Navigation Inside Pipelines with Differential Drive Inpipe Robot

Transcription

1 avigation Inside Pipelines with Differential Drive Inpipe Robot Se-gon Roh School of Mechanical Engineering, Sungkyunkwan University 300, Chonchon dong, Jangan gu, Suwon, Kyonggi do, Korea, , 1 Abstract In this paper, a new approach for moving an inpipe robot inside underground urban gas pipelines is proposed. Since the urban gas supply system is composed of complicated configurations of pipelines, the inpipe inspection requires a robot with outstanding mobility and corresponding control algorithms to apply for. In advance, this paper introduces a new miniature miniature inpipe robot, called MRISPECT (Multifunctional Robotic crawler for Ipipe inspection) IV, which has been developed for the inspection of urban gas pipelines with a nominal 4-inch inside diameter. Its mechanism for steering with differential drive wheels arranged three-dimensionally makes itself easily adjust to most pipeline configurations and provides excellent mobility in navigation. Also, analysis for pipelines with fittings are given in detail and geometries of the fittings are mathematically described. It is prerequisite to estimate moving pattern of the robot while passing through the fittings and based on the analysis, a method modulating speed of each drive wheel is proposed. Though modulation of speed is very important during proceeding thought the fittings, it is not easy to control the speeds because each wheel of the robot has contact with the walls having different curvatures. A new and simple way of controlling the speed is developed based on the analysis of the geometrical features of the fittings. This algorithm has the advantage to be applicable without using complicated sensor information. To confirm the effectiveness of the proposed method experiments are performed and additional considerations for the design of an inpipe robot are discussed. I. Introduction PIPELIES Are the major tools for the transportation of oils and gases and a number of countries employ pipelines as main facilities. Recently, there happen a lot of troubles caused by aging, corrosion, cracks, and mechanical damages from third parties. Continuous activities for inspection, maintenance and repair are strongly demanded. However, those activities need enormous budgets that may not be easily handled by related industries. For these reasons, the application of the robot for the maintenance of the pipelines appears to be one of the most attractive solutions to be delivered now. Inpipe robots can be classified into several elementary forms according to the locomotion mechanism as shown in Fig. 1. Most of them have been designed based on specific tasks. As shown in Fig. 1(a), for example, the pig is one of the most well known commercial one that is passively driven by fluid pressure inside oil pipelines and employed for the inspection of pipelines with large diameters [1]. The wheel type illustrated in Fig. 1(b) is similar to the plain mobile robot and a number of commercialized can be found [2]-[8]. Fig. 1(c) shows the crawler type robot with caterpillars instead of wheels [9]. As shown in Fig. 1(d), the wall press type denotes the robot with flexible mechanism for pressing the wall whatever means that has advantage in climbing vertical pipelines. As depicted in Fig. 1(e), walking type possesses articulated legs and it can produce various motions [10]. The inchworm type given in Fig. 1(f) is usually being developed for pipelines with very small diameter [11], [12]. Screw type one takes the motion of a screw when it advanced in the pipelines as depicted in Fig. 1(g) [13], [14]. Most of inpipe robots take the mechanism derived from one of those basic mechanisms or their combinations. The goals of the inpipe robot have close relations with taskspaces of specific applications. The principal requirement of the inpipe robot is that the robot has to explore anywhere it would go within its taskspace. Existing robots generally travel along This paper was presented in part at the IEEE International Conference on Robotics and Automation (ICRA2001, Seoul, Korea, 2001, and ICRA2002, Washington D.C., USA, 2002).

