver. One

RGB camera and array of highly sensitive and noise suppressed microphone. It can be used in

speech recognition.

Pressure sensor we are using at palm and thumb of the robotic arm for the grasping purpose. We

can train our robot in such a way that it can apply suitable force for each individual objects so

that no objects get harm while grasping process is going on.

2.1.4 Phidget I/O boards:

Phidget I/O board is used to connect the pressure sensor which are used to calculate the exact

pressure on the objects by thumb and palm. This Phidget IO board is easily interfaced with ROS

(robot operating system)

Fig.12 snapshot of the phidget I/O board

Basically any analog sensor can be interfaced to the phidget i/o board like pressure sensor, gyro sensor,

rotating potentiometer etc. These all can be connected to the ros by using the phidget I/O board.

In the phidget i/o board as shown in figure separate 8 bit digital input and 8 bit digital output port is given

taking 8 bit digital out put by giving 8 bit digital in put. One port is given to connect the board to the

P a g e | 24


computer device. 7 separate ports are given to connect the 7 different sensors to the phidget for

taking input.

2.1.5 Power supply (battery and cable):

It’s the one of the main part of our project its very serious thing to supply the power to all

devices used in our projects. For different device we have different power rating so we need to

build the separate regulator circuits for each power rating.

2.1.5.1 Power supply for AX-12 dynamixel servo motors:

As we know that power rating for the dynamixel servo motor is 9 to 12 volt DC supply. It can

maximum have the 900 mA of current.

Fig.13 Li-ion battery

We using Li-ion battery for the main source of the power, which can give the 12 volt DC. From

this source we will provide supply to all.

2.1.5.2 Power supply for the kinect:

Normal power cord of the Microsoft kinect takes power from AC input. You can use AC power

source for static use of the means no movement of the Microsoft kinect. But it is not possible

for the mobile robot, as we know mobile robot moves in whole environment we have to give

the power source to the Microsoft kinect from the DC source.

2.1.5.3 Power supply for phidget i/o board:

P a g e | 25


Power supply is to given from the usb port from the CPU (central processing unit). No need to worry

about the power supply for the phidget I/O board.

Fig.14 usb power phidget I/O board

2.2 Software:

We are using ROS (robot operating system) ,which is a open source software developed by the

Willow Garage. Initially it was started in Stanford University in 2006, and then in 2007 willow

garage started to developed the ROS.

As I already told that ros is a open source software, where you can do your research in software

field in developing of your robot. Many open source libraries are available there, you can use

them without any problem. Some open source libraries are given below.

2.2.1 PCL (point cloud library):

For point cloud processing it is used the PCL (point cloud library). Point cloud library uses many

algorithms which are used in cloud processing. These algorithms include filtering, feature

extraction, segmentation, surface reconstruction, registration and model filtering.

Mainly PCL is used in perception part of our robot. As I already told that we are using Microsoft

kinect for perception. Microsoft kinect has IR (infra red) projector and receiver by which we get

depth image. By using pointcloud_to_laserscan package we convert the depth image into laser

scan. By using previous package we are building our collision map of our environment.

P a g e | 26


PCL library is has great importance in perception like object detection, object recognition etc.

which are describe below.

Fig.15 point cloud of the objects and environment

Object detection and recognition using PCL:

Object detection and recognition is a very important step in the perception. Object detection

means the perception device which used the PCL to detect the object in the environment.

Detection is based on mainly size. For example in turtlebot block detection they use 3 cm block

to detect and pick and place. You can edit in the launch file of the block detection package of the

turtle to detect our own block of different size. They also use the kinect for perception, kinect

projects the IR and where it finds the 3cm height, width and length of point cloud it will simply

show the block.

Recognition process is basically based on the shape of the object color of the object. Suppose a

robot has to recognize a bottle. 1st it will create the point cloud of that bottle, and then it will

match the shape and color of the point cloud of the bottle with the objects database created and

saved in our computer. It will classify the given data and can have the knowledge about that

bottle. This process is known that object recognition.

In some case if it can’t find the any of the match in our own created database into our own PC,

then it will follow the internet and checks in the huge amount of the point cloud database of the

objects. This connection with inter to our robot is done using ROS package roboearth. This

roboearth is used to create the database of the objects upload these to the internet and connect the

P a g e | 27


robot to the internet. In such a way our robot can search in huge amount of database and

recognize the object.

After recognition the objects it has the prior knowledge about the object like how much pressure

should be apply to grasp without damaging the object. It has knowledge of the grasping point of

the object.

2.2.2 Arbotix:

This is a ROS package in which all the devices controller exists. It has the controller for the

dynamixel motor and for other sensor devices, messages for publishing, firmware. It has

following ROS packages.

3 Arbotix_controller

4 Arbotix_firmware

5 Arbotix_msgs

6 Arbotics_python

7 Arbotics_sensors

2.2.3 TF (transform package):

TF is ROS package using by robot which can track the multiple co-ordinate frames over time. TF

shows the relationship in between co-ordinate frames in a tree manner structure buffered in time,

transform points, vectors between two co-ordinate frames at every desired point at any desired

time. As you know that robot has many co-ordinate frames like world frame, base frame, arm

base frame, elbow frame, wrist frame, gripper frame etc. and we know also that each frame is

going to change over each moment of time. Then TF is basically used to track the co-ordinate

frames with respect to any other co-ordinate frame. So by using TF robot has full information

about any co-ordinate frame with respect to any co-ordinate.

