SECOND ON-SITE IARP WORKSHOP on

HUMANITARIAN DEMINING

PRISHTINA, KOSOVO

June 19-20, 2003

Introduction

INTRODUCTION The International Advanced Robotics Programme (IARP) has, till now, organized four workshops devoted to the possible use of Robotics Systems for speeding the demining of infested countries: the map shown on the first page summarizes the importance of the problem and the on this map coloured countries still suffer on this plague. The first IARP Workshop took place in Toulouse (France): a scientific meeting which gave some scientists the opportunity to present their theoretical results in the domain of the Outdoor applications for mobile robots; the most papers focused on the navigation and the control of robots while some papers also described existing advanced mechanical detection-devices (example: the UAV Camcopter) and their possibilities. Following this first workshop, a working group HUDEM has been entrusted to Dr Tom Martin (Germany) who organized a first ON-SITE workshop in Zimbabwe (more precisely at the border Zimbabwe-Mozambica in the vicinity of Mutare, where the company MINETECH, funded by the UNO, hosted the members of the WG ) The first members of the WG Hudem received there a realistic perception of the demining problems, allowing some of them to re-orientate their research activities by taking into account with the operational requirements of such an application. Dr Tom Martin concluded this workshop by inviting the members of the WG to introduce scientific proposals under FW-5 European Programme. The third Workshop, hosted by Professor P.Kopacek in the Technical University of Vienna, and co-chaired by Prof Y.Baudoin, new chairman of the WG Hudem, again gave some scientists the opportunity to present the last results of their research in the domain of the detection of anti-personnel mines, of the mechanical mine-clearance and of the roboticized demining tasks. Prof Yvan Baudoin concluded this workshop by inviting the members of the WG to pursue their efforts and to take into account with the ITEP directives (International Testing and Evaluation Programme), while Prof Peter Kopacek underlined the necessity to extend the detection tasks ( entrusted to mobile robots or other automated systems) to the mine-removal tasks (the effective mine-clearance) The fourth workshop, again cosponsored by Austria and Belgium, is the second On-Site IARP workshop on Hudem and is hosted by the Technical University of Prishtina: a ‘come-back’ to the reality of the Demining problems in very unstructured infested areas where the use of mobile robots is far from obvious… On behalf of the the cosponsoring members of the IARP, let us thank the National Organizing Committee and the participants to this workshop: their experience will allow us to prepare a detailed state-of-the-art report that will be sent to all the members of the Working Group Hudem and to the Secretary of the ITEP as well as to the Secretary of the GICHD (Geneva International Center for Humanitarian Demining) : we hope that this report will allow an efficient cooperation between those institutions and the End-Users in charge of the ON-SITE Demining EUDEM-2, the funded European Network focusing on Hudem, will organize a (by the EC funded) scientific international symposium (Brussels, 15-19 September 2003) : don’t hesitate to join this event: it will be our next Rendez-Vous this year. We also hope to organize a fifth IARP Workshop in 2004, in close cooperation with the ITEP. Be ready to contribute… Yvan Baudoin (BE) Peter Kopacek (Austria)

2

SECOND On-Site IARP WORKSHOP on Humanitarian Demining

Background of of IARP and IARP WG-HUDEM The International Advanced Robotics Programme (IARP) is an international project initiated at the Versailles Economic Summit of 1982. All countries/members of the IARP have agreed on the general objective: ”to foster international cooperation aiming to develop robot systems able to dispense with human exposure to difficult activities in harsh, demanding or dangerous conditions or environments. All members of the WG Hudem have agreed to foster international cooperation aiming to develop performant techniques and robotics systems speeding up the demining of infested countries.”

Why this Workshop in Kosovo?

Short History of Kosovo Kosovo covers an area of 10.887 km2, 53% are mountains and with about 2 million inhabitants. It is surrounded by Albania, Macedonia, Serbia and Montenegro. The capital of Kosovo is Prishtina and other major cities are Prizren, Peja, Mitrovica, Gjakova, Gjilan and Ferizaj. The highest mountains are situated in the southwest (Mali i Sharrit with peaks up to 2750m), in the west (Bjeshket e Nemura with peak Gjeravica of 2.656m) and in the north (Kopaonik). From Kosovo flow rivers toward three seas: Adriatic Sea, Aegean Sea and Black Sea. It is very rich with minerals which is a good foundation for industrial development. In the former times it was an important commercially crossroad between Adria and the Balkans. The earliest known inhabitants of Kosovo were called Illyrians by both Greeks and Romans. In 1389, in the famous Battle of Fushe Kosove / Kosovo Polje, the Serbs and their allies were defeated by the Ottoman Turks and shortly Kosovo became part of the Ottoman Empire. Kosovo was reoccupied by Serbs during the First Balkan war 1912 and remained a part of Yugoslavia until 1999 when NATO-led forces entered and set a United Nations administration. Places of interest: The Kllokot Spa, The Rugova Gorge, Brezovica, The Marble Cave, The Mirusha Canion. Heritage: Decan Monastery, Peja Patriarchy, Bajrakli Mosque in Peja, Hadum Mosque in Gjakova, Gracanica Monastery near Prishtina, Novo Brdo Castle, Prizren Castle etc.

Facts about Prishtina Prishtina is Kosovo’s capital with more than 500,000 inhabitants. It’s a administrative, economic, cultural and education center. The history of Prishtina goes back to the Neolithikum and is situated near the ancient town called Ulpiana. Because of the minerals it was in the mid age (14th and 15th century) a very important market place. Now, it’s a modern city with three national TV stations, University with more than 20,000 students, National Library, National Theatre, National Museum, Academy of Science and Arts, Catholic Church, Orthodox Church and many old Mosques. Near Prishtina is based Airport as a link with outside world. In outskirts of city Thermo Power Plants are installed which supply Kosovo with electric power. Recreative and sportive center is situated about two kilometers from the center in Germia Valley. Memorial for the Kosovo Battle and Sultan Murat’s Tomb are just few kilometers outside Prishtina.

Demining Problem in Kosovo At least 120 people have been killed and 370 wounded by landmines and cluster bombs left over from the Kosovo war since the conflict ended in June 1999. More than 4,500 mined fields were marked by international agencies. By the end of 2001 the United Nations Mine Action Coordination Centre (UNMACC) has disposed of 20,000 landmines and international peacekeepers have removed 7,000 of around 45,000 located. Since then Kosovo Protection Corps (KPC-TMK) is in charge for demining and explosive device clearance. According to international standards mine clearance can be guaranteed up to 99.6% and Kosovo needs up to seven years to reach that rate. Therefore, the Workshop in Pristina, Kosovo will examine how it could be possible to speed up the demining process using sophisticated devices as are mobile (wheeled, tracked or legged) robots and other modern equipments. http://www.ihrt.tuwien.ac.at/HUDEM03/HUDEM.htm - background

3

Sponsorship This Workshop is sponsored by Austria and Belgium and co-sponsored by France, Russia, United States. The IARP representatives of the sponsoring and co-sponsoring countries are : Austria P. Kopacek Vienna University of Technology Belgium Y. Baudoin Royal Military Academy, Brussels Canada E. Dupuis,

J.C. Piedboeuf Canadian Space Agency, Quebec

France G. Giralt LAAS-CNRS, Toulouse Russia V. Gradetsky Russian Academy of Sciences Spain M. Armada Instituto de Automatica Industrial (CSIC) USA E. Marsh National Science Foundation (NSF)

International Programme Committee: A. Almeida (PRT) M. Armada (ESP) Y. Baudoin (BEL) : Chair T. Borangiu (ROM) S. Buza (KOS) N. Caplan (USA) : IARP President J.G. Fontaine (FRA) : Representative P. Gonzalez (ESP) V. Gradetsky (RUS) M. Jimenez (ESP) P. Kopacek (AUT) : Chair B. Kuchen (ARG) C.W. Lee (KOR) E. Marsh (USA) T. Martin (GER) A. Pajaziti (KOS) C.E. Pereira (BRA) J. Sa Da Costa (PRT)

Local Organizing Committee

• I. Gojani (University of Pristina, Mechanical Engineering Faculty, Kosovo) • A. Pajaziti (University of Pristina, Mechanical Engineering Faculty, Kosovo) • Sh. Buza (University of Pristina, Mechanical Engineering Faculty, Kosovo)

Mechanical Engineering Faculty University of Prishtina Bregu i Diellit pn. Prishtina Kosovo, UNMIK Tel/fax. +381 38 554 997

4

PROGRAMME JUNE 19, 2003 09.30-10.00 H Opening - Welcome Dean of the M.E.Faculty

(Ko) Prof P.Kopacek (AU) Prof Y.Baudoin (BE) Dr A.Pajaziti (Ko)

10.00-10.30 H Demining Problem in Kosovo

Dr S.Buza (M.E.F Prishtina)

10.30-11.00 H Demining in Kosovo Major D.Rexha (Kosovo Protection Corps)

11.00-11.30 H Coffee Break 11.30- 12.00 H Anti-Personnel mines

neutralisation JC. Robert-Peillard, Y.Riebernon (Hitachi Furukawa, FR)

12.00-12.30 H Fuzzy-genetic algorithm and obstacle path generation for walking robot

Dr A.Pajaziti, I.Gojani,S.Buza,A.Shala (MEF, Prishtina, KO)

12.30-13.45 H Lunch Break 13.45- 14.15 H Robotics for humanitarian

demining: a need? Yvan Baudoin, Eric Colon, Ping Hong, Jean-Claude Habumuremyi, Ioan Doroftei,, Hichem Shali, Dragomir Milojevic, Jérôme Weemaels

14.15-14.30 H Existence of an Indian market for Robotics Demining

A.Rao (Goldspark, Bangalore)

14.45 – 15.15 H Sensory and Robotics for Humanitarian Demining

Prof M.Acheroy Prof Y.Baudoin (RMA, BE)

15.15- 15.45 H Trajectory tracking by using fuzzy logic controller on mobile robot

Dr A.Shala, R.Likaj,A.Geca,A.Pajaziti,F.Krasniqi (MEF, Prishtina, KO)

15.45-16.00H Coffee Break 16.00-16.30 H Low cost, modular robot

for landmine detection in thick vegetation

E.Cepolina (Un Genova, IT)

16.30-17.00 H Humanoid Robots for demining

Prof P.Kopacek (TU Wien, AU)

17.00-17.30 H Utilisation of Burmester theory in design of planar mechanisms/manipulator

Dr L.Gojani,A.Pajaziti,B.Pira,P.Marango (Polytechnical Un Tiran, ALB)

17.30 H PANEL DISCUSSION IARP-KPC

5

MR. J.C. ROBERT-PEILLARD

COMPANY OVERVIEW

7

HITACHI-FURUKAWA LOADERS EUROPE S.A.S. (H.F.L.E.) is established by leading companies in the industry, namely HITACHI CONSTRUCTION MACHINERY COMPANY and FURUKAWA Co. Ltd. H.F.L.E. is located in the suburbs of Lyon (Genas) and benefits from an outstanding location, in the neighbourhood of the main motorway network, the speed rail links (TGV) and Lyon's international airport. The principal activities of the company are as follows : 1. Manufacture and design of wheel loaders and their related

attachments. 2. Manufacture and design of mine clearance equipments and

systems. Wheel loaders manufactured by the company are distributed exclusively through the HITACHI's sales network.

CATEGORY : ANTI-PERSONNEL MINES NEUTRALIZATION.

EQUIPMENT NAME : A 15 ton Excavator (prime mover) equipped with an Electronic Kit to address and neutralize through "Tool Box Concept" anti personnel mines located in hardly accessible areas.. Description : The HITACHI-FURUAKWA mine clearance equipment and systems offers different solutions to attack and neutralize anti-personnel mines located on hardly accessible ground or surfaces such as slopes, ditches, along walls, fences, trenches, dykes, etc... The purpose of these tools is to concentrate on small areas and to support or complement other demining processes such as manual mine clearance and high scale mechanical techniques. A simple and robust electronic box connected to angular sensors installed as a kit on the excavator, allows coordination of the movements of the system kinematics in space, and consequently at the end of the boom stick. This enables adjustment and automatic operations of the tools attached in the selected area. Normal operation procedures can be achieved through an optional remote control system. A camera monitors and enables the operator to visualize and control the mine clearance sequence.

EQUIPMENT NAME : Mine Sweeper AMDS ("Off the shelf tool").

8

Description : The anti-personnel mine demining system (AMDS) is an area reduction tamping device for identification of the presence of anti-personnel mines in suspect areas and for the neutralization of these mines by detonating any near to surface or surface pressure mines. Vibrating tamping capacity, combined with a simple electronic instrumentation, allows demining of areas that are difficult to access by manual demining or with other doming mechanical equipment. Six independent and articulated footpad kits exert pressure on the ground through tamping, 1 ton per foot, and consequently detonate the anti-personnel mines. Additionally, these feet serve as a fuse to absorb the blow to the detonation. Upon explosion, the plastic pipe feet can be easily and quickly replaced. The total AMDS ground tamping process can also be achieved through a remote control system. A camera monitor set enables the operator to visualize and control the operation. The weight of the AMDS is 700 kg. EQUIPMENT NAME : DiggerExcavator / Sifter UXOES ("Off the shelf tool"). Description : The HITACHI-FURUKAWA excavator sifter (UXOES) is designed for sifting soil in difficult access areas where anti-personnel mines are a threat (i.e. slopes, ditches, channels, dykes, etc.) The UXOES tool has a combination of "mill/rippers" or earth removing teeth and a "sifter drum basket" which sorts out the earth removed up to a depth of 25 cm. The tool can dig and sort out the earth through the front and back swing of the "drum basket". At the end of the sequence the tool allows the automatic repositioning of the mill/ripper. In this manner, no surface remains unexplored. This particular anti-personnel mine tool is ideally suited for sandy ground and weights 350 kg.