2 2 = 2 EC J F A > 9 D A A J F A? + H = M A H J F 9 = F H A I I J F A A 9 = E C J F A B1? D M H J F A C 5? H A M J F A Fig. 1. Classification of inpipe robots horizontal pipelines smoothly, but some of them move along vertical pipelines or elbows(bend pipes, L-shaped pipes). Furthermore, a small minority of them can achieve the selective navigation in either common type of branches(t-shaped pipes) or branches with the radical radius of 1.0D curvature. For effective inspection, however, inpipe robots should have the ability to negotiate elbows and branches, since there exist a number of those special fittings in pipelines; especially urban gas pipelines. )? JEL A = H JE? K = E J, H EL E K A 8 8, H EL E C L A D E? A, EH A? JE L A D E? A 5 JH = EC D J, H EL A 8 = G E J 6 K H, H EL A 8 E = ) H JE? K = J F A >, EBBA H A H EL A J F A Fig. 2. Steerable robots Up to now, several uncommon robots with steering capability have been reported which are classified into typically an articulated type and a differential drive type as shown in Fig. 2. The articulated type is for the robot with active articulated joint like the snake which may be one of the most adequate mechanism configuration though the steering mechanism becomes complicated such as double active universal joint [18], [19], rubber gas actuate joint [11] and oblique swivel joint [5]. As an alternative approach, the differential drive one as shown in Fig. 2 (b) enables the steering easily by modulating the speed of driving wheels although analysis of pipelines configuration should be accompanied. For last several years, we have developed inpipe inspection robots for underground gas pipelines, called MRISPECT series which are mainly focused on the capability of steering [18], [19]. MRISPECT IV introduced in this paper has been developed as the fourth prototype. Though it shares some aspects with previous MRISPECT series, most of the mechanism has been renewed to adequate for 4- inch underground gas pipelines because miniaturization of former robots is not suitable for pipelines with small diameter; for example, the small sized clutch which is a structural element of steering joint is not found easily. MRISPECT IV has the link construction capable of being folded forward and backward independently, and adopts a new steering mechanism which can drive the robot to travel along the selected direction in the

3 3 branched pipelines. As a result, MRISPECT IV has outstanding mobility in complicated configurations of pipelines such as gas pipelines, which is the taskspace of the robot, even if hardware components are simpler than those of previous versions. Its mobility, however, is mainly due to the effective design based on the careful consideration of pipelines and fittings as well as the navigation strategy developed by ourselves. In this paper, MRISPECT IV is introduced in advance. Its mechanical design and considerations on navigation in pipelines are addressed. In the second, strategies for navigation in fittings are proposed. The geometrical features of elbows and branches are analyzed and described with mathematical expression. Based on these works, methods for determining velocities are developed, and its effectiveness is evaluated. In the last, in addition to the brief reference to system setup, critical and practical considerations with respect to the inpipe robot of differential drive type are discussed. II. verview of MRISPECT IV 5 F H E C 5 J F F A H 5 E@ A H E = C A I + A JH = E = C A I K F F H JA H = 8 EA M B > BH= A 4 A = H M D A A I A J. H JM D A A I A J, H EL E K A ' & # & #? 5 F A? EBE? = JE + +, M D A A I A J Fig. 3. MRISPECT IV As shown in Fig. 3, MRISPECT IV is largely composed of three parts ; 1) a body frame that mounts a CCD camera assembly and driving modules with foldable linkages, 2) three driving modules located circumferentialy with 120 apart, and 3) a camera assembly for the navigation and the visual inspection of the pipelines. Its maximum speed, weight and tractive force are 9m/min, 0.7kgf and 10kgf, respectively. It travels in elbows and steers in branches by modulating speeds of driving modules without articulated steering joint devices. Link mechanism plays a role in the steering action. This makes the wheels of each driving module contact against the wall of pipelines effectively, so that the robot can exert driving force enough for steering. Each part of the robot is briefly described in this section. A. Body Frame The body frame is designed as illustrated in Fig. 4 to exert sufficient traction forces and flexibility in navigation. Three driving modules are attached at the end of the legs on the body frame. From the kinematic analysis of the linkage the following relations are derived y = x tan θ xy = L 2 1 x 2 (1) where x and y represent the displacement along x and y directions, respectively, L 1 denotes the length of the linkage, and θ xy is the rotation angle of the linkage. Such kind of structure provides many advantageous