P a g e | 28


Fig.16 transform of pr2

2.2.4 RVIZ (visualizer):

It is a 3 D visualization tool for ROS. It is a open source tool available for ROS. Mainly use of

rviz is to visualize the URDF (unified robot description format). Main application using rviz is

the interactive marker by which you can interact with your robot URDF. You can control your

robot joint by rotating and change the robot position and you can select some special menu

assigned to each marker.

Interactive marker server continuously update the rviz as you will see in your rviz to your robot

model while you interact with your urdf. Interactive marker continuously takes feedback from

the rviz for checking the status of the rviz.

P a g e | 29


Fig.17 Interactive marker control of our robot

As shown in figure you can see the interactive control of the right hand wrist joint which has 3

DOF and in figure 1 you can see that some special menu option in which you can select for IK,

plan the trajectory, trajectory filtering etc.

P a g e | 30


Chapter: 3

3 Theoretical tools – Analysis and develop

In this section we will discuss about theoretical analysis of the autonomous robotic arm. The

dynamics behind any manipulator is based on the three basic concept, these are given below.

3.1 Forward kinematics:

Forward kinematics is the relation between the joint space and the pose and the orientation of the

rigid robot (manipulator). Forward kinematics is used when joint parameter is given for any

manipulator we need to find out the pose goal of the end effector of the manipulator. We use

kinematics equations for this purpose.

As we know our robot autonomous robotic arm is serial chain of the motors, then to find out the

forward kinematics for our robot the transform matrix. These series chain of joint may be any

type like they are fixed joint or they are revolute joint or they are prismatic joint or they may be

more complex joint than the revolute and prismatic like ball and socket joint. As we know that

revolute joint rotate with respect to the single axis and the prismatic joint will provide the linear

movement along the axis. After finding out the transform matrix we will simply multiply the

joint parameter to the transform matrix the we will get the pose parameter or simply we can say

we have the position of the end effactor of our robot.

So forward kinematics is used to calculate the position of the end effector in Cartesian space

when given the joint angles in joint space.

Given parameter: The length of each link and angle of each joint

Target parameter: The position of the end effector (i.e its in (x, y, z) co-ordinate)

Now we are moving forward to calculate the transform matrix. For finding out the transform

matrix we have to use the DH (Denavit-Hartenberg) principle.

P a g e | 31


DH (Denavit-Hartenberg) principle:

Jacques Denavit and Richard Hartenberg introduce this principle in 1955 for standardize the

co-ordinate frames for spatial linkages. This principle is used in robotics specially for selecting

the frame of reference.

For the kinematic chain of joints and links we have separate co-ordinate for every link. So we

need a transform matrix for calculation of the position and orientation of the next link co-

ordinate with respect to the previous co-ordinate frame. In such a way we will use this method

for calculation of the position and orientation of the end effector of the robotic arm.

For analysis of the kinematics of robotic arm we have to attach the c0-ordinate frame for all link

like given in figure Oi (Xi, Yi, Zi).

We attached Oi (Xi, Yi, Zi) for the link i , the meaning of this is that when robot actuated all

points on the link i have the same co-ordinate frame i. The base co-ordinate frame O0 (X0,Y0,

Z0) which is attached to the base of the robot.

P a g e | 32


For calculation of the orientation and position next link like Oi (Xi, Yi, Zi) in reference of the

Oi-1 (Xi-1, Yi-1, Zi-1) that transformation known as homogenous transformation matrix. And

when we calculated homogeneous transformation matrix for each link, then finally we can

calculate the transformation matrix for the robotic arm in which joint parameter is multiplied

for getting the position and orientation of our end effector.

Fig.18 co-ordinate frame attached to each link

For DH parameter four parameter is used , , , which associated with the link i and

joint i. detail of this four parameter is given below.

joint angle, link length, link offset, link twist.

P a g e | 33


Homogeneous transformation matrix is single variable, so one out of four is only vary and rest

three will be constant for each joint. As we know that for revolute joint will vary and for

prismatic joint will vary.

3.2 Inverse kinematics:

As we already discussed that forward kinematics is used to map joint space Q to Cartesian space

W. But inverse kinematics is just the opposite of the forward kinematics it maps Cartesian space

W to joint space Q.

Equation for the inverse kinematics is given by following which maps Cartesian space to joint

space.

Inverse kinematics have very important role in robotics. Because normally input of any robot is

in Cartesian space (i.e. normally we define the initial position and goal position in Cartesian co-

ordinate, whenever we give the target location or initial location we can give only in Cartesian

co-ordinate system). When initial and goal position is given to the robot will solve for the angle

of each joint using inverse kinematics. After solving the inverse kinematics either it can give the

unique solution or it can provide the multiple solution. When it will get the unique solution the

P a g e | 34


normally it will execute the trajectory and follow the resulted joint angle, and when it will get

multiple solution first it will select the optimum solution in such a way that it can move in

environment without collision with other objects i.e. collision free path. Some solution from

multiple solution are not physically reliable.

Fig.19 multiple solutions in inverse kinematics

In inverse kinematics for calculation of angle we use ATan2 function. This ATan2 function is

more helpful in calculation of the joint angle from Cartesian co-ordinate.

Fig.20 figure of ATan2 function

P a g e | 35


ATan2 function is used in many places where co-ordinates are given and you have to find out the

angle. If you are given (3,4) co-ordinate in Cartesian co-ordinate you are asked to calculate the

angle. Then simply you will apply tan inverse function atan(y/x) i.e. atan(4/3).