EQUIPMENT NAME : Vegetation Slasher / Bush Cutter ("Off the shelf tool"). Description :

9

This tool is a solid bush cutter and is designed with modular hood protection to accommodate different working applications. For cutting vegetation, two types of cutters can be easily assembled to cope with different vegetation clearance requirements. The bush cutter can also be transformed to a mini flail head (long straight blade) and has limited applications for clearance of anti-personnel mines down to a ground penetration depth of 5 cm. It is particularly effective on sloped terrain such as channels, trenches and dykes. POINT OF CONTACT : MR. J.C. ROBERT-PEILLARD Mobile : + 00 33 6 09 85 83 04 Phone : + 00 33 4 72 23 28 81 Fax : + 00 33 4 78 90 09 56 e.mail : operations.furukawa@libertysurf.fr COMPANY INFORMATION : 35, rue Roger Salengro – B.P. 211 F – 69742 – GENAS CEDEX (FRANCE)

A FUZZY-GENETIC ALGORITHM AND OBSTACLE PATH

GENERATION FOR WALKING ROBOT WITH MANIPULATOR

Pajaziti, A.; Gojani, I.; Buza, Sh. & Shala, A. Mechanical Engineering Faculty, University of Prishtina, Kosova

Abstract

We present eight-legged walking robot Scorpion with manipulator that is controlled with a Fuzzy –Genetic algorithms basic approach. The goal is to develop walking robots for outdoor environments for the use of landmine detection. The eight-legged robot has to plan its path as well as gait simultaneously, while moving on flat terrain in presence of obstacles and ditches. It is a complicated task and no single traditional approach is successful in handling this problem. Due to the computational expenses there is a need to develop the faster and efficient algorithm for generation of optimal paths and gaits. The Fuzzy-Genetic algorithms approach developed in this paper enabled the tetrapod to do its job with minimum time of travel and with an optimum number of legs on the ground. The effectiveness of the algorithm is tested through computer simulations. Keywords: Walking Robots; Fuzzy Control; Genetic Algorithm; Simulation; Path and gait generation 1. Introduction

Mobile platform systems, consisting of a walking robot platform equipped with one or more manipulators, are of great importance to host applications, mainly due to their ability to reach targets that are initially outside of the manipulator reach. Applications for such systems abound in mining, construction, forestry planetary exploration and the military. A wide category of such systems employs wheeled mobile robots, walking robots, which is the type of system under study in this paper. Walking robots continue the world-wide efforts in realizing such mechanisms with the goal to understand walking on the one side and to push forward a technology on the other side which might be useful for quite a lot of applications where wheel-driven machines are not reasonable. Research on legged robots typically concentrates on determination of vehicle’s trajectory, foothold selection and design of a sequence of leg movements. Motion planning for walking robots with manipulator is concerned with obtaining open loop controls, which steer a platform from an initial state to a final one, without violating the nonholonomic constraints. A comprehensive survey of developments in control of nonholonomic systems can be found in Kolmanovsky and McClamroch (1995). Moving mobile manipulator systems present many unique problems that are due to coupling of holonomic manipulators with nonholonomic bases. Seraji (1998) presents a simple on-line approach for motion control of mobile manipulators using augmented Jacobian matrices. The approach is kinematics and requires additional constraints to be met for the manipulator configuration. Perrier et al. (1998) represent the nonholonomy of the vehicle as a constrained displacement and try to make the global feasible displacement of the system correspond to the desired one. Foulon et al. (1999) consider the problem of task execution by coordinating the displacements of a nonholonomic platform with a robotic arm using an intuitive planner, where a transformation was presented. The same authors introduce other variations of local planners, which are then combined to constitute a generalized space planner (Foulon, Fourquet, and Renaud 1998). Papadopoulos and Poulakakis (2000) presented a planning and control methodology for mobile manipulator systems allowing them to follow desired end-effector and platform trajectories simultaneously without violating the nonholonomic constraints. The problem of navigating a mobile manipulator among obstacles has been studied by Yamamoto and Yun (1995) by simultaneously considering the obstacle avoidance problem and the coordination problem. To reduce the computational complexity of motion controls of mobile manipulator systems, some heuristics have also been developed by several researchers. Fuzzy Logic Controllers (FLCs) have been used by

10

several researchers in the recent past (Deb 1998, Pratihar, 2002, and Mohan, 2002) to solve the walking robot with manipulator navigation among stationary obstacles. The optimization problem involves finding an optimal fuzzy rule base that the walking robot should use for navigation, when left in a number of scenarios of stationary obstacles. Once the optimal rule base by using the Genetic Algorithm (GA) is obtained off-line, the walking robot can then use it on-line to navigate in other scenarios of stationary obstacles. In the present study, we concentrate on navigation problem, where the objective is to find an obstacle-free path between a starting point and a destination point, requiring the minimum possible time of travel. This paper consists of five sections: Section 2 describes the walking robot with manipulator model. The possible solution of the problem using FLC is proposed in Section 3. Section 3.1 discusses the Fuzzy and GA approaches. The results of computer simulations are presented and discussed in Section 4. Some concluding remarks and scope for future work are made in Section 5. 2. Mobile Platform and Manipulator System Model The mobile platform and manipulator system model consists of two parts: walking robot with 8 legs, and manipulator with two segments (see Fig. 1). Because nonholonomy is associated with the mobile platform, while the manipulator is holonomic, the system is studied as two connected subsystems, the holonomic manipulator and its nonholonomic platform. This allows one to find an admissible path for the mobile platform that can move it from an initial position and orientation to a final desired one. Next, using known techniques for manipulators, joint trajectories are calculated for the manipulator so that its end-effector is driven to its destination. An advantage of this approach is that it is very simple to extend the method to mobile systems with multiple manipulators on board.

Fig.1. Eight-legged walking robot Scorpion with manipulator model during mine detection on a flat terrain

2.1. Nonholonomic Mobile Platform Subsystem The eight legged robot was adopted as a mobile platform that is 65 cm long. The minimum height of the robot is 20 cm, the maximum height (legs outstretched) is 35 cm. It consists of a central body and eight legs with three segments each (see Fig. 2). The legs consist of a thoracic joint for protraction and retraction, a basal joint for elevation and depression and a distal joint for extension and flexion of the leg. With 6 degrees of freedom (DOF) of the central body the mobile platform has 32 DOF. Each leg has three DOF, turning around the α, β and γ axes, while the manipulator has two DOF (θ1 and θ2), which is installed on the front side of the mobile platform.

11

α

β

γ

Segment 1

Segment 2

Segment 3

Central body

H

S

θ2

θ1

Detector Manipulator

v

M

C

8 12

Fig. 2. Model of the walking robot with a single leg and manipulator

Position of point M (the point where manipulator is attached to the central body) is given by equation:

[ ] )sin()cos()cos( 321 αβγβ −+++= ssssxM (1)

Velocity of robot platform is given by:

[ ][ ] )cos()cos()cos(

)sin()sin()()sin(

321

32

ααβγβ

αβγβγββ

&

&&&

−+++

+−−−−=

sss

ssvM (2)

s1

β

γ

S

H

x

α

s2

s3

s1

s2s3

x

t

z

y

x

M

M’

s

S

F

swing

stance

Fig. 3. Kinematics parameters of the central body and leg L1

12

During the swing/stance phases (see Fig. 3) the body height H and foot distance S should remain constant: .)sin()sin( 32 constssH =−+−= βγβ (3)

[ ] .)cos()cos()cos( 321 constsssS =−++= αβγβ

2.2. Holonomic Manipulator Subsystem The Cartesian coordinates of the center of mass C1, center of mass C2 and the of detector D to the mobile platform joint M are given by (see Fig. 4). Position of C1 - center of mass for link 1:

)cos(2

)cos( 11

1θϕϕ −−⋅⋅=

ltvx oC (4)

)sin(2

)sin( 11

1θϕϕ −−⋅⋅=

ltvy oC

Position of C2 - center of mass for link 2:

)cos(2

)cos()cos( 212

112θθϕθϕϕ +−+−−⋅⋅=

lltvx oC

(5)

)sin(2

)sin()sin( 212

112θθϕθϕϕ +−+−−⋅⋅=

lltvy oC

Position of detector D on xy plane is defined by:

tvd oM ⋅= ,

l1

dM

l2

x1

y1

θ1

θ2

D

ϕ

ϕ

x

y

mobile platform

C M

M’

C2

C1

Fig. 4. Kinematics parameters of the central body and manipulator

13

or by its components:

)cos()cos()cos( 21211 θθϕθϕϕ +−+−−⋅⋅= lltvx oD (6)

)sin()sin()sin( 21211 θθϕθϕϕ +−+−−⋅⋅= lltvy oD Velocity of detector D is defined by:

)sin()()sin(

)()sin()cos(

212121

11

θθϕθθϕθϕ

θϕϕϕϕ

+−+−−−⋅

⋅−+⋅⋅−=&&&

&&&&

l

ltvvx ooD

)cos()()cos(

)()cos()sin(

212121

11

θθϕθθϕθϕ

θϕϕϕϕ

+−+−+−⋅

⋅−−⋅⋅+=&&&

&&&&

l

ltvvy ooD (7)

or in matrices form:

⎥⎦

⎤⎢⎣

⎡=

D

DD y

xp

&

&& , (8)

with

⎥⎦

⎤⎢⎣

⎡⋅⋅⋅+⋅⋅⋅−

=− )cos()sin()sin()cos(

ϕϕϕϕϕϕ

&

&

tvvtvv

Joo

oorw ; ; ; ⎥

⎦

⎤⎢⎣

⎡+−−−+−−−

=)cos()cos()sin()sin(

21211

21211

θθϕθϕθθϕθϕ

llll

J m

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

+−−=

21

1

1

θθϕθϕ&&&

&&&q

where is [ ]mrw JJJ −= central body and manipulator Jacobian, that consists of two sub-jacobians, Jw-r and Jm, where Jw-r multiplies the mobile platform velocities corresponding to the uncontrolled degrees of freedom, and Jm multiplies the manipulator joint velocities corresponding to the controlled degrees of freedom. is the vector of central body and manipulator joint velocities. The matrix form of the equation (8) is given by:

q&

qJpD && ⋅= (9)

2.3. Mobile Platform and Manipulator System Dynamics Kinetic energy of the manipulator is given by equation:

221

21

222

221

22

)(21)(

21)(

21

)(21)(

21

2122

11

θθϕθϕ &&&&&&&

&&&&

+−+−+++

++++=

CCCC

CCDDDk

JJyxm

yxmyxmE

(10)

where the moment of inertia for center of link 1and 2 is:

21112

11

lmJ C = , 22212

12

lmJ C = ,

Second order Lagrange equations of motion for the mobile platform and manipulator is:

τ=∂∂

−⎟⎟⎠

⎞⎜⎜⎝

⎛∂∂

qE

qE

dtd kk

& (11)

14

where τ - represents forces and torques exerted by walking robot and manipulator joint torques, Ek is the kinetic energy of system, while q and is the vector of walking robot and manipulator joint positions and velocities.

q&

2.4. Gait patterns Fig. 5. left shows the tetrapod gait, that describes the phase characteristics of the leg movement of the walking robot. The tetrapod gait, where four legs are on the ground at the same time, can be observed at low speeds of walking robots. During walking, a eight legged robot uses its 2 tetrapods not unlike a bipped stepping from one foot to the other – the gait is simply shifted alternately from one tetrapod to another. Since 4 legs are on the ground at all times, this gait is both “statically and dynamically stable. A cycle consists of a leg promotion period (swing ) and a leg remotion period (stance ). Phase, swing, and stance are the parameters of the robot’s motion pattern (Fig.5, right) together with the step size of a leg.

1 2

R4 L3 R2 L1

L4 R3 L2 R1

Fig. 5. Tetrapod walking gait (left) and an according motion pattern beginning from the start position

(right)

L1 R1

L2 R2 L3 R3 L4 R4

3. Solution of the problem by using Fuzzy Logic and GA Controller The problem can be stated as follows: A eight-legged robot will have to plan its path as well as gait simultaneously, while moving on flat terrain with occasional hurdles such as ditches and in presence of stationary obstacles. The eight legged robot needs to do this job by avoiding to collide with any obstacles and not falling into ditches and all within the minimum time of travel and with an optimum effort to gain ratio. Moreover, its body height H and foot distance S should always be same to ensure static stability. To perform above tasks, in practice, the walking robot will have to do the following sub-tasks optimally (with minimum travel time):

1. Move along straight-line paths (only translation). 2. Take sharp circular turns (only rotation).

In order to simplify the problem the following assumptions are made: 1. The contact of the feet with flat can be modeled as point contacts. 2. Each obstacle is represented by its bounding circle. 3. The terrain is discretized into cells and the center of each cell is considered as a candidate

foothold. 4. The mass of the legs is reduced into the body and the center of gravity is assumed to be at the

centroid of the body. 5. The detector should cover the surface with constant foot distance S during the forward motion of

the walking robot. 3.1. Used controllers

15

In our proposed fuzzy and Fuzzy Genetic algorithm, two potential tools, namely fuzzy logic controller (FLC) and genetic algorithm (GA) have been merged to utilize the advantages from both. It is to be noted that the performance of an FLC depends on its rule base and membership function distributions. It is seen that optimizing rule base of an FLC is a rough tuning process, whereas optimizing the scaling factors of the membership function distribution is a fine tuning process. The FLC and GA controllers which are used for solving the problem of combined path and gait generations simultaneously of a walking robot with manipulator. The FLC finds the obstacle-free direction based on the predicted position of the obstacles in the next time step. The inputs, namely position and velocity of the central body and manipulator are fed to the FLC and there is the control vector to compensate the uncertainties of the model. A GA is used to control the body height H and foot distance S. The walking robot with manipulator will have to reach its destination starting from an initial position in minimum traveling time and its generated gaits will be optimal (with minimum number of ground-legs). This criteria is considered on the selection of the GA parameters, namely population size, crossover probability, mutation probability and the number of generations. Thus, the output from the GA block was also the control vector that compensated the gait errors. 4. Simulation results For simulation, according to the biometric approach the patterns are transferred to the model of robot, we have use sine function f(t) to avoid too abrupt accelerations, Fig. 6. The function f(t) for leg L4 is repeated by the other legs as g(t) after a partition of cycle.

⎪⎪⎪⎪

⎩

⎪⎪⎪⎪

⎨

⎧

+⋅≤

+−

⋅

⋅≤

−⋅

=

25.0)5.2,mod(5.0

)22

5.0sin(30

5.0)5.2,mod(0

)2

2sin(30

)(

π

ππ

π

ππ

tif

t

tif

t

tf (12)

The phase ϕ between two legs remains constant independent of the velocity.