4 4 4 A = H M D A A L E EH A? JE B9 D A A = EI. H JM D A A J H? = I E C, G 5 F H E C #, Fig. 4. Link mechanism feature in moving horizontal and vertical pipelines, valves, flanges, reducers, elbows, etc., and is particularly essential for travelling along branches. Two wheels of one driving module move independently along the direction of y-axis because there exist the partial constraint of links and the interaction between elastic force at the center spring of the robot and reaction forces from the wall. Distance between the center shaft of the robot and the wheel varies according to the action of the link. It makes the wheel effectively contact with the inside wall of pipelines and assure the robot of stable travelling as well as providing sufficient traction forces no matter how much the diameter of pipelines changes. Based on kinematics, motor casing connecting the front wheel and the rear wheel force each wheel not to move independently. From the view of engineering, however, motor casing hardly effects the locomotion of the wheel set. Thus, it should be noted that motor case illustrated in many figures of this paper represents not the linkage with joints but the mere casing of a robot. The front wheels of driving modules constrain each other. Three wheels in front of driving modules are named as front wheel set as shown in Fig. 3 and rear wheel set has the configuration symmetrical to front wheel set. The constraint of the wheels inside each wheel set makes the robot avoid the influence of gravity. Also, it lets the central axis of the robot coincide with that of the pipelines and forces the robot have sufficient contact with the pipeline wall when the robot turns in branches. B. Driving Module 4 A = H = K EE= H M D A A 1@ A M D A A J H -? A H M 4 A = H M D A A. H J = E M D A A ) EI B H F M A H JH = I EI I E 5 F K H C A = H 9 H M D A A 9 H C A = H M. H J M D A A = + BEC K H= JE > JE B M M J H M 5 F K H C A = H M 9 H C A = H Fig. 5. Driving module The driving module largely consists of a geared DC motor(maxon, 4.5W) with an encoder, several wheels and casings as shown in Fig.5(a). The front wheel and the rear wheel are driven simultaneously by gear transmission as shown in Fig.5(b), where ωs denoting the vectors of the rotating directions of the transmission units. As the driving module can be easily disassembled from the body frame, the convenience in maintenance is ensured. Driving units can be controlled independently and thus they amplify traction forces as well as provide steering capability. MRISPECT IV can negotiate vertical elbows with sufficient traction force because it has three driving modules capable of having contact with the elbow, where the number of three is suitable for the full contact of curved inner wall and this will be mentioned after.

5 5 C. CCD assembly + +, I K > M D A A EC D J + +,? = A H = * A = HE C + +, M D A A 4 E C I EJA B+ +, M D A A I A J Fig. 6. Construction and function of CCD wheel set As shown in Fig. 6 the CCD camera assembly is composed of a CCD camera, three lights for illumination, a frame and a special mechanism called for the CCD wheel set. This set assists the robot in steering in a branch. When the robot turns in a branch, This set has a contact with the wall of the pipeline and eight couples of small wheels rotate along their own axes. Simultaneously this set itself rotate along the x-axis shown in Fig. 6. Such locomotion guides the robot to the desired direction. The CCD wheel set prevents the body of the robot from having contact with the wall of pipelines when it is difficult for the robot to find moving trajectory or when the robot overruns forward because of operator s mistake, and it cannot turn. It is not easy to estimate of this locomotion before seeing actually, so that the role of this set is impressive enough while the robot is turning in a branch. III. Problem Statements In urban gas pipelines, their geometries are relatively simple since their dimensions and configurations are regulated by law, which is quite advantageous aspect in developing the robot. evertheless moving inside pipelines with a differential drive robot produces severe problems in practical applications. In navigation on plain surfaces with mobile robots, steering is simply accomplished by modulating the speed of wheels according to the desired moving direction. It is not even required to know how configurations of the workspace such as curvature of the floor are. Currently only internal status variables such as position, velocity etc. are used in the control. In the case of inpipe navigation, the situations are quite different from that of the plain surface. Just knowing the internal variables of the robot is not sufficient any more and additional information such as the geometry of the pipelines and the locations of the contact points are prerequisite. It is required to sense both internal variables and external environments simultaneously, which is more serious in the case of differential drive robots. In fact, independent drive wheels of differential drive robots may have the advantage to provide steering capability in fittings such as elbows and branches. However there are problems to be solved as much as. For example it requires elaborated control to drive wheels in straight pipelines as well as fittings. During movement in elbows desired speeds of the wheels are different from each other depending on the contact points with the wall of the pipelines. It is sure that it is undesirable features of a differential drive robot because a robot with a single active wheel (which is called single drive robot hereafter) never has the same kind of problems even if it can not provide steering capability in branches. In case of differential drive robot, accurate control of speeds of each wheel is prerequisite to prevent the slippage of the wheels based on the path that the wheel moves along. In reality it is impossible without knowing where the wheels have contact with the wall and it is not easy to sense the location of contact points, too. The cases are more serious in branches. The problem is there is no adequate way of figuring out the locations of contacts of the wheels with conventional way of sensing. Actually it may be the main reason preventing the application of the differential drive robot in pipelines. A simple and easy way of getting the information of contact point and modulating speeds are required. Up to now, these problems have not been discussed in depth though a few robots may be found taking the configurations of mechanism similar to MRISPECT IV [8]. In the following sections, rugged and simple algorithms estimating the contact point based on the analysis of the geometrical features of elbows and branches are proposed. Then a method for modulating speeds of the wheels are proposed.