But when you are given the co-ordinate (-3, -4) and again you are asked to calculate the angle, if

you apply same atan(-4/-3)=atan(4/3),and you will get same result. But both points are not

having the same angle, both are in different quadrant. (3,4) is in 1st quadrant and (-3,-4) is in 3

rd

quadrant.

That’s why we use atan2 function to discriminate between these two and this function will give

the accurate result.

3.3 Jacobean matrix:

Simply Jacobean is defined as the time derivative of the kinematics equation. Jacobean gives the

relation between the joint rate of the joint and linear and angular velocity of the end effector of

the robot.

P a g e | 36


Jacobean also gives the relation of the joint torque by joint and the force and torque produced by

the end effector.

For any given manipulator we are given the joint angles and the link length, we have to calculate

the linear velocity of the end effctor while angular joint rates are given. Or reverse can be

possible while joint rates are given and you have to calculate the linear velocity of the end

effector.

P a g e | 37


Chapter: 4

4 System Architecture:

For safe planning of the trajectory and execution of the trajectory we will go to the following

architecture to perform our manipulation. In this section we are going to describe all software

part step by step development.

Fig.21 block diagram of our system architecture

1. Perception pipeline

2. Motion planning algorithm

3. Motion execution monitor

These are the 5 step which you have to follow for designing the software part of the manipulator.

4.1 Perception pipeline:

P a g e | 38


Perception pipeline plays the key role of building a representation of world into a collision map.

Perception pipeline contributes basically two function:

(a) Perception pipeline combines the planner and give the problem to perform its

manipulation in cluttered atmosphere using data from the sensor.

(b) Perception pipeline is used in the field of handling occlusion and noisy data perfectly. It

is also used in noisy laser scan and it can deal with shadowing effects.

Perception pipeline is a complete package of the these three parts which are describe further.

4.1.1 Sensor input:

as I already told that I have used Microsoft kinect for perception purpose. The input data from

this data is in the form of point cloud. Kinect use IR sensor for scanning the environment then

data is transformed to laser scan by using pointcloud_to_laserscan package. Point cloud is just

the set of points in the space that corresponds to detect the objects. This sensor provides the input

in the form of point cloud of the 3D environment.

Fig.22 point cloud of the environment

P a g e | 39


4.1.2 Noisy point cloud dealing:

As we know that sensor point cloud data is often noisy that should be take care of it. When robot

is performing its manipulation its arm may be in front of the perception device and during the

point cloud generation the point cloud of the projected arm is also treated as obstacle. But this

projected part of the arm is not the obstacle of the environment. So we need to deal with these

noisy data.

So we have to separate these points from the sensor input points cloud. We will use a simple

method to separate these points i.e. point cloud of the projected part of the arm. “The system

have to check any part of the point cloud input data is the same as the point cloud of the

geometric shape of the projected part of the arm, then this portion of the point cloud will not be

treated as obstacle of the environment.”

One more problem generally occurs which is shadowing effect. It will occur when kinect grazes

some part of the body of our robot. Points seem by edges of the robot arm now seems to be away

and treated as part of the environment. When robot arm will move in the environment then these

veiling points seems to be on desired path of the robot arm, so our process will be haulted. “So to

remove these veiling points a small padding distance is summed up in the collision

representation of our robot links.”

P a g e | 40


These two problem is need to be arise which need to be filtered from the input point cloud.

4.1.2.1 Filtering:

To remove this shadows points which are coming in the point clouds in the environments we

used filters like voxel grid filter which are using to filter the shadow points as given in figure

Fig.23 filtering in PCL

As you can see in this figure shadow points which are present in the 1st image which are removed

after using the filter (voxel grid filter).

P a g e | 41


4.1.2.2 Segmentation:

After successful creation of point cloud of the environment, this environment point cloud is

unstructured so system will take so much time to process. So we need to segment these

unstructured point cloud data.

Segmentation is nothing but Euclidean Cluster Extraction. It uses several clustering methods like

Nearest Neighbor Estimator, RANSAC etc. to segment these objects.

Fig.24 segmentation of objects on table

4.1.3 Construction of collision environment or collision map:

Collision environment is similar to the collision map of the environment. A proper building of

collision map with frequent sensor updates can handle correctly the occluded data. We can

handle this by update the collision map with the next sensor update i.e. we have current collision

map and next collision map due to sensor update data then we can handle the occlusion by these

collision maps.

P a g e | 42


Fig.25 collision map of the environment

Suppose we have collision map C (initially empty) and we have new collision map due to sensor

update N. Then we will 1st calculate the difference D in between these two collision maps and we

can find the occluded data or removal data from the environment.

D = C – N

D in the collision map is either moving obstacle or part that is occluded by robot arm. For this

part D is occluded data or not , we will check that line segment in between d as a function of D

and the sensor origin intersects a body which is the part of the robot arm. If it happen then the

obstacle is said to be occluded and D will be added to N for complete data after removing the

occlusion.

Collision map is the very important input for the motion planning and path execution process.

For this our robot has restriction of a cube of size 2 meter forward, 1.5 meter in each side and 2

meter upword with respect to the base of the robot. This cube box has entire workspace when

arm can reach and do its manipulation.

4.2 Motion planning algorithm:

P a g e | 43


After successfully building the collision map the filtered collision environment data is transferred

to the motion planner. Motion planner is used to plan the trajectory between initial position and

goal position. Basically three types of planners are used for planning the trajectory.