⎪⎪⎪⎪

⎩

⎪⎪⎪⎪

⎨

⎧

+⋅≤

++−

⋅

⋅≤

+−⋅

=

25.0)5.2,mod(5.0

)22

5.0sin(30

5.0)5.2,mod(0

)2

2sin(30

)(

π

ϕππ

π

ϕππ

tif

t

tif

t

tg (13)

During the walk, all legs of robot, when they are in swing-position, distance S = 17 [cm] (distance between central body axe and foot - end side of segment 3) is constant. Also during the walk, central body axe height is constant value H = 15 [cm]. Length of segments and manipulator links are: ][51 cms =

][282 cms = , , and . Desired velocity of robot - central body is v][233 cms = ][121 cml = ][242 cml = o = 0.5 [m/s]. The rotational motion of the joints is transferred to the forward motion of the walking robot.

16

5 5.5 6 6.5 7 7.5 8 8.5 9 9.5 1030

15

0

15

30

trace 1trace 2trace 3trace 4

L4 Time [sec] R3 L2 R1

deg

f (t)

Fig. 6. A tetrapod group of thoracic hip joints for swing/stance = 3/5 and phase ϕ = 0.1 Control scheme (Fig. 7) of walking robot is realized in Matlab/Simulink.

Fig. 7. Control scheme of walking robot using Fuzzy – Genetic Algorithm In control scheme shown in Fig. 7, through “Inverse dynamics” block, as output we have actuator torque on leg joints and manipulator joints (τ). From block “Fuzzy Logic Controller” also as output is torque (τflc). Through the Genetic Algorithm we have design as optimization algorithm for body height H and foot distance S of the walking robot, which are needed to be constant values. Using instructions given by Kalyamoy Deb where GA implementation is using binary and real coded variables, we have developed a Genetic Algorithm Block. Inputs are estimated body high H and foot distance S. Outputs are torques (τga). Block “Walking robot and manipulator” represents the “Direct dynamics” solution of dynamic model for walking robot and manipulator given by differential equations:

gaflcqqhqqD τττ ++=+⋅ ),()( &&& (14)

where D(q) is symmetric inertia matrix, is the vector of Coriolis, centrifugal and gravitational torque’s and other unmodeled disturbances, and q is the vector of joint acceleration.

),( qqh &

&&All simulated variables can be viewed namely, the paths, velocities and errors.

17

Pat

h [c

m]

x

y

z

Path of walking robot – central body Path of detector D

Legend:x direction [cm] 500 375 250 125 0 70

27.5

15

57.5

100

Fig. 8. Path of the walking robot-central body and detector D

x direction [cm]

x

y

z

Path of walking robot – central body Trace of foot L1

Legend:

500 375 250 125 0 70

26.5

17

60.5

104

Pat

h [c

m]

Fig. 9. Path of walking robot - central body and foot L1

velo

city

[m/s

]

0 4 8 12 16 20 24 28 32 36 40 0.48

0.49

0.5

0.52

0.54

Time [s] Fig. 10. Velocity of walking robot - central body

Fig. 8 shows the path of central body and detector, while Fig. 9 shows the path of central body and foot L1, by using the Fuzzy-Genetic algorithms. The velocity of walking robot is rather constant during the entire simulations (see Fig. 10).

18

Erro

r E-

S [c

m]

Time [s]

0 4 8 12 16 20 24 28 32 36 400.05

0

0.05

0.10

0.15

Erro

r E-

S [c

m]

Time [s]

0 4 8 12 16 20 24 28 32 36 400.66

0.33

0

0.33

0.66

a) b)

Fig. 11. Stance – foot distance S error: a) with GA and b) without GA

Erro

r E-

H [c

m]

Time [s]

0 4 8 12 16 20 24 28 32 36 400.12

0.06

0

0.06

0.12

Erro

r E-

H [c

m]

Time [s]

0 4 8 12 16 20 24 28 32 36 40 0

0.10

0.20

0.30

0.40

a) b)

Fig. 12. Error on walking robot - central body height H: a) with GA and b) without GA In case of keeping the constant foot distance there were significant differences between the used algorithms with GA and without GA (see Fig. 11 a and b). The GA algorithm has also influenced in an optimal manner to reduce the central body height error (see Fig. 12). 5. Conclusions Position of walking robot – central body and leg L1 had no significant deviation from desired values, but after 10 seconds the deviations increased considerably. The velocity error of the walking robot – central body in comparison with desired values was not so significant. The performances of controller GA keeping constant values of the body height H and foot distance shows that the GA decreased four times. This work can be extended further where the fuzzy rule will be designed by a GA that would be suitable for on-line implementation due to its computational complexity. References Deb, K., Pratihar, D. K., and Ghosh, A.1998. Learning to avoid Moving Obstacles Optimally for Mobile

Robots Using a Genetic-Fuzzy Approach. A.E. Eiben et al. (Eds.); PPSN V, LNCS 1498, pp. 583-592. Foulon, G., Fourquet, J.-Y, and Renaud, M. 1998. Planning point to point paths for nonholonomic mobile

manipulators. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS), pp. 374-379. Foulon, G., Fourquet, J.-Y, and Renaud, M. 1999. Coordinating mobility and manipulation using

nonholonomic mobile manipulators. Control and Engineering Practice, vol. 7, pp. 391-399. Gojani, I., Pajaziti, A., and Shala, A. 2002. Mobile robot navigation using cognition models and genetic

algorithm-based approach. The 13-th International DAAAM Symposium, pp. 189-190.

19

Klaassen, B., Linnemann R., Spenneberg D., and Kirchner, F. 2002. Biometric walking robot SCORPION, Control and Modeling, Robotic and Autonomous System, 41, pp. 69-76.

Kolmanovsky, I., and McClamroch, H. 1995. Developments in nonholonomic control problems, IEEE Control Systems, pp. 20-35.

Mohan, A., and Deb, K. 2002. Genetic-Fuzzy Approach in Robot Motion Planning Revisited, N.R. Pal and M. Sugeno (Eds.): AFSS 2002, LNAI 2275, pp. 414-420.

Papadopoulos, E., and Poulakakis J. 2000. Planning and model-based control for mobile manipulators. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS), pp. 1810-1815.

Perrier, C., Dauchez, P., and Pierrot, F. 1998. A global approach for motion of nonholonomic mobile manipulators. Proc. Of the IEEE Int. Conf. on Robotics and Automation, pp. 2791-2796.

Pratihar, D. K., Deb, K., and Ghosh A. 2002. Optimal path and gait generations simultaneously of a six-legged robot using a GA-fuzzy approach, Robotic and Autonomous System, 41, pp. 1-20.

Seraji, H. 1998. A unified approach to motion control of mobile manipulators. The International Journal of Robotics Research, 17(2), pp. 107-118.

Yamamoto, Y. and Yun, X. 1995. Coordinated obstacle avoidance of a mobile manipulator. Proc. of the IEEE Int. Conf. on Robotics and Automation, pp. 2255-2260.

20

Robotics for Humanitarian Demining : a need ?

Yvan Baudoin, Eric Colona*, Ping Honga, Jean-Claude Habumuremyia, Ioan Dorofteia, , Hichem Shalib, Dragomir Milojevicc, Jérôme Weemaelsd

a Robotics Centre, Royal Military Academy, Avenue de la Renaissance 30, B-1000 Brussels Belgium b Department of Electronics and Information Processing VUB-ETRO-IRIS, Vrije Universiteit Brussel,

Pleinlaan 2, B-1050 Brussels c Service des Systèmes Logiques et Numériques, Université Libre de Bruxelles 50, Av. F.Roosevelt,

CP 165/57 B-1050 Bruxelles d SCMeRO, Université Libre de Bruxelles 50, Av. F.Roosevelt, CP 165/42 B-1050 Bruxelles

phone: +32 2 737 6552 fax: +32 2 737 6547

eric.colon@mapp.rma.ac.be

Abstract. In the current situation, there is a need to correctly define the usefulness and requirements of robotics solutions , essentially in pre- and post-mine detection (minefield delineation and quality assurance), to develop a network of research-centers focusing on this kind of solutions , to define and continuously update generic modules of the used Robotics Systems . Beside the correct orientation of research activities , deduced from such definitions, it will be necessary to develop test methods and procedures in order to assess the performances of the 'System' in highly, cost-effective and most generic way. The Network , with teams focusing on work-packages related to the modules defined in the picture 1, could help to clarify the role of the Robotics Systems (or Mechanical assistance) and assist future T&E activities of ITEP. (International Test and Evaluation Programme). This paper summarizes the results of the research activities on Robotics in Humanitarian Demining, conducted by the RMA, as well as some important conclusions regarding the needs of such research objectives? Keywords: Robotics, Control,

1. Introduction. The terrible antipersonnel landmines plague represents a real challenge for the research community. Antipersonnel mines kill or mutilate tens of people every day. Humanitarian deminers still use classical manual methods because heavy demining vehicles cannot achieve a satisfying

destruction percentage. This work is very slow, tedious, dangerous and costly. Furthermore, the detection is not always reliable. Improvements can be made by developing new sensors, by automating the detection sequence and by using different sensors simultaneously (Baudoin & Colon, 1999). The Royal Military Academy, leading the Belgian project Hudem, is focusing on the development of new data processing and fusion algorithms (Milisavljevic 1999), on improvement of Ground Penetration Radar (GPR) (Scheers 1999) and on robotic systems that carry mines detection sensors. Among the different ways robots could help human deminers, the scenarios described below are the most realistic. Small autonomous vehicles equipped with different sensors could run around an area to delimit the surface that is really polluted with mines. This phase when done manually is the most dangerous one because deminers are working faster and are taking more risks than during systematic detection. To study this first aspect, a small wheeled electrical vehicle named Tridem has been developed (figure 1). A first indoor prototype was build to demonstrate the concept feasibility (Alexandre, Weemaels, Doroftei & Colon 1998). A second version with more powerful motors and a modified frame has been developed for outdoor trials. Once the actual mined area is delimited, a systematic scanning process can begin. It has been proved that the use of different sensors could drastically improve the detection efficiency and reliability. However, the data fusion process requires the registration of the data acquired by the different sensors. This requirement justifies the work done on the two

23

other systems: the first one which is a sliding pneumatic robots with a 2 degrees of freedom scanner (figure 2) and the second one which is a modified Explosive Ordnance Disposal (EOD) vehicle that carries a three-dimensional Cartesian scanning system (figure 3) (Colon, Baudoin & Alexandre 1998). The fact that the three systems have very different robotic architectures introduced some additional challenges in the development of the control and interface architecture. Also the different sensors used in the project and more particularly their interface influenced the system implementation.

The last aspect considered in this project is the determination of the robot's location in the field. This is required for navigation but also for automatic production of detection maps. For this purpose, a visual servoing system based on a pan-and-tilt colour camera has been developed. (Hong, Sahli & Baudoin 1998). This system tracks a colour beacon mounted on the robot and sends in real-time the three-dimensional position of the sensor to the main control computer. The principle is shown in figure 4. Besides mobility trials, the systems have been used on dummy domestic minefields to record data for the searchers working on sensor development and data fusion algorithms.

Figure 1. The Tridem with the Metal Detector Figure 2. The pneumatic walking robot AMRU 4

24

Figure 3 The Hunter with a blue beacon Figure 4 The tracking system principle

Figure 5a. AMRU-5 : the six-legged robot Figure 5b. Possible configuration of AMRU-4

Finally, although this may be considered as a long-term study, we also decided to develop a six-legged robot (figure 5) and to study multi-tools systems for the AMRU 4 (figure 5b). This text continues with an overview of the different components of the robotic acquisition

systems. Next, the control and communication architecture is described. The emphasis is put on the control of the scanning and on the synchronisation between the distributed processes. Finally some results are presented and future work is discussed.

2. System Overview Basically, a Robotics system may be described by the next scheme:

25

Vehicle status

Motion control

Sensordeployment

Robotpositioningsensor

Cutter control

Controltransceiver

Sensor dataprocessing

Sensor datatransceiver

Data acquisitionprocess

Sensor

Data acquisitionprocess

Sensor 2

Data acquisitionprocess

Sensor 3

Data acquisitionprocess

Sensor n

Transmit /Receive

Transmit /Receive

Vehicle host

Sensor datafusion

Missionmanagement

HMI

DATANETWORK

VEHICLE CONTROL NETWORK

CONTROLSTATIONBACKPLANE

MIS

SIO

N D

ATA

REP

OSI

TOR

Y

CO

MM

UN

ICA

TIO

N L

INK

SENSOR SUITE

During our project we focused on the shaded modules. Other work groups of our Hudem-team focused on the other ones. The scanning robotic systems are composed of the following elements: - The vehicles with their possible scanning

device, - The mine detection sensors, - The tracking and location system. As described in the preceding section the robotic systems can be divided into two categories: the ones having a scanning device that can be equipped with different sensors and the wheeled robot that simply carries a single sensor. In this case, the scanning of an area can be obtained by moving the robot body itself. The mechanical systems need to be controlled and their motion must be synchronised with the data acquisition. The vehicle motion is also synchronised with the tracking and location process. The vehicles and the scanners are described in this section; the control and communication architecture is presented in section 3. The first vehicle available in the project is called Hunter (figure 3); it is a small caterpillar vehicle that was used by the Belgian Army for anti-terrorism operations. The original manual

control system of the vehicle has been interfaced with a microcontroller such that the vehicle can now be controlled with a computer. The scanner is actually a Cartesian robot with three degrees of freedom. A DC servomotor coupled with a planetary gear-head actuates each axis. An optical digital encoder is used for position and speed feedback and contact switches have also been placed at the end of each axis. The useful area is 850 by 500 mm and the vertical axis has a travel distance of 500 mm. The system can be used on both sides of the vehicle and is powered with batteries or with a power supply. The second robot (AMRU4) is a pneumatically actuated walking machine (figure 2). Two sliding frames allow a linear motion and a rotating cylinder is used for changing the motion direction. The upper part of the robot is a two degrees of freedom scanner that can carry different kinds of mine detection sensors. As mentioned in the introduction, to develop efficient detection methods, precise and reliable data are essential. This fact justifies the development of those two systems, which are able to acquire multi-sensor data . The third vehicle called Tridem (figure 1) has a triangular shape and stands on three wheels. The wheels are connected to the frame as a three arms star. The frame supports the electronics and the batteries. The robot is equipped with three driving and three steering

26

motors. The power is transmitted to the driving motors by copper brushes. This solution gives the wheels a complete rotation freedom around the steering axis. System modularity was one of the basic requirements for this robot. The wheels can be removed and replaced very easily because all the wheels are identical and are fastened with fast screw connections. The

electrical connections are also very easy; the wires (control signal and power) are connected through standard DB9 connectors which are embedded into the frame. It is an omnidirectional vehicle that has been designed for highly unstructured grounds. This platform could be used to detect mines in areas which are dangerous for human deminers.