6 6 The moving trajectory relying on the mechanism construction of the robot needs to be analyzed in detail to modulate the velocity of each driving module. In addition, the algorithm or the strategy for moving in fittings should be planned. These sections explain how the mechanism of MRISPECT IV can follow this process and discusses why the previous robots with various mechanism can hardly travel in various fittings. IV. Strategy for avigation in Elbows A. Description of an Elbow Geometry and Behavior of the Robot in an Elbow + EH? A * #, + EH? A ) H? B - > M Fig. 7. Formation of an elbow As illustrated in Fig. 7, an elbow is geometrically built by the rotating Circle A that corresponds to the cross-sectional area of the elbow. Let us set a z-axis along the axis of rotation of Circle A and the radius of Circle B which is generated by the rotation of Circle A along z-axis designated with r c. Referring to the regulation of urban gas supply equipment, r c is 1.5times larger than the diameter of Circle A such as r c =1.5D when D represents the diameter of Circle A. x and y axis are assigned along the arbitrary direction of z-axis. Then the equation for the elbow can be written by P elbow (φ, λ) = ([(r c +0.5D cos φ)cosλ], [(r c +0.5D cos φ)sinλ], [0.5D sin φ]), (2) where φ is the parameter representing the polar location of the pipelines wall on Circle A and λ denotes the rotation angle of Circle A. A 5 JH = EC D J2 EF A + H + K H L A 4 ) H? 2 + ) * 5 A? JE + ) * - > M 5 A? JE ) * Fig. 8. Behavior of the robot approaching an elbow Fig. 8 shows the moving phase of the robot in an elbow. Section C-A denotes the range where the center of the robot moves between the straight pipeline and the elbows, while section A-B is the range where the robot is completely placed in the elbow. Curve R represents the moving trajectory of the center of the robot, and