1. Sampling based motion planning [using OMPL library]

2. Trajectory optimization techniques. CHOMP (Covariant Hamiltonian Optimization

For Motion Planning)

4.2.1 Sampling based motion planning:

Sampling based planning is generally takes the samples points in between the initial and goal

points taking care of obstacles, and then it will calculate the inverse kinematics for each points

thus trajectory will be generated. It gives the collision free trajectory very quickly.

This sampling based planner used a tree based algorithm from OMPL library. This planner used

following algorithm also:

(1) LBKPIECE- lazy bidirectional kinematic motion planning by interior-exterior

cell exploration.

(2) SBL- Single-query Bi-directional probabilistic roadmap planner with Lazy

collision checking

(3) KPIECE- Kinematic Motion Planning by Interior-Exterior Cell Exploration

(4) RRT- Rapid exploring Random Tree.

When robot receives a planning request then planner selects algorithm based on type of planning

request. If goal is available i.e. if goal is retrieved from the request then LBKPIECE or SBL are

used. And when goal is not given then KPIECE and RRT is used.

Planner selects the algorithm on basis of the priority, this priority will be either increased or

decreased if planner does not find the solutions. We know highest priority is of LBKPIECE and

lowest priority for the RRT.

4.2.2 CHOMP:

P a g e | 44


CHOMP stands for Covariant Hamiltonian Optimization for Motion Planning. It is a trajectory

optimization and it is based on the covariant gradient descent technique. It is also use to smooth

the path which are generated by sampling based planner. CHOMP is a cost function which is

summation of two cost function of the smoothness and cost function of the collision.

Smoothness cost function comes from summing the square of derivative of each joint along the

trajectory. And collision cost function comes from the Signed Distance Field. Collision cost is

in Cartesian space which can be transformed to joint space by using Jacobean matrix. For our

robot CHOMP optimizes trajectory in 1 to 4 seconds depends on the type of request.

4.3 Motion execution monitor:

Motion plans can be calculated for a robot in a collision map of an environment which is

continuously updated in the real time using perception pipeline and motion planning. This

motion planning is then sent to the trajectory control, which tries to follow them as perfect as

possible. In the motion execution monitor two motion planners are used:

(a) Long range planner

(b) Short range planner

Long range planner is used in sampling based planning in which it quickly generates the path

between initial point and the near to the goal point or to the goal point if goal point is not in

collision state. Short range planner is used in CHOMP based planning. It is apply when goal

position in collision or very near to the obstacles.

If any goal is in collision state or very close to the obstacles then 1st we execute the trajectory

from initial point to near point of the goal by sampling based motion planning quickly and from

near to the goal point to the goal point it will apply CHOMP based planning so that it can grasp

the object without getting any collision. CHOMP based planning is slow.

P a g e | 45


Fig.26 motion planning and execution

P a g e | 46


Above shown algorithm is used about working of motion planning and motion execution system.

This algorithm is used for finding the states in goal region in stead of single goal state. If a state

for which robot endeffector satisfies all condition and having no collision then long range

planner can be executed. If it does not find any valid goal state then it will apply genetic

algorithm to find a valid state which is as close to goal.

Due to proper updation of the environment data i.e. collision map updatation in each and every

moment of time it will check whether our robot arm is in collision state or not if robot arm in

collision state then it will terminate the execution and it will request for the trajectory replanning.

P a g e | 47


Chapter: 5

5 Software hardware development:

This chapter is actual work done by me. I have developed both hardware and software part. As

my thesis topic is Real Time Collision Avoidant Arm Navigation for a Low Cost Robotic Arm

Manipulator using ROS. Main aim of my thesis is to developing a low cost real time collision

avoidant robotic arm. A robotic arm which having the low cost so that many young researchers

and scientists who wants to work on the arm navigation field and do not having the sufficient

funding for their research. This thesis will surely much helpful for those young researchers and

scientists.

Our robotic arm can operate in any unstructured and cluttered environment. It can avoid any

obstacle presence in the environment in between the initial position and goal position. Mainly it

can be used in pick and place purpose. It can be used in table top manipulation also. Table top

manipulation like suppose I want a red pen, and so many different colors pen are kept at the

table. I will give the command to the robot to “give me the red pen”. Actually this command can

be given by either through speech or through the command. It will receive the information and

then it will 1st create the point cloud of the objects which are kept on the table. Then it will

search the red pen point cloud database in the objects data base for recognition of the red pen. If

it will succeed in finding the red pen point cloud database then its ok otherwise it will search on

internet for the database of the red pen point cloud. After searching the database of the red pen it

will match the input point cloud taken from the table and classify the data to recognize the red

pen. If this will recognize the red pen it has the co-ordinate of the grasping point and it has the

goal position to place their order. Then it will plan the trajectory between the initial position to

goal position avoiding the obstacles in between the initial and goal points. Then it will executes

the trajectory and actuators will actuate and perform proper operation. This is the overview of

my work what I am trying to do.

For developing this type of advance autonomous robotic arm we have to develop both the

hardware and software part. Now here I am going to describe step by step development of the

hardware and software.

P a g e | 48


5.1 Hardware development:

Hardware development is the initial step of the project. 1st we have to design to design the

hardware of the autonomous robotic arm.

For designing of the robotic arm we need a BIOLOID ROBO KIT. Basically BIOLOID kit is

used for simple humanoid robot, which can dance, can do various pose of the gymnastic, which

can salute you, which can run etc. You can do your different research on that robot. It BIOLOID

kit these hardware are available.