3. Control Architecture

3.1 The components of the

control architecture The whole system has a multi-processing architecture and comprises the following components (figure 6): - The HMI computer, - An embedded computer for data

acquisition and communication with the HMI computer,

- The motion controller (microcontroller), - The visual tracking and location computer.

Figure 6 The general system architecture The following function are distributed among the different processing systems: - Motion control

27

- Processes Synchronisation - Data Acquisition In this application, two kinds of interfaces are needed: communication interfaces (serial, Ethernet) and acquisition interfaces (serial, GPIB, frame grabber). The subsequent subsections give and overview of the different components of the control architecture. The last subsection describes the integration and synchronisation of the different systems and processes.

3.2. The Master PC The control architecture comprises two loops: a high level control loop that runs on a PC and synchronises high level processes and a low level control loop running on a microcontroller which performs common control functions (position and speed control). The Master PC is responsible for the synchronisation of the robot motion with the positioning system and for the synchronisation of the scanning with the data acquisition. It sends configuration and trigger commands to the microcontroller trough a serial link and performs the data acquisition. The role of the Master PC can be played by the HMI PC, when it is directly connected to the vehicles and to the sensors, or by the embedded PC which communicates via a TCP/IP link with the HMI PC. This remote acquisition can for the moment only be performed with the metal detector (RS232 interface). The direct connection is required when acquiring data from the GPR because it is an instrumentation system that cannot be easily displaced. It uses a normal PC HPIB interface card connected to a scope for acquiring data. When a portable sensor is available, it will be possible to easily integrate it into the mobile system. 3.3. The motion and scanning controller The three robotic systems use the same basic electronics; the low-level control is realised with a 32-bit microcontroller that communicates through a serial link with the Master PC. They can be remotely controlled through the same HMI, a program called CoRoDe (Control of Robots for Demining). The HMI manages all control aspects: vehicle and scanning control, communication and data acquisition. A high interactivity and user-

friendliness is reached through the use of efficient programming techniques (see section 3.7). The 32-bit microcontroller is very versatile and powerful; it performs different functions depending on the system: - It controls electrical motors (scanner,

Tridem), - It monitors signals coming from contact

switches, - It generates commands (replacing a PLC)

for controlling the valves on the pneumatic robot,

- It generates analogue signals for the tracks of the caterpillar vehicle,

- It hosts the main communication loop with the Master PC.

The specific characteristics of the control methods are given in the remainder of this subsection. What concerns the Hunter, two control levels can be distinguished: a software only loop that manages communication through the serial interface and the motion control functions which are interrupt-based. The serial port is polled and a switch-case instruction fires the function corresponding to the received code. Main high level control functions for the Hunter are: set scanning limits, set position, set speeds, move to area limits, start scanning ... During the scanning, synchronisation signals are sent to the Master PC to trigger the data acquisition. Two scanning modes are available: a continuous mode and a discrete mode. In the discrete mode, the scanner stops at every step waiting for the end of the data acquisition. The acquisition step can be changed in the set-up dialog box of the HMI. In the moveTo function, the desired goal co-ordinates are copied into shared variables that are read by the motion control functions. These functions are called by interrupts at fixed intervals (a 4 ms control period is used). The PI controller controls the motor speed by using the Time Processing Units (TPU) of the microcontroller which are configured to generate PWM signals. A cascade negative feedback control combining two loops has been implemented: the first loop corresponds to a speed negative feedback control (the inner loop) while the other one is for position negative feedback control (the outer loop). The microcontroller also generates analogue signals to drive the tracks of the Hunter. Two TPU's channels produce two independent

28

PWM signals for right and left tracks. The PWM signals are filtered to generate continuous bipolar analog signasl compatible with the old interface. As the driving motors do not have encoders it actually acts as an open loop controller. In the case of the walking robot, the microcontroller is essentially used as a software PLC. The control is implemented like a finite state machine and all outputs are binary signals. The TPU's are used to read the position signal from incremental linear sensors placed along the pneumatic cylinders. The main loop manages the communication with the master PC. The control of the Tridem is similar to the one of the Hunter but requires all TPU's of the

controller. The TPU's generate six PWM signals and capture the digital encoders' signals from the steering motors. The speed of the driving wheels are measured with tacho-generators whose signals are converted by the internal AD converters. A classical PI algorithm is used for controlling the wheels. Different configurations are possible (figures

7a, 7b and 7c respectively):

- Translate: all wheels are parallel, - Spin: the wheels are tangent to a circle, - Free: the normal motion mode in which

the three wheels are separately commanded to follow arbitrary trajectories.

This later mode is preferred for the detection because it keeps the metal detector pointed along the motion direction.

Figure 7a Translate mode Figure 7b Spin mode Figure 7c Free mode

3.4. Sensors and acquisition interfaces

Three different sensors have been successfully used in the project: a metal detector (MD), a Ground Penetration Radar (GPR) and an infrared camera. A fourth one, a pyrometer, has been tested but abandoned due to unsatisfactory results. The data acquisition process requires different interfaces: the metal detector has a serial interface, the GPR data are read through the GPIB interface of a high speed oscilloscope, the images coming from the infrared camera are captured with a frame grabber (through a cable or a wireless connection). We will see in the subsection 3.7 how the sensors' characteristics influence the

scanning process and the way the control is realised.

3.5. Communication A serial communication allows the transmission of commands between the Master PC and the microcontroller (the transmission speed is 9600 baud). Radio Ethernet links (protocol 802.11) are used to communicate between the HMI PC and the embedded PC.

3.6. Location and tracking A location system determines the robot's positions which are used to automatically generate a map of the detections (Hong 2001). A pan-and-tilt colour camera tracks a coloured

29

target mounted on the vehicle. Every frame the ball object is extracted from the image (figure 8) and an estimation of its position is computed. As the robot moves, the camera follows the target in order to keep it in the centre of the image. Furthermore, the size of the target is kept constant by using a motorised zoom and the value of the zoom gives an estimation of the distance. As the orientation

of the camera is also known, the ball is completely located in the field. In favourable conditions (uniform background, constant illumination) a precision of 30 cm at distances up to 15 m is reached, which is enough for this application. In order to increase the precision of the location process in varying conditions the coupling of the camera with a laser telemeter is now considered.

Figure 8 Colour target identification

3.7. Human Machine Interface

The graphical user interface of the control program CoRoDe is shown in figure 9. This program offers the following functions: - Control of the vehicle, - Configuration and control of the scanning

system, - Configuration of the sensors, - Data visualisation, - Data archiving, - Mapping. The main area is devoted to data visualisation. It can alternatively display a map with successive locations of the robot or the sensors' data. Displayed data can result from two different acquisition methods: - Global acquisition:

Infra-red images Video images (used to view and

record pictures from the scanned area) - Sequential acquisition:

Single values organised as a 2D image (MD)

One-dimensional data merged into a 3D volume (GPR).

The map contains a reference frame and consecutive positions of the robot and scanned areas. The user can switch between either configuration by clicking on a button in the toolbar. Data acquired during the scan process are saved in two different formats: first as binary data for later processing (double for GPR, double word for MD) and as 8 bits grey scale raw images for direct visualisation. The data acquisition, scanning, location computation and vehicle motion are integrated into a sequence that is controlled by the user with button commands lying in a single toolbar. The interface is simple and intuitive thanks to the use of well-known symbols (VCR-like) and standard colours (see figure 8). In this application, it was a requirement to let the user keep the control of the process; at every moment the user can pause, resume or stop the operations.

30

Figure 9 The CoRoDe data visualisation window It is also essential to provide information during internal processing or timeouts. In this case, sensors' data are drawn on the screen as the scanning progresses. The position of the scanner relative to the maximum positions, the status of the scanning sequence and the main options are also presented to the user and regularly updated. Finally the use of additional communication threads (see next section) preserves the interaction with the user interface. The next section provides implementation details about the communication and the synchronisation of processes.

3.8 Control Process, communication and

synchronisation The robotic systems used for sequential data acquisition have a more complex control architecture than the wheeled robot. The division of the control between the microcontroller and the Master PC requires a permanent synchronisation of the processes. The main program runs on a computer under the Windows operating system (OS) (Windows98 was used because at that time it was the only windows version compatible with

the force-feedback joystick drivers, now Windows2000 or XP could be used as well). The communication with the microcontroller runs in its own thread in order to keep the interactivity of the user interface. Each event coming from the microcontroller (begin or end of line, position trigger for data acquisition) through the serial port is intercepted by the communication thread. For each event, corresponding messages are posted to the view window (the program is implemented with the MFC document-view scheme). Each message results in a call to the corresponding function of the window class (examples of such messages are acquiredata, savedata,...). When using the discrete scanning mode, this process is synchronous and the microcontroller blocks until it receives the acknowledgement of the main process. The consequence is that the scanning is not continuous but stops at every step (typically every 2 cm). This was required by the metal detector because when triggered, this sensor has an acquisition latency of about 200 ms which corresponds to 2 cm for a speed of 10 cm/s and consequently to a 4 cm shift between the forward and backward scanning lines. This synchronisation also removes the need to take into account the delays introduced by the OS message queuing mechanism and the acquisition duration. The communication

31

thread then enters in a blocking section and waits until the data acquisition has been performed and the lock released to send the acknowledge message to the microcontroller. Realising the communication between two threads by messages allows to completely uncouple them (the other way is to pass objects' pointers to the control thread and to directly call functions from it). At each step, the position of the scanner in the user interface is updated, data is acquired, plotted and saved. However, this asynchronous control has two drawbacks: the scanning process is slower and it can produce vibrations due to resonance phenomena. As mentioned earlier, the first limitation is also due to the sensors and cannot be avoided without changing the sensors themselves. Introducing random delays to avoid resonance frequencies can eliminate the second problem. Other possibilities for dealing with the acquisition latency is to calibrate the delays and to shift the data every two lines at the end of the acquisition or to always scan in the same direction. The first solution was rejected by the data processing team because it introduces some extra pre-processing and can alter the data. The second solution has as consequence to double the scanning duration and to cancel the benefit of the continuous scanning; it was consequently not considered. What concerns the Tridem, there is no such synchronisation between the robot and the

Master PC because the communication process is essentially asynchronous. The microcontroller receives configuration or motion commands and executes them. An acknowledgement is sent after the reception of the command, the execution is started and a new command is read from the serial port.

4. Results and future work The two systems equipped with scanners have been successfully used to test the GPR prototype and to acquire registered multi-sensors data (figure 10). These data have been very valuable for the searchers working on fusion methods. The wheeled robot has also been demonstrated as a remotely controlled mine detection system. Different elements have contributed to the success of this project: - The use of the same microcontroller for

different robots, - The use of object-oriented programming

techniques and language that allow to easily reuse building blocks in developing new applications or modifying existing ones,

- The hierarchical model of the control, - The user-friendly user interface.

-

Figure 10 Data acquisition using the prototype GPR and a metal detector

32

But despite these benefits, the application presents several limitations: - It is a monolithic application where

control and interface cannot be easily separated,

- It is OS specific, - Adding a robot with new capabilities or a

sensor with new display requirements would demand important modifications of the program.

These drawbacks are not unique for this application but are common characteristics of classic programs. New ways for implementing the control and visualisation software are consequently considered: components and distributed components like CORBA or Java RMI are possible candidates. The requirements for the new architecture are listed below. It should: - Provide an universal User Interface, - Have Plug & Use capability, - Be open to existing and future robots, - Allow the concurrent control of different

robots, - Allow the easy reconfiguration of the

application (addition or removal of components: robots, sensors, users),

- Allow several users to collaborate, - Allow the users to incorporate their own

control algorithms.

5. First Conclusions This summary has presented the results achieved by the Robotic Workgroup within the Belgian funded project Hudem. Robots using different locomotion techniques have been successfully used to acquire sensor data and to test and validate utilisation methods in different scenarios. In this summary, the control and programming architectures of these systems have been described. The use of object oriented techniques for application development and the reuse of the same control hardware have contributed to the success of this project. However, in the next future new programming techniques will allow developers to write more open and reusable applications. At this stage, it has not been proven that robotic detection of mines works better or faster than human deminers. But the obtained results are encouraging and pave the way for an integrated solution that will some day help to solve this terrible plague.

6. Final conclusions. Need for

Robotics in Humanitarian Demining ?

There exists a big difference between the military and the civilian mine clearance. The military demining operations accept low rates of Clearance Efficiency (CE). For these purposes it is often sufficient to punch a path through a minefield. For the humanitarian demining purposes, on the contrary, a high CE is required (a CE of 99.6% is required by UN). This can only be achieved through a ‘keen carding of the terrain and an accurate scanning of the infested areas’ that implies the use of sensitive sensors and their slow systematic displacements, according to well-defined procedures or drill rules, on the minefields. The robots, carrying the mine-detectors, could play here an important role. The automatization of an application such as the detection and removal of AntiPersonnel mines implies the use of autonomous or teleoperated mobile robots following a predefined path, sending the recorded data to their expert-system (in charge of processing the collected data), marking the ground when a mine is detected with a probability of predefined level and/or, possibly removing the detected mine. This automatization is unrealistic : the technologies allowing it exist, but the integration of those technologies in a Robotic System moving in unpredictable outdoor environmental conditions is not yet mature. Several workshops, a.o. organised by the IARP (International Advanced Robotics Programme) allowed a discussion on the possible R&D activities for solving the problem(s): the robotic systems was not felt as the most promising solutions, due to their high cost, the use and maintenance difficulties, the varying (daily changing) terrain conditions, etc. However, specific tasks could be entrusted to mechanical mine disposal systems (or, if efficient, robotised sensor-carriers): 1) the cutting of the vegetation : it’s a

mechanical (tele-operated or not) assistance that doesn’t need research activities, but adaptation of existing mobile cutters,

2) the detection tasks in very dense and dangerous areas : such tasks could imply the realisation of effective multi-

33

legged robots, a difficult long-term challenge,

3) the delimitation of the borders of a suspected area : task that could be entrusted to aerial tools : the EU will probably encourage the R&D focusing on this task: UAVs (Unmanned Aerial Vehicles) could be useful for this aim.