7 7 Arc P means the part of Circle B going through the center of the torus as shown Fig. 7. Curve R does not coincide with Arc P because the wheels of the robot with longitudinal length have three-dimensional contact with the curved inner surface of the elbow. If the difference between Arc P and Curve R is designated with d, it is changing while the robot moves from a straight pipe to an elbow and it varies depending on the posture of the robot. This means that the legs with wheels are placed variously around the central axis of the robot when it enters the elbow. To travel effectively along an elbow, it should be noted that the curvature of the elbow varies depending upon the inner areas of the pipelines having the contact with the wheels of the robot. The wheels of ne or two driving module among three modules are forced to slip when speeds of all the wheels are the same. It is because the moving trajectory of each wheel is not the same. This consequence causes overload to the driving system, and gives quite detrimental effects on the overall usage of the robot in workspace such as gas pipelines consisting of many elbows. To settle this problem, previously, the method making other wheels except one to be idled without driving force has been employed [18], [19]. However, this method can not provide the driving vehicles with the traction force enough to haul the other equipments in vertical elbows and bends because traction force is generated by only one wheel. Therefore, each driving module of the robot needs to be actively driven such as to modulate the speeds of the wheels according to the configuration of pipelines, which is not easy because it includes the sensing of environments and itself, and requires control in three-dimensional space. B. Approach for Speed Modulation in Elbows : Simplified Case To modulate the moving speed of each driving module or exactly speaking, to control the angular velocity of each wheel, it is the sensible and appropriate way that the trajectories of wheels having contact with the inner wall of elbow is expected geometrically and known. This is originated that the proper accommodation to action by sensing itself and its own circumstance is difficult because of three dimensional curved space of the elbow. + EH? A - H F H G H H G A F H A H H H G F G H H F + EH? A - G A,? F H #, Fig. 9. Analysis for the moving of disk in an elbow Fig. 9 is the figure to explain simply the method for computing the trajectories of wheels in an elbow. Let us simply assume that d is zero and consider that Circle E describes the cross-section of the elbow, where p, q and r are the points of wheels having contact with. This means that the shape of the robot is simplified as a thin disk and the trajectory of the center of the robot is coincided with Circle B in Fig. 7. Then the plane connected by p, q, and r would be the equilateral triangle inscribed in Circle E because each driving module is under radial restraint and the trajectories of wheels would be arcs with radiuses of r p, r q and r r from not e but z-axis. The ratios of velocities of three wheels are replaced with those of r p, r q and r r because these trajectories are the moving distances of the wheels in the elbow for the same period. The relations are represented as V p : V q : V r = r p : r q : r r (3)

8 8 = 1.5D 0.5D cos ψ c :1.5D 0.5D cos(ψ c 120 ):1.5D 0.5D cos(ψ c ), where ψ c is the angle between the planes. ne is the plane including p and the center of the robot and the other is xy-plane. Fig. 10 shows the comparison of the velocities of wheels depending on ψ c. bviously the velocity of each wheel varies depending oh the posture of the robot, which is designated with ψ c I 8 H 8 F 8 G $ & "!! $?, A C Fig. 10. Comparison of velocities in case of disk C. Approach for Speed Modulation in Elbows: Actual case In the previous subsection, the assumption that a robot were simplified as a thin disk makes the curve happen to be a circle. Therefore, the trajectories of wheels are determined easily, so that the velocities of wheels is computed simply. In the real world, however, it is not simple to compute the locomotion of a typical robot in elbows because the robot has the specific size and shape(i.e., the overall length of a robot, and the diameter and thickness of wheel). Fig. 11 illustrates the geometrical analysis as to this difficulty of computation. To accurately modulate the velocities of wheels of a robot in elbows, the trajectories of wheels are computed by using a geometrical method based on the actual behavior of the robot in the elbow. The computation requires to solve simultaneous equations with following two conditions. The first condition is that the curve crossing between the specific plane and elbow exists, where the specific plane means the one including the contact points of wheels and the elbow. The second is that the polygon formed by connecting the points of wheels on the curve is a equilateral triangle due to the kinematical constraint among driving modules whatever the posture of the robot is. The curve mentioned in the first condition for the velocity modulation is neither circle nor ellipse. The curve, by the name of Curve W real is like the cross section of an egg because a robot has a peculiar length. Also, because the has a finite diameter, Plane α including Curve W real is not parallel to the yz plane perpendicular to the central axis of the robot and the centers of wheels are not on the Plane α. Moreover, due to the thickness of a wheel Plane α is not perpendicular to the xy plane and P real, Q real and R real, which are points on the inner wall of the elbow having contact with wheels of the robot, are not on the plane perpendicular to the axes of wheels including the centers of wheels. In other words, Plane α of which normal vector h with arbitrary parameters of γ x, γ y and γ z is not determined easily. Also, Curve W real on Plane α is similar to the shape of the cross section of an egg, but is not symmetric with respect to any points or lines. Consequently, Curve W real cannot be found easily, to say nothing of the computation of the curve. This is because the wheel has the shape such as a cylinder or a tire with the diameter and thickness, and thus the wheel has three-dimensional contact with the wall of the elbow with compound trajectories. Although it is quit difficult to obtain the equations of the trajectories, they are able to be simplified into forms easily handled. The body and frame of the robot can be simplified to lines. Joints and wheels can be simplified to points as illustrated in Fig. 12. Let us assume that P, Q and R are the points of wheels having contact with the elbow, and are on Curve W which means the curve of the first condition for the velocity modulation introduced at the beginning of this section. Because of the simplification of the robot, the plane