1. One Bioloid CM-5 mcu controller

2. 18 AX-12 dynamixel servo motors

3. 1 dynamixel AX-S1 sensor

4. 2 programming utility

5. 1 switch mode power supply

6. Serial cables

7. Comprehensive parts (F1,F2, F3, F4, F5, F6, F7, F8, F9, F10, F11, F12, F13, F14,

F15, F16, BU,WA, SP1, SP2, fuse, sticker, screws etc)

These all parts are given to you in the BIOLOID kit if you buy it. Its cost is $899.00, which are

little bit costlier. But you can afford it if not you can order only those things which you need this

will be cheaper.

Requirements:

For our projects we only need few things these are dynamixel servo motors, serial cables,

usb2dynamixel controller, 12 volt battery, Microsoft kinect and cables. One more thing we will

use a circuit for the power supply for Microsoft kinect from DC source. And for developing the

base of the robot we need steel plate, steel rod, wooden circular base etc.

5.1.1 Base development:

1st thing we have developed is the base for your robot at which all devices can be kept easily,

stably. So we need a strong stable base for fixing the robot body, autonomous robotic arm,

P a g e | 49


perception devices. For developing the robot base we use 4 wooden circular base. We use these

circular wooden plate are in a stack manner for creating the stable and mechanically strong.

After successfully creating the base we add two torso link that is upper torso link and lower torso

link. We fixed our autonomous robotic arm to the upper torso link. We add our Microsoft kinect

which is the perception device to the upper torso link. And the upper torso link is connected to

the lower torso link. And lower torso link is connected to the base of the robot.

Fig.27 description all part of our robot (ATOM)

P a g e | 50


As shown in figure you can see our full robot, in which two 4 DOF autonomous robotic arm.

And two Microsoft kinect one is used for the arm navigation and other is used for the mapping

and localization of the environment.

5.1.2 Head development of the robot:

This is the second step of the developing the robot, for development of the head of the robot 1st

fixed one dynamixel servo motor to the upper torso link. This dynamixel motor is used for the

pan joint by which robot can rotate its head. And then we use the another dynamixel servo motor

to the pan dynamixel servo motor. This dynamixel servo motor is used as providing tilt joint. By

using the tilt joint robot can tilt its head. So our head has two Degree of Freedom of pan joint and

tilt joint.

Fig.28 head joints

P a g e | 51


As you can see in the figure of head of our robot two dynamixel servo motors are used in such

manner that robot can move its head in pan direction and in tilt direction. And then we add a

steel plate for base of the stereo camera and for the Microsoft kinect. At the steel plate two stereo

vision HD camera we have fixed and Microsoft kinect is also fixed on the steel plate. For stable

fixing the Microsoft kinect we use steel strip to bound the kinect with the head steel platform.

5.1.3 Arm development:

Now here the main work is started to develop the perfect arm using the AX-12 dynamixel servo

motor. We have faced a lot of problem during developing of the arm. Firstly we designed the

arm using 5 dynamixel servo motor, single motor for each joint. You can see in the figure as

shown. You have to use the different motor IDs in whole arm and only one without ID motor

Fig.29 initial development of the robotic arm

P a g e | 52


Can be used in your whole robot. You know every servo motor has different motor IDs and some

of then have no ID when you are going to use those motor you will treat as ID 1. Initially we

have developed the 5 DOF (degree of freedom) robotic arm. These joints are shoulder_pan_joint,

shoulder_tilt_joint, elbow_tilt_joint, elbow_flex_joint, wrist_roll_joint. So we have used 5

dynamixel servo motor for developing this, single servo motor for each joint. And two motor for

the gripper control. We have designed the two sided gripper initially.

After designing this robotic arm we have faced so many problem, as we know that it is a 5

degree of freedom arm so 5 servo motors are used in that robotic arm and 2 servo motors are

used in gripper so the initial motor i.e. shoulder lift joint’s motor did not take enough torque to

lift the whole arm as at the end of the robotic arm these two gripper control servo motor provide

the large torque.

So to overcome this problem we use single sided gripper control i.e. only one dynamixel servo

motor is used in the gripper and one end of the gripper is fixed.

After facing a lot of problem then we thought about the dual motor concept that means for each

joint we have used two dynamixel servo motors. This dual motor concept is used to overcome

the torque problem. As we have developed arm initially that robotic arm is not able to take

enough torque. For that arm when we increase the load then it will not work properly. That’s

why we have used two servo motors for each joint except in shoulder_pan joint. This new arm is

4 degree of freedom arm in which joints having the number of motors as given below.

Joints

no.

Joints name Number of dynamixel servo

motors used

1 Shoulder_pan_joint 1

2 Shoulder_lift_joint 2

3 Elbow_flex_joint 2

4 Wrist_roll_joint 1

Gripper control 1

As you can see in the figure below the 4 DOF autonomous robotic arm which uses the 1 servo

motor for shoulder pan joint, 2 motor for the shoulder lift joint, 2 motor for the elbow flex joint

P a g e | 53


Fig.30 dual motor robotic arm

1 servo motor for wrist roll joint and finally 1 motor for the gripper control.

This dual motor arm is capable for lifting the coke cane because two motor provides to resist the

enough torque. So this arm is more helpful for us for designing purpose. And this arm is working

properly in our lab. This arm can be used to lift some heavy thing like bear bottle, soda bottle,

coke cane etc. so over all this autonomous arm can be used in our daily life as our helper to help.