4) the inspection of an area after manual demining or mechanical clearance.

We are convinced that both tasks 2 and 4 may be entrusted to Robotics Systems for so far it will be taken into consideration with the next recommendations: 1) No any R&D may be conducted without a

permanent realistic contact with the End-Users

2) No any ‘dummy’ minefield may lead to realistic trials: experiences conducted in laboratories on ‘sandboxes’ distort the comprehensive perception of the problems.

3) R&D AND parallel trials on real minefields are needed, even limited to minor betterment of existing techniques (example: adaptation of existing agricultural vehicles for cutting the vegetation): some companies , NGOs , etc. are ready to welcome and test proposed solutions. Their involvement from the beginning of a project to its end must be seen as a warranty of success.

4) Improvement of existing sensors by well-trained local demining teams and temporary help of the teams by well-accepted and optimally interfaced (HMI) small light-weight, easy controllable and maintainable mechanical systems for well-defined tasks (vegetation cutting, sensor carrying in high-dense minefields, pre-inspection or post-inspection of a suspected or cleared area, etc.) are welcome for so far they improve the cost-effectiveness and the safety of the demining operations.

Acknowledgements This project would not have been possible without the contribution of all the searchers involved in the Hudem project and without the funding by the Belgian Ministry of Defence and the Secretary for Cooperation and Development. We also want to thank our partners from the European Network CLAWAR (Climbing and Walking Robotics) and from the WG Hudem of the IARP.

I also want to thank my assistant, technical director of our UGV (Unmanned Ground CeVehicle) center, Eric Colon, who composed the major part of this paper.

References Alexandre P., Weemaaels J., Doroftei I. & Colon E., (1998) Development of a high mobility wheeled robot for humanitarian mine clearance, Proc. Robotic and semi-robotic ground vehicle technology, Aerosense - SPIE, Orlando, USA. Baudoin Y. & Colon E. (1999) Humanitarian Demining and Robotics: a difficult challenge, Proc. Clawar99, Portsmouth, UK. Colon E., Baudoin Y., Alexandre P. (1998) Development of mobile robots for mines detection, Proc. Mechatronics98, Skovde, Sweden. Hong P., Sahli H. & Baudoin Y.(1998) Color Target Detection and Tracking, Proc. Clawar ‘98, Brussels, Belgium Hong P. (2001) Visual servoing for robot navigation: application in humanitarian demining, Ph. D. Thesis, Free University of Brussels, Belgium. Milisavljevic N. (1999) Mine shape detection and data fusion considerations, Proc. Hudem Symposium, RMA, Brussels, Belgium. Scheers B. (1999), Development of a laboratory UWP GPR system for minedetection, Proc Hudem Symposium, RMA, Brussels, Belgium.

34

Demining in the Indian context

Akshay Rao Goldspark Bangalore

Over the last Century, India has suffered many conflicts, though the wounds of war may have healed but the danger still lies buried in the ground, waiting to kill and maim even years after the conflict is over. These are the true wounds of war and require a concerted effort by the world community to heal them. India has a long and hostile border where extensive mining and de-mining operations are held due to continued escalation and de-escalation along the border. This as the rest of the world understands is an unjustified problem, but in the race to balance powers across our border this is one of the ways which are justified in stopping hostile personnel from crossing the border into India. A great part of this border is mountainous and rocky with lots of caves and rock faces where it is possible for hostile personnel to hide. Kargil and Afghanistan are examples of this. Surface penetrating radar can find these, thus helping the location of these hostile personnel India and Pakistan have laid large numbers of such mines along their common border since coming close to war over Kashmir in December 2001, the International Campaign to Ban Landmines said in a report. India is not a State Party to the Mine Ban Treaty and uses and produces antipersonnel mines. At 4-5 million it has the fifth largest stockpile of the weapon in the world. Since December 2001, Indian Government forces have been involved in what appears to be the largest mine-laying operation in years and possibly decades.

Following the attack on the Indian parliament on December 13th, 2001 India began the largest mine laying operation along the 2900 km border. Initially only the 440-kilometre Line of Control was mined but following December's attack on parliament, the international borders have also been mined. The regions of mining in India are

1) Mountainous part of Kashmir Valley,

2) Plain and fertile land of Jammu & Punjab

3) Desert of Rajasthan. 4) Landmines are also used heavily by the "Naxalites" in Andhra Pradesh, Unfortunately in the Indian context, India and Pakistan are respectively the fifth and sixth largest landmine countries in the world and have an estimated stockpile of about 11 million landmines between them. Parts of the border between the two countries are already regarded as some of the most densely-mined areas in the world. India and Pakistan have not signed the 1997 Ottawa Treaty banning anti-personnel mines .The United Nations estimates that landmines still kill about 10,000 people a year around the world, and activists said the devices injure about another 10,000.Up to 40 percent of all mine victims are children under 15, according to the United Nations The above statistics are not new, but show how far we have gone into making a “mess” for ourselves. This problem can be solved technologically in a way that does not jeopardize human life. At the current moment there are many technologies that can be used for this purpose but the world is still to a certain extent “groping in the dark” .This groping leads to technological improvements. Like the television, internet and so many other technologies we know, it only gets better, and the vision of improving our “today” leads us into a better tomorrow.

21

There are many technological aspects of this problem. Like the development of many other robots, the robotic de-miner will have to undergo much iteration to get to a more efficient model which can rid the world of landmines – in an economical way. Goldspark is developing these robotic models with our eye on the latest technology. There are two aspects of the de-mining problem. Firstly location of the mine and secondly the defusing of the mine. For the first there are many ways which are being used. Here again we answer another question, do we just locate and bypass the mine or do we also defuse it? Goldspark is developing track vehicles to be used in robotic de-mining. These will be tracked vehicles to help them negotiate uneven and uncertain terrain. Thanks to the help and guidance given by Prof James Trevelyan and Prof. Claudio Bruschini whose advice has been provided within the framework of the EC EUDEM2 project, we are in the process of making a technologically superior vehicle for this purpose. There are many interesting ways of approaching this problem. The development of the Fireant is an example of how the defusing technology improves itself with a proportional reduction in the cost. Bringing down the cost from 1000$ to 15$.also Goldspark is looking at chemical neutralization which will render the mine inactive through chemicals being injected into it. The chemical neutralization will be further achieved by means of dispensing mechanisms which we will mount upon the mobile robot vehicle in order to render the mine useless. The reason Goldspark looks at chemical neutralization is that the army claims that always blasting the mine causes noise which could alert the enemy in a hostile scenario. Therefore chemical neutralization and other non-blasting techniques can be employed. The advantage of this is that the mine is destroyed without actual blasting thereby preventing alerting the enemy in the battle scenario. Also to be used in our de-mining vehicle is a Low-Depth Ground Penetrating Radar. Such de-miners can also be used in their idle time to locate other objects buried in the ground. Our development process would essentially consist of the following. After the conceptual design, performing of Finite Element Analysis on various aspects of the vehicle and its motion, capability of taking shocks and vibrations, also done are necessary analyses for shielding for all parts of the system requiring it. In addition to this the design and validation of the motion control and sensing systems and their integration into the main system, to be accompanied with the adequate design of the Robot Arms, Dispensing mechanisms and other parts which facilitate de-mining. Most important is the final integration of the whole system at which stage the entire system, functions to actually sense the mine and defuse it The Indian Army currently uses conventional de-mining techniques; these are well known as causes of death and incapacitation. Currently the Army is involved in Operation Sarp Vinash, which is to block infiltrators from across the border. This has taken a deployment of an additional 10,000 in two regions of our northern state Efforts are on to introduce technology into this process. We are uniquely placed technologically to handle developments in the field of robotic de-mining. We have been acknowledged the world over for our software prowess. Having companies which possess the capability of making advanced systems requiring an integration of the nature which is required for this product, India already designs and manufactures transmitters and receivers covering from metric to millimetric ranges of the electromagnetic spectrum. Further is the advantage of being strategically placed in the subcontinent to lead this effort in this region. This part of the world is infested with landmines and India can become the leader in this context. The immediate neighbours of India will also be able to benefit from the lead that India will take in this regard. In India, we believe that we can take this lead as this is one of our focus areas. A solution which works in one region can be implemented in another region too. In summary we can say India is one of the countries largely affected by land mines. De-mining in India is being done by the conventional method. Once the lead is taken we can surely reduce the death rates by introducing robots in de-mining process. Goldspark contains the team of experts in this context who can make de-mining an easy task by implementing the latest technologies in an economical way.

22

Humanitarian demining: sensor technology status and signalprocessing aspects

Marc AcheroyDept. of Electrical Engineering, Royal Military

AcademyAvenue de la Renaissance 30, B-1000 Brussels,

Belgium

Abstract— This paper presents the status of sensor tech-nology, including operational characteristics without aimingat being exhaustive. Signal processing aspects and importantlessons on data fusion are also discussed briefly. The detectionis considered as a global process in which the outputs of thesensors, considered as skilled specialists, are integrated in afusion operation.

I. INTRODUCTION

A huge amount of antipersonnel (AP) mines are pol-luting the environment in about 60 countries. Thanks tothe Ottawa Convention, mine clearing operations havebeen organized in a more controlled and effective way.Nevertheless, mine clearance remains a very slow process.It is estimated that, on average, a deminer is clearingan area of 10 m2 every working day if he is usingconventional tools, i.e. metal detectors and prodders. Togive an idea of the extent of the problem, in Cambodia,only approximately 260 km2 have been cleared duringthe last ten years. Therefore, humanitarian mine clearanceoperations must be understood and designed correctly,keeping in mind that their main goal is to provide efficientaid to innocent people, who may be severely injured bythis dreadful pollution.

Furthermore, the analysis of actual demining campaignsnot only reveals the far too long time needed to clearpolluted terrain, but also brings to the fore a far toolarge false alarm rate, the threat of plastic mines (whichare difficult to detect by classical means i.e. by metaldetectors), and the large variety of mine clearancescenarios, depending on the country, the region, theclimate and the place of the pollution (houses in villages,roads, agricultural fields, etc).

The important parameters, which characterize the minedetection problem, are the mine occurrence probability,the detection probability of a given material and the falsealarm probability of a given material [9]:

� The mine occurrence probability in a given positionof a minefield expresses the local mine density ofthat minefield as well. Obviously, it is impossibleto control this parameter because it depends on thereality of the terrain. Nevertheless, this parameter is

very important for assessing the probability of analarm in a given location of the minefield.

� The detection probability is the probability of havingan alarm in a given position of a minefield for agiven detection material, if there is a mine in thatposition. This probability gives indirectly a measureof the non-detection probability of that material aswell.

� The probability of false alarm is the probability ofhaving an alarm, for a given material, in a givenlocation if there is no mine in that location.

The two latter definitions are extremely important tounderstand the humanitarian demining problem and fordesigning demining systems.It is indeed particularly important that the detection prob-ability should be as close as possible to one. It is easyto show that evaluating the detection probability alsoamounts to evaluating the risk of the occurrence of amine which has not been detected. This risk is concernedwith human preservation and is therefore of the utmostimportance. No such risk is acceptable and it is thereforean absolute requirement that a demining system shoulddecrease the probability of such a risk to the lowest upperbound possible.Besides, although one indirectly saves human lives bydecreasing the false alarm risk thanks to the accelerationof the demining operations, the false alarm risk is alsoa question of cost. Indeed, a demining method whichminimizes the false alarm rate results in an accelerationof the demining operations which results in spending lessmoney.

Therefore, any demining operation enhancement mustresult in the highest possible detection probability (closeto one) and in the smallest possible false alarm rate andthat at the lowest price. Generally, it is accepted that themost efficient way for increasing the detection probabilitywhile minimizing the false alarm rate consists in usingseveral complementary sensors in parallel and in fusing

the information collected by these sensors.As a matter of fact, it is imperative to evaluate the de-

tection probability when optimizing the performances of asystem. However, the detection probability, as it is definedbefore, assumes that a mine is present in the consideredposition. Since, during organized trials, the position of themines is well known, the condition of the occurrence ofa mine in the given position where the performances ofa system must be evaluated is always realized. This latterremark is of particular importance because it justifies theorganization of trials and the construction of models, tobe validated by trials, in order to evaluate the detectionprobabilities.

Furthermore, assuming in the following as the firstapproximation that the sensors are independent1, thedetection probability can be maximized by optimizingseparately the design of each sensor and of the associatedsignal processing. Next, it can easily be shown that thedetection probability increases if the number of differentsensors increases and that maximizing the overall detec-tion probability of a set of independent sensors clearlycomes to the same as maximizing the detection capabilitiesof each individual sensor. This justifies the use of severalcomplementary sensors and of data fusion techniques toincrease the detection probability. Among the most citedsensors one finds the metal detectors, the radars and theinfrared sensors. A lot more are presented in section 2.

Finally, the false alarm risk, i.e. the probability ofhaving an alarm if there is no mine, cannot be as easilyevaluated as the detection probability because of the useof data fusion methods which favor the manual or auto-matic cancellation of false alarms. Furthermore, it is verydifficult to evaluate the risk of false alarm because it isvery difficult to define in a general way what is not a mine.In this context, it should be particularly inappropriate thata demining system, whatever it may be, makes decisioninstead of the final user whose own physical security isinvolved. Therefore, a well designed system should helpthe user in the decision making, not by replacing him, butby implementing efficient data fusion methods. For thispurpose, methods which are able to deal with uncertaintyby making proposals including the doubt to the user seemto be promising.