9 9 C A C C D A D 2 = A = + K H L A 9 H A = 2 H A = A 3 H A = ) H? 4 A #, A 3 H A = A 2 H A = 4 H A = ) H? 4 H A = Fig. 11. Analysis for the robot in an elbow including all the P, Q and R is parallel to the plane perpendicular to the central axis of the robot whatever value of ψ is. ψ is the angle between two planes, where one plane includes the closest point T on Curve W from x-axis and the central axis of robot, and the other plane includes the point P and the central axis of robot(in this paper, the posture of a robot means the posture being related to ψ). Distance between yz plane and the plane including Curve W is L when the overall length of the robot is 2L. Thus, the positions of P, Q and R can be given by P =(L,y P,z P ), (4) Q =(L,y Q,z P ), R =(L,y R,z R ). Arc W P, Arc W Q and Arc W R show the trajectories travelled by each driving wheel. Each velocity is proportional to the length of each arc or R P, R Q and R R, and the velocities ratios of wheels at P, Q and R can be computed by using Eq. (6) V P : V Q : V R (5) = R P : R Q : R R = L 2 + y 2 P : L 2 + y 2 Q : L 2 + y 2 R, To modulate the velocities of wheels, therefore, P, Q and R should be known because the radii R P, R Q and R R are derived from these points. Because all the points are derived from the Curve W, the equation of Curve W should be solved in advance to modulate velocities. The equation of Curve W, on which P, Q and R exist, can be obtained by using Eq. (6). This presents the relations that the value of x coordinate is L in Eq. (2) of the elbow. Thus, the equation of Curve W is obtained as follows: (1.5D + 0.5D cos φ)cosλ =L. (6) W (φ, λ) = (L, [(1.5D + 0.5D cos φ)sinλ], [0.5D sin φ]) (7) = (L, Ltanλ, 0.5D sin φ).

10 10 A 4 2 A ) H? ) H? 9 3 ) H? K H L A 9 4 A = 6 H = A? J H EA I BM D A A I +?? ,, 6, , < EH? A - 5 A? JE L EA M + K H L A > 4 A = JE > A JM A A = H > = A > M Fig. 12. Modulation of the velocities of wheels Since sin φ = ± 0.5D from the relation of φ and λ in Eq.(6) we have 1 ( L 3) 0.5D 2 cos λ 0.5D W (λ) = (L, Ltanλ, ± ). (8) 1 ( L 0.5D cos λ 3)2 Because the points P, Q and R are on the Curve W, Eq. (5) is rewritten as and the triangle made of P, Q and R is an equilateral triangle as P = W (λ P ), (9) Q = W (λ Q ), R = W (λ R ). U PQ = U QR = U RP. (10) The solution to Eq. (10) can make λ Q and λ R replace with the function of λ P by using the relation that P, Q and R is on Curve W as given in Eq. (10). Thus, Eq. (6) for the navigation by differential drive method in elbows can be expressed as the equation of the λ P. This process is because the points of P, Q and R of Eq. (6) are not obtained directly. When a robot move along elbows, L and D are the known values, and ψ is the value obtained from the image of CCD camera. λ P which is the parameter of R P, R Q and R R should be substituted with ψ, L and D because the ψ is the only parameter sensed during the elbow navigation with the known value.l and D. The relation of λ P, ψ, L and D is obtained by solving the additional equation as follows. cos ψ = U CP 2 + U CT 2 U PT 2, (11) 2 U CP U CT