P a g e | 54


After designing the this dual motor arm, we developed a 4 degree of freedom single using servo

motor for the block manipulation purpose. We have taken some hint from the TURTLEBOT

robot arm for the block manipulation purpose. 1 servo motor for the joint shoulder pan, 1 motor

for the joint shoulder lift, 1 motor for the joint elbow flex, 1 motor for the joint wrist flex and 1

motor for the gripper control. We have used single sided gripper.

Image

This type of arm is very easily applicable to the block manipulation. It can pick and place the

light objects like 3 cm wooden block cube, marker pen etc.

5.1.4 Gripper development:

For any robotic arm gripper is most important part by which it will pick the objects and place the

object at the goal position. You have to take care of the maximum force provided by the gripper

on the objects without damaging the objects. Some objects are very sophisticated like egg, egg is

crckable if you apply the more force on it during the grasping. So you have to take continuous

feedback from the gripper so that object does not slip, if object slips during the grasping gripper

force should increases.

Basically two types of the gripper are used in the robotics field 1st is one sided gripper and 2

nd is

two sided gripper.

Fig.31 two sided gripper control

P a g e | 55


Above shown in figure is the two sided gripper control in which two servo motors are used. This

both sided gripper can hold the object tightly. It is actuated from both side.

Fig.32 single sided gripper

Above shown figure is showing the single sided gripper in which only one servo motor is used.

One end is fixed and other end is connected to the motor which can rotate to grasp.

5.2 Software development:

In this chapter we are going to discuss about the software development in our project. As I

already told you that my robot is fully ROS based, we are using ROS packages for our projects

which all are open source means available free of cost. We will discuss about the software

development here step by step.

Before starting the work we have to install ROS and other supporting tools and packages which

are available free of cost and ROS compatible. We know ROS and all tools and packages are

open source so need of waste money for software.

P a g e | 56


5.2.1 Urdf of Robot:

URDF stands for unified robot description format. URDF is the model of your robot so first step

to our development of the software part we have to create the perfect URDF. If your URDF is

perfectly build.

Fig.33 URDF Tree structure

URDF gives the perfect description about your robot. URDF is the tree structure of the robot. As

shown in figure 1st thing you have to define the links, and you know that joint exists in between

any two links. So after defining the links we have to start to create the tree structure by defining

the joints and defining the parent Link and child Link. Like as shown in figure we have defined

links Link1, Link2, Link3 and Link4, then we have to define joint1 which exist between Link1

and Link2 in which parent Link is Link1 and child Link is Link2, joint2 exist between Link1 and

Link3 in which parent link is Link1 and child link is Link3, joint3 exist between Link3 and

Link4 in which parent link is Link3 and child link is Link4. Such a way you can see the tree like

structure of the links.

In URDF you have to give the perfect dimension of the links in meter, moment of inertia of the

links, material, color, origin, orientation of the links and types of the joint like it’s a fixed joint or

rotary joint or prismatic joint, using mesh for the links.

P a g e | 57


In the block diagram you can see the dimension of the link, orientation of the links are given.

Fig.34 1st developed URDF

P a g e | 58


This is the our first URDF for our robot developed, in which we use only blocks for the links and

the servo motors. So that is why it doesn’t looks so perfect and good. In this URDF perfect size

and dimension are not given so this URDF is perfect. You can see the joint state publisher in the

figure by which we can control the every joint.

After developing this URDF tried to make it as it looks like in original. Means we used mesh for

the dynamixel AX-12 rather than a black cuboids, and we used meshes for the links connector

like F1, F2, F3…., F16 which are used in the arm as a links and the connectors. By using meshes

it will look like same as it is in original means dynamixel servo motor looks like original

dynamixel servo motors, and the links and connectors are looks same as they are in original

piece. So after using the meshes we don’t have to take care of measuring the size. We just use

those connectors which are used in the original arm, so size is the same as in the real robotic arm.

As you can see in the figure this is the developed URDF used dual motor for single joint which

are described already. Here this URDF is look like in original what it look in the real.

Fig.35 dual motor URDF using meshes

P a g e | 59


Fig.36 latest URDF developed

After developing this dual motor URDF as well as in real we tried to work on the block

manipulation, so we took help from the turtle bot robot and we designed our 4DOF arm using 4

dynamixel servo motors and the one motor for the gripper. This arm is used in the block

manipulation, which can pick the block of size 3cm and place it on the goal position.

5.2.2 Generation of packages for our robot:

For working on the arm navigation 1st thing you have to generate the robot model i.e. URDF. If

you have your perfect design of URDF then 2nd

step is to generate the kinematic package for our

robot using planning description configuration wizard.

P a g e | 60


Fig.37 package generation of our robot model using planning description configuration wizard

P a g e | 61


Planning description configuration wizard generates the package for our robot model i.e. our

robot’s URDF. It will generates the launch files of motion planning, IK solver, planning

components visualize, planning scene ware house viewer etc. for our robot model.

Then we have developed our main launch file for robot arm, in which each nodes will be

launched like our URDF, joints parameter, robot_state_publisher, joint_state_publisher,

dynamixel follow controller, TF etc. we have created our parameter file also in format of .YAML

in which it contains the joint name, respective motor ID, max and min speed, motor temperature

etc. this file is loaded to our URDF in this launch file.

5.2.2.1 Joint_state_publisher/robot_state_publisher:

This package is to use in synchronization between our robot model (URDF) and our real robot.