This paper will only consider close-in detection in twosteps. In the first step sketched in section 2, will consistsin describing potential technologies for mines detectionand, when appropriate, their associated processing. Someof these technologies are already in use (e.g. prodders andmetal detectors). Some will be fielded very soon (e.g. theground penetrating radar (GPR)). Others will only appearin the field in the long term because they are not mature

1the independence is a particular case of the complementarity: two in-dependent sensors are complementary, but the contrary is not necessarilytrue.

and still need a lot of developments. The second stepis sketched out in section 3 and consists in fusing theinformation produced by the different sensors with theirdedicated processing tools.

II. SENSOR DESCRIPTION

In this section, it will be tried to describe succinctlysensors of different types without claiming exhaustivity.we will subdivide the description into four categories:

1) Prodders, seismic and acoustic sensors2) Electromagnetic sensors (Metal detector, GPR,

Micro-wave radiometer, Electrical Impedance To-mography, Electrography, Imaging with handheldsensors)

3) Electro-optic sensors (visible, IR, multispectral, hy-perspectral, LIDAR)

4) Explosive detectors (NQR, X-rays, Neutron activa-tion, Biosensors, Trace explosive detection)

The three first sensor categories are not able to discrim-inate between an explosive material and any materialwith the same electro-magnetic, thermal and/or opticalproperties, but often offer good localisation capabilities aswell as 2-D and even 3-D capabilities for some of them.The last category aims at detecting explosive material,often offers poor localisation capabilities and often lacksfor spatial resolution as well as for 2-D or 3-D capabilities.Most of the time these sensors require a long integrationtime, which makes them more suitable as confirmationdevice. In the latter case, they are used in combinationwith sensors of the three first categories.

For each of these categories, a table will describe foreach sensor its status of maturity (“R&D”, “in devel-opment, “in use”), its cost (“Low”, “Low to medium”,“High”, “Very High”), its clearance speed (“Low”, “Lowto medium”, “High”) and its effectiveness (“Unknown”,“Low”, “Low to medium”, “High”).

A. Prodders, seismic and acoustic sensors

Manual prodders can also be enhanced by addition tothem of an ultrasonic sensor at the prodding extremity.This so-called “Smart Prodder” allows to have a betterguess of encountered objects and to exercise less pressureon these objects (in this way, decreasing the risk ofaccidents).

Seismic devices are listening to the response of theground to a shock applied at the ground surface from asafe position.

In the case of an acoustic sensor, an ultra-sonic wave issent into the ground and the backscattering of this waveagainst buried objects and the environment is analysed.The application of this method is limited to soil witha very high level of moisture or in the water (paddyfields) for which the propagation of ultra-sonic wavesis possible. This latter characteristic makes ultra-sonic

sensors complementary to the ground penetrating radar(GPR), the performances of which are optimal when themoisture level of the soil is low.

The operational characteristics of these sensors aredescribed in table II-E.6.

B. Electromagnetic sensors

1) Metal detectors: There exist three families of metaldetector: the first one, based on electromagnetic induction(EMI), sends a primary magnetic signal in the ground in anemitting phase during which it creates eddy currents in theburied metallic objects which in turn create a secondarymagnetic field. During a listening phase, the emission isstopped and the system listens to the secondary magneticfield which induces eddy currents in the coils of thedetector. These currents are characteristic of the buriedmetallic objects and of the soil. There exist two types ofEMI devices: the first one sends a magnetic pulse, thesecond one a continuous wave at different frequencies ina stepped frequency mode.

In the second family, the detector, called magnetometer,mesures the local perturbations of the earth magnetic field.

In the third one, the detector, called gradiometer, mea-sures the magnetic field gradient in a given directiondepending of the sensor configuration.

The most used family of detectors is the first one, basedon EMI.

Surprisingly, the metal detector of the first family(which is the most common detector), considered as animaging device, can also provide very useful informationon the shape of metallic pieces included in mines.

Unfortunately, the point spread function (PSF) of ametal detector is a function of the depth (see Fig. 1) and ofthe nature of the buried metallic object (eddy currents aredifferent in a close and in an open circuit) and the imageformation process is non linear. However, the in depthmodeling of the metal detector behavior as a function ofthe type of buried object by the RMA [13], has shownthat it is possible to derive the depth of a buried objectfrom the original data and thus to derive the correspondingPSF to allow a correct de-convolution (see Fig. 2, 3and 4). Further, information on the symmetry propertiesof the buried metallic objects can easily be extracted.This subject is still under investigation. This interestingconsideration shows that the metal detector, known as acheap mine detection system, remains a promising device.

2) The ground penetrating radar (GPR): Useful defini-tions to understand what follows have to be given first. AnA-scan is a one-dimensional signal taken perpendicular tothe ground surface and is the basic echo signal producedby a GPR. A B-scan is a two-dimensional signal resultingfrom a collection of adjacent A-scans along a straightline horizontal to the ground surface. A C-scan is a

Fig. 1. Metal detector — deconvolution of a metallic ball at 5.5cm and8.0 cm

Fig. 2. Metal detector — deconvolution of a straight metallic wire

two-dimensional horizontal slice (parallel to the groundsurface) in a set of adjacent B-scans.

The GPR includes an emitting system (transmitter) anda receiving system (receiver). The transmitter emits a pulsewave or a continuous wave at given frequencies. The re-ceiver collects the waves backscattered by discontinuitiesin permittivity. Discontinuities can be provoked by buriedobjects like landmines (useful signal) but also by naturaldiscontinuities of the soil (clutter). This means also that aGPR is able to detect plastic objects buried in the ground.There are mainly two important types of GPR dependingthe emitted signal: the first one sends a short pulse intothe ground (Ultra wideband pulse GPR), the second onesends a continuous wave in a stepped frequency mode. Theadvantage of the second type is that it provides directlythe Fourier transform of the received signal and that moreenergy can be send into the ground at a given frequency.

Ground penetrating radars and passive radiometers areintended to function as anti-personnel mine detectors.Their performances depend upon parameters such as typeand texture of soil, soil water content, soil density andoperating frequency. In order to evaluate the performancesof microwave technologies in land-mine detection, theelectrical properties of soils must be extensively evaluated(see [10]).

Current GPR are working in a frequency range com-prised approximately between 0.4 and 6.0 MHz.

Fig. 5 shows a sample A-scan of a PMN mine in loam.

Fig. 3. Metal detector — deconvolution of a metallic open loop

Fig. 4. Metal detector — deconvolution of a metallic closed loop

Fig. 6 presents two B-scans of a PMN mine respectivelyin loam and in sand and a C-scan of a PMN mine in loamat 4cm depth.

Fig. 5. Example of a PMN A-scan in loam

One advantage of GPR systems is that it is possibleto become a 3-D representation of objects buried in theground. In order to recover the correct 3-D shape of buriedobjects, RMA [4][5][6] has developed algorithms based onthe convolution, by modeling the behavior of the GPR inthe time domain. The developed algorithms are faster thanthe classical migration methods and provide better resultsas it can be seen on Fig.7.

3) Microwaves radiometers: The microwave radiome-ter is a passive ground penetrating radar (only receivingantennas are present) which uses the EM waves (a few K)emitted by the sky and reflected on the ground surface andsubsurface. They also can generate clear two-dimensionalimages of surface, shallowly buried and buried objects(metallic and plastic). The spatial resolution and thepenetration depend on the frequency. As for the GPR, theperformances are also depending on the soil conditions:a high level of moisture can largely limit the detection

Fig. 6. From top to bottom, PMN B-scans in sand and in loam and aPMN C-scan in loam

capabilities. In the scope of the HOPE project2, The DLRhas developed a MWR which is capable to work at morethan 32 different frequencies between 1 and 8 GHz. Anexample of images obtained with a manual scan is givenin Fig. (8).

4) Other electromagnetic sensors: In Electricalimpedance tomography, the soil impedance is measuredbetween selected locations on the ground. By solving anon-trivial and non-linear inverse problem, it is possibleto detect anomalies.

In electrography, the corona effect is used to detectexplosives (typically TNT) in a liquid phase.

The operational characteristics of electromagnetic sen-sors are described in table II-E.6.

2the HOPE handheld system, project funded by the European Com-mission, includes a metal detector (Vallon, GmbH & RMA), a steppedfrequency GPR (RST, GmbH) and a multifrequency MWR (DLR), allwith imaging capabilities through the use of a high precision positioningsystem (RMA)

Fig. 7. From left to right: raw GPR 3-D image of a PMN mine, restored3-D image of a PMN mine and restored 3-D image of a 10cm barbedwire with three barbes

Fig. 8. DLR microwave radiometer images of a metallic plate, obtainedwith a manual scan, at 8 different frequencies between 1 and 6 MHz

C. Electro-optic sensors

Classical cameras have a poor detection capability evenfor mines laid on the ground and shallow buried ob-jects. LIDAR and therahertz imaging systems have stillto demonstrate their usefulness for mine detection. Theyindeed use shorter wavelengths than ground penetratingradar and hence suffer from significant limitations in soilpenetration. Further, wild groing vegetation offers a stronglimitation to most of electro-optic devices. Nevertheless,special attention must be paid to hyperspectral, thermalinfrared sensors and Scanning Laser Doppler Vibrometry(SLDV).

1) Hyperspectral sensors: Hyperspectral techniquestake into account the very selective properties of the ma-terial reflectivity. Laboratory experiments[7], in the courseof which very narrow wavelength bands have been used,have demonstrate the capabilities of wavelength tuning todiscriminate between different surface laid materials.

2) Thermal infrared: Mine detection by means of ther-mal infrared sensors can be achieved in two differentapproaches. The first approach consists in measuring theappearing temperature difference of the soil, induced bythe differences in emissivity and/or by the differences inthermal flux due to the presence of a shallow buried orburied object (see Fig. 9and [12]). A second approachconsists to take advantage of the polarisation propertiesof manufactured surfaces, by analysing the informationcontained in the images of shallow buried mines producedby the three Stokes parameters which characterize the po-larisation state, i.e. the degree of polarisation, the azimuthand the ellipticity. Those parameters can be evaluated by

recording by means of a linear polarizer four different im-ages corresponding to four different polarization directions[15].

Fig. 9. From left to right: original image of a buried mine (5 cm belowthe ground surface), preprocessed image, detected edges and originalimage with detected mine-like shape (ellipse)

3) Scanning Laser Doppler Vibrometry (SLDV): Inthis technology, which is not properly said an electro-optic technology, an acoustic power transmitter sends anacoustic wave in the ground. If an object is present in thesoil, at the ground surface a backscattered wave inducessoil vibrations measured by a laser Doppler vibrometer.This technique has been tried by FGAN (Germany) on atest site in ISPRA (European Commission Joint ResearchCentrum). Results are shown on Fig. 10 The operational

Fig. 10. buried mines acquired with a SLDV (FGAN - Germany)

characteristics of electro-optic sensors are described intable II-E.6.

D. Explosive detectors: biosensors

1) Dogs and educated rodents: Actually, one of themost efficient “sensor” for mine detection is the dog. Butit appears that rodents are easier to educate and to feedand that they can work longer than dogs. Furthermore,the rodents are much lighter and have a better olfactorycapacity and a better immunity. The non profit organiza-tion APOPO and the University of Antwerpen (Belgium)are currently making operational tests in six differentcountries, which are representative of the mine threat.Rats have already booked interesting results in Tanzaniain 2002.

Fig. 11. Rodent

2) Artificial nose: Some reseach activities have beendevoted to technologies that try to mimic the olfactorysystem of a dog. But up to now no significative resultshave been booked in the field of humanitarian demining.

Another approach consists in using antibodies sensitiveto TNT. these antibodies are fixed on a quartz crystal.When the sensor is in contact with TNT free molecules,the fixed antibodies leave the crystal. The result is that theweight of the quartz crystal is changed. This weight lossis measured by measuring the crystal frequency change.

A relatively long integration time, due to the extremelylow concentration of explosive vapours in the air, makesthis technology more suited to confirmation than to thedetection itself. Therefore, this technology should be ac-companied by a detection equipment such as a metaldetectors or a GPR or a combination of them.

The operational characteristics of biosensors are de-scribed in table II-E.6.

E. Explosive detection: nuclear and chemical methods

This family of technologies includes NuclearQuadripole Resonance, Thermal Neutron Activation(TNA), Fast Neutron Activation (FNA), Trace ofexplosive detection using chemical processes, X-raybackscattering and X-ray fluorescence. Again, therelatively long integration time needed to detect theexplosive molecules and their high cost make thistechnology more suited to confirmation than to thedetection itself. Therefore, this technology should beaccompanied by a detection equipment such as a metaldetectors or a GPR or a combination of them.

1) Nuclear Quadripole Resonance: Advanced NuclearQuadrupole Resonance (NQR) techniques can be used todetect explosives in any surroundings. The quadrupolecharge distribution of the atom results in alignments ofnuclear spins. A radio frequency pulse (RF-pulse) gener-ated by a transmitter coil causes the excitation of nuclearspins to higher quantized energy levels. When the nuclearspins return to their equilibrium position, they follow aparticular precession frequency. This specifies the atoms

and functional groups in the molecules. Nitrogen is aquadrupole atom that appears in every type of explosive.Because of very distinct NQR frequencies the false alarmrate due to other nitrogen containing materials is extremelylow [1]. Fig. 12 shows respectively a time responseof aNQR system and the corresponding spectrum.

Fig. 12. From left to right, time domain NQR response and correspond-ing spectrum

2) Thermal Neutron Activation (TNA): Explosive ma-terials are rich in Nitrogen-14 (14N) which is the pre-dominant stable isotope of Nitrogen. When a neutron iscaptured by a Nitrogen nucleus the following reactionsoccur:

n� 14 N � 15N ��� 15N

� γ

The excited Nitrogen nucleus (denoted by � ) de-excites toits ground state in less than a picosecond by emitting oneor more gamma-rays with unique energies up to 10.83MeV . The gamma-rays can be detected by conventionaldetectors such as NaI and/or GeLi. Specific software canbe used for signal processing and identification.