11 11 where the position C of the crossing point between the central axis of the robot and the plane including Curve W is given by C =(L, y P + y Q + y R 3, z P + z Q + z R ), (12) 3 and the position T on the Cureve W is given by T =(L, D 2 L 2, 0). (13) Unfortunately, it is not easy to solve these equations because (Eq. 10( containing the relation of Eq. (10) is a complicated non-linear equation. In this case, off-line computation is applicable. Fig. 13 (a) shows Curve W + EH? A - 3! + K H L A 9 $ ' 2! ' ) > K J 4, $ $ & ", A C + K H L A + = > $ I $ & "!! $, A C? Fig. 13. ff-line computation for differential drive and Curve C. These curves are similar figures each other, where Curve C is the trajectory of the point C represented in Eq.( 12) depending on ψ. Curve w of Fig. 13 (a) is exaggeratively represented through scaling in order to be understood the shape of it more distinctly. Fig. 13 (a) also shows that the center C of the robot is always above the center of Circle E regardless of ψ, where Circle E mentioned in previous subsction is the cross section of the elbow. As shown in Fig. 13 (b), l is the distance between the central axis of the robot and P depending on ψ. l is changeable depending on ψ because C is not fixed. Finally, values of V P, V Q and V R are computed as shown in Fig. 13 (c). Fig. 13 are plotted by the values applied to MRISPECT IV. As shown in Fig 14, V P, V Q and V R are compared to V p, V q and V r which are the velocities as shown in Fig. 10. As shown in Fig 14 (a) V P is compared with V p depending on the each posture ψ and ψ c, where it should be noted that the maximum and minimum magnitude of two velocities are scaled to be equal for the comparison about the tendency and shape between two. In Fig 14 (b), the velocity ratios of other wheels to one wheel are compared with each other. Considering V R V P and Vr V p, it is considerable that the value of ψ in the max V R V P is about 28 and it is not equal exactly to the value of ψ c in the maximum Vr V p. The longer the overall length of a robot is, the more large the the velocity difference between the real world robot and the simplified one of thin disk-type increases. This

12 12 I 8 F # 8 H 8 F # 8 G 8 F ! $ ' # & " %!!!! $! $ ' # & " %!!!! $ H?, A C = 8 A? EJ BM D A A I > 8 A? EJ H= JE BM D A A I H?, A C Fig. 14. Comparison of velocities according to the shape of a robot difference means that the velocity control of the robot is simply not replaced with that of the disk-type vehicle. As for both two types, however, the differential drive method is distinguished from the single-drive. The difference of wheel velocities in the elbow navigation always exist and the difference is too much to neglect. The maximum ratio of velocity is from 1.5 to 1.8 according to the overall length of robot as shown in Fig 14 (b). In conclusion, the navigation of the robot in elbows using the differential drive method has many advantageous aspect. It gives less slipping and overrunning, and could be realized with minimum number of sensors. It is rather difficult to understand the condition of the robot concerned with the elbow, but it is relatively simple to control robot for the navigation of elbows. ψ is decided by the information of the CCD camera, which is the only information required to control the driving module. V. Strategy for avigation in Branches A. Description of a Branch Geometry and Behavior of the Robot in a Branch As shown in Fig. 15 branches can be considered as consisting of two elbows, a straight normal pipe and two flat areas called V-shaped area between the elbows. V-shaped area has a flat surface one and the surface - > M ) - > M * 8 I D = F = H A = H = 2 EF A Fig. 15. Formation of a branch can be found out only in V-shaped area throughout the pipelines. Consequently, V-shaped area may be an obstacle to the robot because it has been designed to be suitable for the curved surface. When a robot enters a branch, initially the diameter of the pipeline does not change a lot until front wheel Set reaches the line B-B after passing through the line A-A illustrated in Fig. 16, but the robot cannot rotate in this region whatever speed differences are given. When front wheel set comes to the line C-C, the diameter of the pipeline changes greatly and the robot seems to turn toward turn drive space represented in Fig. 16. This is because front wheel set is forced to be placed turn drive space reacting the change of diameter and unfolding linkage locomotion of the robot. The robot, however, cannot turn by using speed differences because front wheel set has still contact with the inner surface of pipe and rear wheel set is confined absolutely the inner surface of pipe. In other words, the wheel sets are not fully released enough to turn. Practically, only slip occurs between the wheels and the inner sides of the pipelines and the robot cannot turn even if the speed difference of wheel is given. In this space, the