By using this real robot publishing its joints state to the ROS and then our URDF subscribe that

joints state of the real robot and follow that state with the real robot..

P a g e | 62


Fig38. demonstration of synchronization of URDF and real robot

Figure 1st and second are just showing the synchronization between the real robot and URDF and

3rd

figure is showing the same with transform coordinate frames.

5.2.2.2 Dynamixel follow controller:

Dynamixel follow controller is used to follow the joint trajectory generated by the planner of our

robot. It will receive the input in form of angles and give the command to perform actuation to

the dynamixel motor. We have designed this controller for our robot’s dynamixel motor to

follow the joints angle given by planner.

P a g e | 63


Chapter: 6

6 Results and analysis: Our aim is to develop a low cost mobile autonomous robotic

arm which can perform the various application. This robotic arm can be used in table top

manipulation, in kitchen environment, in office environment and for the block

manipulation etc. We have checked in our lab only for the block manipulation using this

Fig39. block manipulation by our robot arm

P a g e | 64


autonomous robotic arm. Using this robot arm I can pick and place the block from any position

to the given goal position. As I already told that this robot arm can be used for multi purpose but

due to limited resources in our lab we have only used this arm for block manipulation. Block

manipulation activity you can see in the figure.

Collision avoiding warehouse viewer:

P a g e | 65


Fig.40 collision free trajectory generation using planning scene ware house viewer

We have developed collision free trajectory for our robot using planning scene ware house

viewer. These figure shows how we generate the trajectory using ware house viewer. Using this

ware house viewer we can filter the trajectory, and we can create the some obstacles which are

used to avoid by our robot during the trajectory execution. So we have successfully generated

trajectory for our robot in this tool.

P a g e | 66


Fig41. torque analysis of all four joints during block manipulation

Above presented in figure this is the torque analysis of each motors during performing the block

manipulation. Figure 1 shows torque analysis of the shoulder pan joint, figure 2 shows the torque

analysis of the shoulder lift joint, figure 3 shows torque analysis of the elbow flex joint and last

one shows the torque analysis of wrist flex joint.

P a g e | 67


Chapter: 7

7 Conclusion:

Main objective of this thesis is to develop a low cost mobile manipulator which can perform the

operation of pick and place in cluttered environment also. We gave our best to develop this

autonomous robot arm in real time which can avoid the collision and can perform object

manipulation.

For development of this robot arm we have developed first hardware part of our robot arm using

BIOLOID kit (AX12 dynamixel servo motor, connectors, screws, nuts etc).

After successful development of the hardware we focused ourselves towards software part. For

developing software part we used ROS (robot operating system). 1st step in this field is to

develop a perfect URDF model for our robot arm. After this using this URDF file we have to

generate our atom_arm_navigation package using planning description configuration wizard. In

this package we have forward kinematics package, inverse kinematics package, motion planning

package, perception package etc exist. After that we have to create our robot launch file, by

launching this file our robot will be online and ready for doing your application whatever

application you want to make perform. This launch file contains so many ROS nodes like

uploading the actuators YAML file for every joints, it will launch the URDF, it will launch the

perception node, it will launch the robot_state_publisher and joint_state_publisher and so on.

Now we concentrated ourselves in the block manipulation for our robot. We first build the

simple_arm_server for our robot by which our real robot arm cam contact to the ROS. After

successfully building of simple_arm_server we use turtlebot_block_manipulation package for

performing the block manipulation by our real time robot. By using all of these our robot

performs block manipulation i.e. it can detect the blocks of 3cm, it can autonomously plan its

trajectory after receiving the goal, then it can pick and place the block to the goal position.

This is the overview of my these “Real Time Collision Avoidant Low Cost Autonomous

Robotic Arm using ROS”.

P a g e | 68


Chapter: 8

8 Recommendation and future work:

This robot arm is basically a testbed on which anyone can test their developed application. Any

researcher who wants to do their research on it he can use it as a testbed and try to develop it.

Young researchers can work on the recognition of the house object like pen, cup, spoon, box of

tooth paste, soap etc.

They can work on the grasping of those objects by developing their algorithm they can find the

best grasping point and they can test on our robot arm.

As we know we are using a very low cost dynamixel AX 12 servo motor which is not giving the

better performance. So for better performance we have to use 1st high cost servo motor, so that

error due to this servo low cost motor can be removed make our system better.

As I told that we have no mechanical workshop in our organization so that we can make our

robot’s mechanical structure strong and stable. So our mechanical structure of the robot is not so

stable due to this limited resources in our laboratory. Due to this loose mechanical structure we

have faced so many problem during development of such a robot arm.

So I will suggest first to remove these problem which are discussed above i.e. use the high cost

servo motor which give low jerk and error due to temperature and build a stable frame for your

robot for the better performance.

P a g e | 69


References:

[1] J. Kramer and M. Scheutz, “Development environments for au-tonomous mobile

robots: A survey,” Autonomous Robots , vol. 22, no. 2, pp. 101–132, 2007.

[2] Ioan A. S, Ucan Mrinal Kalakrishnan, Sachin Chitta.: “Combining Planning

Techniques for Manipulation Using Realtime Perception” IEEE ROBOTICS AND

AUTOMATION MAGAZINE.

[3] Sachin Chitta, E. Gil Jones, Matei Ciocarlie, Kaijen Hsiao.: “Perception, Planning,

and Execution for Mobile Manipulation in Unstructured Environments”, IEEE

ROBOTICS AND AUTOMATION MAGAZINE.