3) Fast Neutron Activation (FNA): Fast (14 MeV)neutrons and the associated alpha particles (3.5 MeV) aregenerated by a fast neutron source. The neutrons produceprompt gamma rays in inelastic scattering with the nucleiof the materials. Each neutron is always produced in as-sociation with an alpha particle which are always emittedat 1800 to the neutron direction; hence, direction of eachneutron is known from the direction of each alpha particleassociated with it. Alpha particles are detected with anarray of scintillation detectors; the position of the detectorhit by the alpha determines the alpha direction which,in turn provides the neutron imaging (see Fig 13). Theanalysis of the gamma-rays provides information aboutthe stoichiometric composition of hitted materials in termsof Carbon, Nitrogen and Oxigen. Depth information of agiven neutron is provided by measuring the time of flightof the corresponding alpha particules.

4) X-ray backscattering: X-ray backscattered radiationis detected during active illumination of the ground withX-rays, and basically determines whether or not an objectis made up predominantly of light chemical elements (i.e.low atomic number Z). The technique is intended forbulk explosive detection, although AP mines have beenimaged as well; smaller, man-portable detectors basedon radioactive sources have also been proposed. Thesystems which have been developed are said to be able

Fig. 13. Image produced by a FNA system, the colours represent thestoichiometric composition in C, N and O (image provided by courtesyof High Energy (CA - USA))

to produce a 2D image with a resolution of some cm.Potential problems come from shallow penetration, systemcomplexity, sensitivity to soil topography, sensor heightvariation, and safety aspects due to the use of ionisingradiation [16].

5) X-ray fluorescence: X-rays are not penetratingdeeply in the ground. Therefore, they are most of thetime not detecting directly the explosive encapsulated ina mine but molecules of explosive which are migratingfrom the mines to the ground surface. When migratedexplosive molecules are illuminated by X-rays a series ofchanges occurs in the electron configuration resulting inthe emission of photons characteristic of the material andthat can be captured and analysed.

6) Chemical methods: Chemical methods are also usedto detect explosive in the air and in the ground. The diffi-culty consists in collecting enough explosive molecules inthe air or in solution because of the very low concentrationof explosive available.

The operational characteristics are described in table II-E.6.

III. SIGNAL PROCESSING AND DATA FUSION

For each type of sensor, specific signal processingtechniques are used in order to extract useful information.The used techniques mainly include signal conditioning orpreprocessing (e.g. signal detection, signal transformation,noise reduction, signal restoration and enhancement (see[2], [3] and [8], which are a very important steps beforefurther processing) and pattern recognition techniquesaiming at increasing the expertise of each sensor sepa-rately. Nevertheless, it has been shown in the previoussection that no sensor is perfect for all scenarios and allconditions (moisture, depth, cost, etc).

The analysis of the principles of operation of differentsensors, their complementary information, and the factorsthat affect their operability, have led to the conclusionthat their fusion should result in improved detectabilityand reduced number of false alarms in various situations(different types of mines, of soil, vegetation, moisture,etc ����� ).

Low level fusion can be performed even using a het-erogeneous set of sensors if the data are co-registered.

Our experience has shown that higher level data fusionis possible but accounting for the following facts:

� Learning processes are very difficult and risky be-cause of the inter and intra variability of the scenar-ios. The heterogeneous character of a given minefieldand of the huge set of possible minefields makesgeneralisation unpractical if not dangerous.

� High level fusion must rely on qualitative instead ofquantitative a priori knowledge, therefore methodslike Bayes decision theory will often fail.

� The absolute (objective) confidence in specific sen-sors resulting from extensive trials must be includedin the fusion model (principle of objective discount-ing).

� The relative (subjective) confidence that the deminerhas in specific sensors must also be (interactively)included in the fusion model (principle of subjectivediscounting).

Since in this domain of application one has to deal withuncertainty, ambiguity, partial knowledge, ignorance andqualitative knowledge, it is important to chose for anapproach where they can be appropriately modeled, e.g.belief functions within the framework of the Dempster-Shafer theory. A main motivation for working within thisframework is to be able to easily model and include exist-ing knowledge regarding: chosen mine detection sensors,mine laying principles, mines, and objects that can beconfused with mines [11].

In any case, we need to be aware that the ultimatedecision must belong to the deminer because his life isinvolved.

IV. CONCLUSIONS

The purpose of this paper is to present the statusof sensor technology, including their operational charac-teritics without having the pretension to be exhaustive.Furthermore, this paper presents the detection as a globalprocess wherein the outputs of the sensors, considered asskilled specialists, thanks to their associated processing,can be integrated in a fusion process.

ACKNOWLEDGMENTS

The authors wish to thank all the researchers of theHUDEM3 and MsMs4 projects without whom it wasimpossible to write this paper.

3The Belgian project on humanitarian demining (Hudem) has beenfunded by the Belgian Ministry of Defence and the Belgian StateSecretariat on Development Aid

4The Multi-sensor, Multi-signature project is supported by the Euro-pean Commission Joint Research Centre (ISPRA)

Sensor family Sensor Maturity Cost Speed Effectiveness

Prodder In Use Low Very Low HighProdders & Smart Prodder In Use Low to Medium Very Low HighAcoustic In devel.

Seismic & acoustic R&D High Medium High (in wet soil)EMI devices In Use Low to Medium Low to Medium HighMagnetometer In Use Low to Medium Low to Medium HighGradiometer In Use Low to Medium Low to Medium High

Electro- GPR In Use Medium to High Low to Medium High (in dry soil)magnetic MWR In Devel. Medium to High Low to Medium Medium

Electrical Imp. Tom. R&D Low to Medium Low to Medium UnknownElectrography R&D Low to Medium Low to Medium UnknownVisible OK Low to Medium Medium LowInfrared OK High Medium MediumInfrared Polar. R&D High Medium Medium

Electro-optic PrototypeMulti & hyperspectal R&D High Medium MediumLIDAR R&D Very high Medium LowTerahertz R&D Very high Medium LowSLDV R&D Very high Medium Medium to highDog OK Medium to high Medium to high Medium to high

Biosensors Rodents In devel. Medium Medium to high Medium to highArtificial nose R&D Medium to high Medium MediumNQR R&D Medium to high Medium Medium

PrototypeTNA R&D High Medium Medium

PrototypeFNA R&D Very high Medium Very high

Nuclear & X-ray backscattering R&D High Medium LowChemical Prototype

X-ray fluorescence R&D High Medium to high MediumPrototype

Chemical detectors R&D High Medium Unknown

TABLE I

SENSOR OPERATIONAL CHARACTERISTICS

V. REFERENCES

[1] A. Engelbeen. Nuclear quadrupole resonance minedetection. In CLAWAR’98, pages pp.249–253, Brus-sels, Belgium, November 1998. BSMEE.

[2] A. Pizurica. Multiresolution techniques for im-age restoration in mine detection problems. InCLAWAR’98, pages pp.225–230, Brussels, Belgium,November 1998. BSMEE.

[3] A. Pizurica, W. Philips, I. Lemahieu and M. Acheroy.Speckle Noise Reduction in GPR Images. In Interna-tional Symposium on Pattern Recognition “In Memo-riam Prof Pierre Devijver”, Brussels, Belgium,Februari 1999. Royal Military Academy, RMA.

[4] M. Piette B. Scheers. Short-pulse response of anti-personnel landmines to UWB GPR signals. In Sev-

enth international conference on ground-penetratingradar, Kansas, USA, 1998.

[5] B. Scheers, M. Piette, A. Vander Vorst. The detectionof AP mines using UWB GPR. In IEE Second In-ternational Conference, The Detection of AbandonedLand Mines, Edinburgh, UK, October 1998. IEE.

[6] B. Scheers, M. Piette, M. Acheroy and A. VanderVorst. A laboratory UWB GPR system for landminedetection. In GPR2000, Sidney, Australia, June 2000.

[7] F. Dupuis. Acquisition d’images hyperspectrales.Master’s thesis, RMA, Brussels, Belgium, 1998.

[8] L. van Kempen, H. Sahli, J. Brooks, and J. Cornelis.New Results on Clutter Reduction and ParameterEstimation for Landmine Detection Using GPR.In GPR 2000, Eighth International Conference onGround Penetrating Radar, Gold Coast, Australia,

Australia, May 2000.[9] M. Acheroy. L’infrarouge thermique, principes et

applications au deminage humanitaire. Revue HF,(3):p. 13–24, 1998.

[10] M. Storme, I. Huynen, A. Vander Vorst. Characteri-zation of wet soils from 2 to 18 GHz, experimentalresults. In CLAWAR’98, pages pp.237–239, Brussels,Belgium, November 1998. BSMEE.

[11] N. Milisavljevic, I. Bloch, and M. Acheroy. A firststep towards modeling and combining mine detectionsensors within Dempster-Shafer framework. In The2000 International Conference on Artificial Intelli-gence (IC-AI’2000), Las Vegas, USA, 2000.

[12] N. Milisavljevic. Comparison of Three Methods forShape Recognition in the Case of Mine Detection.Pattern Recognition Letters, 20(11-13):1079–1083,1999.

[13] P. Druyts and L. Merlat and M. Acheroy. ModelingConsiderations for Imaging with a Standard MetalDetector. In Detection and Remediation Technologiesfor Mines and Minelike Targets V, volume 4038,pages 1431–1451, Orlando, FL, USA, April 2000.SPIE Press.

[14] P. Verlinde, M. Acheroy, G. Nesti, and A. Sieber.First Results of the Joint Multi-Sensor Mine-Signatures Measurement Campaign (MsMs Project).In Detection and Remediation Technologies forMines and Minelike Targets VI, volume 4394, Or-lando, FL, USA, April 2001. SPIE Press.

[15] D.L. Jordan G.D. Lewis. Infrared polarisation signa-tures. In Proceeding NATO-IRIS Vol. 41 N5, pagesp.71–82, BRUSSELS, 1996. NATO-IRIS.

[16] EUDEM2 - Mine Action Technology website -http://www.eudem.vub.ac.be.

TRAJECTORY TRACKING BY USING FUZZY LOGIC CONTROLLER ON MOBILE ROBOT

A. Shala, R. Likaj, A. Geca, A. Pajaziti & F. Krasniqi,

shalaa@un.org, ramelikaj@yahoo.com, ageca@yahoo.com, agronpajaziti@hotmail.com, fehkrasniqi@yahoo.com. Mechanical Engineering Faculty

Pishtina - Kosova Abstract: In this paper a Conventional PD Controller and Fuzzy Logic Control for trajectory tracking of mobile robot platform is presented. Using computed torque method is given input – torque on mobile robot wheels. The locomotion control structure is based on integration of kinematics and dynamics model of mobile robot base. The proposed Control scheme and Fuzzy Logic Algorithm could be useful for building an autonomous non-destructive testing system based on wheeled mobile robot. Key words: mobile robot, Conventional, Fuzzy Logic, trajectory tracking, platform, wheel INTRODUCTION Recently, the development of engineering technology drives people to devise goods for convenience in everyday life, and the research utilizing the robots for the better of the life is conducted in many areas. And such robots are applied not only at the industrial sites but also at the houses, medical uses, and etc. However, due to the diversity and complexity of the project formation, different from the former robots, which performs simple repetitive works only, the robots with the ability to make appropriate decisions as to fit the situation and those with ability to appropriate correspond are needed. Thus, the system proposed in this paper was constructed in order that the mobile robot can follow desired trajectory. It was made possible by using the Conventional PD Controller and Fuzzy Logic Control. KINEMATICS MODELING We have use for modeling mobile robot with two degrees of freedom (DOFs): x - translation and either y - translation or z-rotation.

Cxv

CyvCv

Czω

b

a C

1

2

r

yC

xC

y

x ϕO

Figure 1: Kinematics coordinate system assignments. The nonholonomic constraint states that the robot can only move in the direction normal to the axis of the driving wheels i.e., the mobile base satisfies the conditions of pure rolling and non-slipping:

0sincos =⋅−− CzCxCy bvv ωϕϕ ................. (1) From Fig. 1:

ϕωϕ sincos ⋅⋅−= CzCCx bvv ϕωϕ cossin ⋅⋅+= CzCCy bvv

Czωϕ =& or in matrix form:

vqJq ⋅= )(& .............................................. (2)

44

⎟⎟⎠

⎞⎜⎜⎝

⎛⋅⎟⎟⎟

⎠

⎞

⎜⎜⎜

⎝

⎛⋅⋅−

=⎟⎟⎟

⎠

⎞

⎜⎜⎜

⎝

⎛

Cz

CCy

Cx vbb

vv

ωϕϕϕϕ

ϕ 10cossinsincos

&

⎟⎟⎟

⎠

⎞

⎜⎜⎜

⎝

⎛=

⎟⎟⎟

⎠

⎞

⎜⎜⎜

⎝

⎛=

ϕϕ &

& Cy

Cx

C

C

vv

qyx

q ; , and ⎟⎟⎠

⎞⎜⎜⎝

⎛=

Cz

Cvv

ω

⎟⎟⎟

⎠

⎞

⎜⎜⎜

⎝

⎛ϕ⋅ϕϕ⋅−ϕ

=10

cossinsincos

)( bb

qJ .

where: - are translation velocities of the CyCx vv ,

robot body, Czω - is the robot z-rotational velocity,

1wω , 2wω - are wheel rotational velocities, r - is actuated wheel radius,

ba, - are distances of wheels from robot axes and

)(qJ - is Jacobian matrix. Constants a, b and r should be set according to the robot proportions. Implicit values are: a =20 cm, b = 5 cm and r =10 cm.

b

a R

C

C’

trajectory

][616.2022min cmbaR =+=

Figure 2: Minimal trajectory radius during turning left or right. DYNAMIC MODELING The Lagrange formulations are used to derive the dynamic equations of the mobile robot.

τ=∂∂

−⎟⎟⎠

⎞⎜⎜⎝

⎛∂∂

qL

qL

dtd

&....................................... (3)

where:

⎟⎟⎟

⎠

⎞

⎜⎜⎜

⎝

⎛=

ϕC

C

yx

q - generalized coordinates,

pk EEL += - Lagrange function,

⎟⎟⎠

⎞⎜⎜⎝

⎛=

2

1

ττ

τ - torque on wheels 1 and 2.