[4] R. B. Rusu, I. A. S¸ ucan, B. Gerkey, S. Chitta, M. Beetz, and L. E. Kavraki, “Real-

time perception guided motion planning for a personal robot,” in International

Conference on Intelligent Robots and Systems, St. Louis, USA, October 2009.

[5] M. Quigley, E. Berger, and A. Y. Ng, “STAIR: Hardware and Software

Architecture,” in AAAI 2007 Robotics Workshop, Vancouver, B.C, August , 2007.

[6] K. Wyobek, E. Berger, H. V. der Loos, and K. Salisbury, “Towards a personal

robotics development platform: Rationale and design of an intrinsically safe personal

robot,” in Proc. of the IEEE Intl. Conf. on Robotics and Automation (ICRA) , 2008.

[7] N. Ratliff, M. Zucker, J. A. Bagnell, and S. Srinivasa, “Chomp: Gradient

optimization techniques for efficient motion planning,” in IEEE International Conference

on Robotics and Automation, 12–17 May 2009, pp. 489–494.

[8] R. Diankov, N. Ratliff, D. Ferguson, S. Srinivasa, and J. J. Kuffner., “Bispace

planning: Concurrent multi-space exploration,” in Robotics: Science and Systems,

Zurich, Switzerland 2008.

[9] D. Berenson, S. S. Srinivasa, D. Ferguson, A. Collet, and J. J. Kuffner,

“Manipulation planning with workspace goal regions,” in IEEE International Conference

on Robotics and Automation, May 2009.

[10] D. Katz, E. Horrell, Y. Yang, B. Burns, T. Buckley, A. Grishkan, V.

Zhylkovskyy, O. Brock, and E. Learned-Miller, “The UMass Mo-bile Manipulator

UMan: An Experimental Platform for Autonomous Mobile Manipulation,” in IEEE

Workshop on Manipulation for Human Environments, Philadelphia, USA, August 2006.

[11] C. Borst, C. Ott, T. Wimbock, B. Brunner, F. Zacharias, B. Baeum, U.

Hillenbrand, S. Haddadin, A. Albu-Schaeffer, and G. Hirzinger, “A humanoid upper

body system for two-handed manipulation,” in IEEE International Conference on

Robotics and Automation, April 2007, pp.2766–2767.

[12] S. Srinivasa, D. Ferguson, M. V. Weghe, R. Diankov, D. Berenson, C. Helfrich,

and H. Strasdat, “The Robotic Busboy: Steps Towards Developing a Mobile Robotic

Home Assistant,” in Intl. Conference on Intelligent Autonomous Systems (IAS-10), July

2008.

P a g e | 70


[13] T. Ihme and U. Ruffler, Motion Planning Based on Realistic Sensor Data for Six-

Legged Robots, ser. Informatik aktuell, Autonome Mobile Systeme, pp. 247-253.

Springer Berlin/Heidelberg, 2007.

[14] P. Sermanet, R. Hadsell, M. Scoffier, U. Muller, and Y. LeCun,“Mapping and

planning under uncertainty in mobile robots with long-range perception,” in International

Conference on Intelligent Robots and Systems, 2008, pp. 2525–2530.

[15] Y. Kuwata, G. Fiore, J. Teo, E. Frazzoli, and J. How, “Motion planning for urban

driving using RRT,” in International Conference on Intelligent Robots and Systems,

2008, pp. 1681–1686.

[16] H. Choset, K. M. Lynch, S. Hutchinson, G. A. Kantor, W. Burgard, L. E. Kavraki,

and S. Thrun, Principles of Robot Motion: Theory, Algorithms, and Implementations.

MIT Press, June 2005.

[17] http://www.ros.org/wiki/ompl.

[18] http://www.ros.org/wiki/sbpl.

[19] I. A. S¸ ucan and L. E. Kavraki, “Kinodynamic motion planning by interior-

exterior cell exploration,” in International Workshop on the Algorithmic Foundations of

Robotics, Guanajuato, Mexico, December 2008.

[20] G. S´anchez and J.-C. Latombe, “A single-query bi-directional proba-bilistic

roadmap planner with lazy collision checking,” International Journal of Robotics

Research, vol. 6, pp. 403–417, 2003.

[21] W. G. Inc., “Intern challenge,”

http://www.willowgarage.com/blog/2009/08/17/intern-pr2-challenge-2009, 2009.

[22] J. Kuffner, K. Nishiwaki, S. Kagami, M. Inaba, and H. Inoue, “Motion planning

for humanoid robots,” in In Proc. 20th Int’l Symp. Robotics Research (ISRR’03), 2003.

[23] http://www.ros.org/wiki/pr2_interactive_manipulation

[24] http://www.trossenrobotics.com/bioloid-comprehensive-robot-kit.aspx

[25] http://www.pointclouds.org/documentation/tutorials/cluster_extraction.php

[26] http://www.pointclouds.org/documentation/tutorials/planar_segmentation.php#planar

-segmentation

[27] http://docs.pointclouds.org/trunk/group__filters.html

[28] Geoffrey Taylor, Lindsay Kleeman Monash University Intelligent Robotics

Research Centre Department of Electrical & Computer Systems Engineering Monash

University, VIC 3800 Australia.: “Visual Perception and Robotic Manipulation”,

Springer Tracts in Advanced Robotics Volume 26.

developing a real time collision avoid-ant low cost manipulator using ROS

Documents