In our case 0=pE because the trajectory of the mobile robot base is constrain to the horizontal plane. That means the system can’t change its vertical position. After the calculation of kinetic energy, the dynamical equations of the mobile base in Fig. 1 can be expressed in the matrix form:

λτ ⋅+⋅=⋅+⋅ )()(),()( qAqBqqqHqqD &&&& ....... (4) where:

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

ϕ⋅⋅−ϕ⋅⋅ϕ⋅⋅−+ϕ⋅⋅+

=IbmbmbmmM

bmmMqD

ww

ww

ww

cossincos0

sin0)( ,

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡ϕ⋅ϕ⋅⋅ϕ⋅ϕ⋅⋅

=000

sin00cos00

),( &

&

& bmbm

qqH w

w,

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

−ϕϕϕϕ

=aa

rqB sinsin

coscos1)( ,

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡

−

−=

bqA ϕ

ϕcossin

)( and

ϕ⋅ϕ+ϕ−=λ &)sincos( CyCxw vvm , where: M – mass of mobile robot platform, mw – mass of wheel, all three wheels has same masse. I – moment of inertia for mobile robot. The mass and inertia of the wheels are considered explicitly. INVERSE DYNAMIC As we know this consists on that for known inputs (torques on wheels), find outputs (position and velocity of mobile robot). Based on equation (2), equation (4) become:

λ⋅+τ⋅==⋅⋅+⋅+⋅⋅

)()()(),())()(()(

qAqBvqJqqHvqJvqJqD &&&

.... (5)

where: )(qJ& - derivate of Jacobian.

Based on equation (5), is needed to find : v&

λ⋅+τ⋅+⋅⋅⋅−⋅⋅−=⋅⋅

)()()( ),()()()]()([

qAqBvqJqqHvqJqDvqJqD &&& (6)

45

As we know, mobile robot is redundant and have positions of singularities, equation (6) become:

{ }

])()( )(),()()([

)]()([)]()([)]()([1

λ⋅+τ⋅++⋅⋅−⋅⋅−

⋅⋅⋅⋅=−

qAqBvqJqqHvqJqD

qJqDqJqDqJqDv TT

&&

&

(7)

CONTROL DESIGN a. Conventional PD Controller The trajectory tracking problem for mobile robot is passed as follows. Given is a desired trajectory:

dx , , respectively velocities: dy dϕ

dx& , , . dy& dϕ&

Based on equation (2): ddd vx ϕ= cos& , , . ddd vy ϕ= sin& dd ω=ϕ&

or .............................................. (8) T

dddd yxq ] [ ϕ= , . Tddv ] [ ω=dv

The proposed control scheme for the mobile robot trajectory tracking is presented in Fig. 3. Computed torque input is defined as follow.

)(tτ

Output from mobile robot is his velocity: Tv ] [ ω=v . Based on equation (2) we have:

v⋅= Jq& and with integration is founded q. Trajectory tracking error is:

)( qqTe de −= , where

⎥⎥⎥

⎦

⎤

⎢⎢⎢

⎣

⎡ϕϕ−ϕϕ

=1000cossin0sincos

eT , and the auxiliary

velocity control input achieves tracking is given by:

⎥⎦

⎤⎢⎣

⎡++ω+

=3322

111sin

cosevkevk

ekev

ddd

dcv ................. (9)

Velocity tracking error is defined as: vve cc −= .

MOBILE ROBOT J

PD

∫q&

vc

qd yd (xd) y (x)

FuzzyLogic

Controller

ϕ d

v T e

vd

Figure 3. Fuzzy Logic trajectory tracking control scheme.

This error now is input to conventional PD controller and output from that is:

ce⋅= ] [)( vp KKtu ................................... (10) u(t) represent torque on wheels generated by conventional PD controller. Conventional PD controller is used both as an ordinary feedback controller to guaranty asymptotic stability during motion period, and as a reference model for responses of the controlled mobile robot.

b. Fuzzy Algorithm The fuzzy controller in our approach is based on the fuzzy control principles. Each input space is partitioned by fuzzy sets as shown in Figure 3. To obtain the desired response during the tracking-error convergence movement by compensating for nonlinear model dynamics, a Fuzzy Logic Controller is designed to become a nonlinear controller.

46

Inputs for FLC are: - trajectory tracking-error or distance error:

)()( xyxyd dd −=

Figure 4.1. Input variable 1 – Distance

Error (d) Where for input 1- distance error: NVL – distance is negative very large, NL – distance is negative large, NM – distance is negative medium, Z – distance is zero, PM – distance is positive medium, PL – distance is positive large, PVL – distance is positive very large, Angle error:

ϕ−ϕ=ϕ de .

Figure 4.2. Input variable 2–

Angle Error (e

ϕ) Where for input 2 – angle error: a-NL – angle error is negative large, a-NM – angle error is negative medium, a-Z – angle error is zero, a-PM – angle error is positive medium, a-PL – angle error is positive large. Outputs from FLC is additional torque on wheels Fig. 5.

Figure 5.1. Output 1 – additional torque for wheel 1.

Where for output 1 – additional torque to wheel 1: T1-Z – additional torque is zero, T1-S – additional torque is small, T1-M – additional torque is large, T1-VL – additional torque is very large.

Figure 5.2. Output 2 – additional torque for wheel 2.

Where for output 2 – additional torque to wheel 2: T2-Z – additional torque is zero, T2-S – additional torque is small, T2-M – additional torque is large, T2-VL – additional torque is very large. Here, asymmetrical triangular function which allow a fast computation, essential under real-time conditions, are utilized to describe each fuzzy set. The fuzzy sets of the two inputs dj, and ϕj is calculated thru 27 rules and the final output of the unit is given by a weighted average over 27 rules. Tabele 1. Rules proposed by Authors

Angle Error Distance Error a-NL a-NM a-Z a-PM a-PL NVL T1-Z

T2-VLT1-Z

T2-VL T1-Z

T2-VL - -

NL T1-Z T2-L

T1-Z T2-L

T1-Z T2-L

- -

NM T1-Z T2-M

T1-Z T2-M

T1-Z T2-M

- -

NS T1-Z T2-M

T1-Z T2-S

T1-Z T2-S

- -

Z T1-Z T2-M

T1-Z T2-S

T1-Z T2-Z

- -

PS - - T1-S T2-Z

T1-S T2-Z

T1-S T2-Z

PM - - T1-M T2-Z

T1-M T2-Z

T1-MT2-Z

PL - - T1-L T2-Z

T1-L T2-Z

T1-L T2-Z

PVL - - T1-VL T2-Z

T1-VL T2-Z

T1-VLT2-Z

47

SIMULATIONS

0 20 40 60 80 100 120 140 160 180 200

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

Time (Seconds)

Ang

le [r

ad]

10-2

Figure 6.1. Angle Error with PD without FLC

0 20 40 60 80 100 120 140 160 180 200

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Time (Seconds)

Ang

le [r

ad]

10-3

Figure 6.2. Angle Error with PD and FLC

0 20 40 60 80 100 120 140 160 180 200

-3

-2.5

-2

-1.5

-1

-0.5

0

0.5

Time (Seconds)

Dis

tanc

e [c

m]

Figure 7.1. Distance Error with PD without FLC

0 20 40 60 80 100 120 140 160 180 200

0.3

0.4

0.5

0.6

0.7

0.8

Time (Seconds)

Dis

tanc

e [c

m]

Figure 7.2. Distance Error with PD and FLC Figure 8 shows using FLC and PD designed for path-planning, where mobile robot have e sensor which is giving information’s about distance and angle of “obstacles-green lines”.

Figure 8. Simulated Trajectory of mobile robot for detecting a Area

48

CONCLUSIONS AND FUTURE WORK This paper has compared the design procedure and performance of a PD controller and a Fuzzy Logic Controller used for trajectory tracking a constraint robot system. The paper has achieved the objective of constructing a simple controller that mimics the handling a task that requires coordination of the limbs from a human motor control perspective. The rules for the

FLC have decrease error. The simulation results carried out showed the effectiveness of the proposed control strategy. The disadvantage of having a Fuzzy Logic Controller is number of rules and combinations to have effective trajectory tracking. How would the controller react to the additional systems and trajectories? This and other points such as robustness will be considered in future work.

REFERENCE [1] David Kortenkamp, R.Peter Bonasso, and Robin Murphy, Artificial Intelligence and Mobile Robot, AAAI Press/The MIT Press. [2] Dei-Jeung Huh, Woo-Young Lee, Uk-Youl Huh, "A Study on Intelligence Navigation for Autonomous Mobile RobotUsing Fuzzy Logic Control" ICCAS 2001. [3] Young Ki Kim, “A Method of Obstacle Avoidance for a Mobile Robot using Fuzzy Reasoning", Inha University, 2002. [4] Using MATLAB Graphics. The MathWorks, Natick, MA, USA, 2000. [5] Getting Started with MATLAB. The MathWorks, Natick, MA, USA, 2000.

49

UTILISATION OF BURMESTER THEORY IN DESING OF PLANAR MECHANISM/MANIPUTALOR

I. Gojani, A. Pajaziti, B. Pira

Mechanical Engineering Faculty, University of Prishtina, Kosova P. Marango

Politechnical University of Tirana, Albania

Abstract: A lot has been done so far in the use of Burmester curves in designing four-bar mechanisms that prescribe three and four positions but not so much for five positions and even more in design of six-bar mechanisms prescribing four and five positions. This paper briefly describes a way to utilize Burmester curves in design of planar mechanism/manipulator prescribing five precision points as a part of a mobile robot. In this paper we would like to present an application of Burmester Theory in practice in design planar mechanisms/manipulator that prescribes five positions using. 1. INTRODUCTION Looking at the world around, approximately every few seconds one encounters a mechanism of some kind without being aware of its presence. Although the wheel is one of the greatest inventions of the man, manipulating with it was just as well challenging. For any object that’s needs moving from one position to another a mechanism of some sort is required. In other words, mechanisms ore nothing more than a matter combined with a simple geometry. Although mechanisms can be tree dimensional, most of them are limited in two dimensions. The objectives of this paper are to utilize the Burmester Theory in a design of a four-bar mechanism that will prescribe five precision points as well as keep the desired angle of the element moved by the mechanism. An example where the Burmester Theory is utilized will be presented with this paper. 2. BURMESTER THEORY When a body moves through two and three prescribed positions there are three respectively two infinities of solutions in the design of the mechanism; however, when the body passes through four and five prescribed positions then there is one-infinity respectively few possible designs of a mechanism. Burmester found the solutions to problem of four prescribed positions of the plane using complex numbers [1], figure 1.

50

Figure 1: The unknown dyad’s W,Z which can guide the moving plane π from the first to the j-th position

Notice that the vectors defined above form a closed loop, including the first and j-th positions:

Weiβj+Zeiαj-δj-Z-W=0 (1)

For four positions the above formula becomes:

W(eiβ2-1)+Z(eiα2-1)=δ2

W(eiβ3-1)+Z(eiα3-1)=δ3 (2)

W(eiβ4-1)+Z(eiα4-1)=δ4

After the compatible sets of β2 are obtained, then any method for solving simultaneous complex equations can be used to find Z and W. Then the circle point k and the center point m are given by the following expressions (figure 1).

k = R1 - Z m = k – W

The above pair of these two points, is called Burmester Point Pair (B.P.P.), and the locus of these pairs for different values of β2 form the Burmester Curves

3. APPLICATION OF BURMESTER THEORY Burmester Theory utilized by Chan Y.M. in 1994 with a specially written computer program [2]. It calculated circle-point and centre-point circles (specific for Burmester Theory) for four positions, figure 2 2. This program was later modified initially by Noussas D. and finally by the coauthor of this paper, Pira B., for mechanisms that prescribe five precision points [3, 4]. Using the Burmester curves for four prescribed positions, a set of centre-point and circle-point circles, show the solution, figure 2. For five positions, if a solution exists, the solutions lies by superimposing two sets of four positions, position 1-2-3-4 and 1-2-3-5. These two solutions can yield the solution to the 5-position problem by finding the intersections of the centre and circle curves, figure 3.

Figure 2: Burmester Curves for 4 positions

Centre pointsCircle points

Centre pointsCircle pointsCentre pointsCircle points

Figure 3: Burmester Curves for 5 positions

4. CASE STUDY Figure 4 shows a small bucket/gripper containing various explosives (inactive or even active) that is to be dumped into a large container. There is a need to synthesize a four-bar mechanism, as a part of mobile robot, which will lift the bucket/griper and dump its content into the container. Ground points of the mechanism are attached to the container. The container has to go through five prescribed precision points at a specific angle for every position.

51

Figure 4: Case study Once the computer program is executed, two sets of curves are drawn. At the intersection of two centre-point curves (two intersection) two ground points of the mechanism are identified while at the intersection of the of two circle-point curves two moving points of the mechanism are also identified, figure 5.

Figure 5: Ground points and moving points By connecting these points together a four-bar mechanism is designed that will prescribe all five precision points, figure 6, with A and B as ground points and A1 and B1 as moving points.

Figure 6: Mechanism/manipulator in its first position To make sure that the above mechanism prescribes all five positions, DeMech computer software was used to animate its movements. Figure 7 shows the mechanism in its five positions.

Figure 7: The mechanism/manipulator in its 5 prescribed positions

5. CONCLUSION There have been several approached to design four-bar mechanisms that prescribe five precision points using Burmester Theory. This method shows how useful and how easy is to design planar mechanisms using Burmester curves. Although it looks as if it is very strait forward to design four-bar mechanisms using Burmester curves, six-bar mechanism seem to be a different story all together. Regardless of that, we are going to try to use Burmester curves to design six-bas mechanisms in the future. 6. REFERENCES [1] Sandor G. N. & Erdman A. G. “Advanced

Mechanism Design: Analysis and Synthesis Volume 2”, PRENTICE HALL 1984, pp263-299.

[2] Chan Y. M. “Some Case Studies in Mechanism Design MSc thesis” Imperial College Mechanical Engineering Department 1994.

[3] Noussas D. “Case Studies in Mechanism Design MSc thesis” Imperial College Mechanical Engineering Department 1996,

[4] Pira B. “The use of Burmester Curves in Design of Planar Mechanisms” Imperial College Mechanical Engineering Department 1999.

1

5

4

3

2

x

Centre points

Circle points

Centre points

Circle points

First Position

Second Position

Third Position

Fourth Position

Fifth Position

Ground Points

Moving Points

B1

B

A