13 13 ) * +, * =? 9 D A A 5 A J, H EL A + D E? A 5 F =? A 5 JH = EC D J, H EL A 5 F =? A 5 A? JE ) ) 5 A? JE * * 2 H A E E = H 5 F =? A. H J9 D A A 5 A J ) * +, 6 K H, H EL A 5 F =? A 5 A? JE A? JE,, Fig. 16. Constraint space in a branch robot can prepare to turn or drive straightly, and thus this space is said to be preliminary space. Whenfront wheel set is close to the line D-D, either one or two wheels, which are placed at the turn drive space, does not have contact with the inner side of the pipeline. Such a phenomenon indicates that there exists no inner side of the pipeline having contact with the wheel, which may prevent the robot from moving to the turn drive space. In this region, the robot can turn along the desired direction with speed differences. The robot, however, still select either going straightly or turning, and thus this space is said to be drive choice space. To travel in branches, the robot basically follows the method similar to the method in elbows, but there are several characteristic features which make it difficult to estimate its motion. Those can be summarized as follows. 1)As the robot moves along branches, it meets a variety of cross sections depending on the posture of 5 JH = EC D J, H EL A ) ) H? 2 + K H L A 4 6 * + 6 K H, H EL A 2 + K H L A A? JE 2 ) 5 A? JE 2 * 5 A? JE 2 + Fig. 17. Moving trajectory in a branch and various cross sections the robot as shown in Fig ) Depending on turning direction, the robot is influenced by gravity, so that turning courses are also different accordingly, where Curve R T and Curve R T are the turning trajectories of the robot in a branch when gravity is +y and y direction. 3) The robot is designed to move while making six wheels have contact with the inner side of the pipeline radially. evertheless, when the robot enters the branch, some of the wheels may not have contact with the wall as shown in Fig ) The steering of the robot is three-dimensional, and wheels are caused to be distorted and slide. Thus, the trajectories at the wheels become nonlinear arcs which are difficult to be computed. 5) n moving in a branch, front wheel set and rear wheel set are initially unfolded, and then folded, which cause the wheels to slide. Curve R T as illustrated in Fig. 17 has different meaning from Arc R e of the elbow because the arc of the elbow is a deterministic path produced by the robot while its wheels keep contact with the inner wall of the elbow without sliding. The robot does not turning in the branch having contact with inner surface of the pipeline. Therefore, the strategy to travel in branches is different from that of elbows.

14 14 5 JA F K 6 ) K EE= H M D A A, H EL E C M D A A 5 JA F 5 EI J H JE 6? J=? J5 F =? A BM D A A 5 A? JE 6 6 Fig. 18. Characteristic feature of turning in a branch B. Velocity Modulation in Branches >? = > M? G 7 > 8 7 = = 7? M = M > = M > M =? M? M F M = Fig. 19. = > Strategy for turning in branches As mentioned, the equation of Curve R T can hardly be solved and it is almost impossible to compute the correct trajectories of wheels. However, to solve the Curve R T or is not required necessarily in order to travel in branches. To turn in branches, just the travel direction of the robot is need to be determined and the speed of each driving wheel is modulated accordingly. When wheels of the robot has different velocities, the robot turns smoothly along the inner surface of the branch. Fig. 19 illustrates the strategy for turning in a branch. When a robot enters the branch, it decides the direction of rotation using the CCD camera attached in the front of the robot, and then adjusts the rotational speeds of individual driving modules. In Fig. 19 (a), a, b and c indicate the points where each wheel contacts the inner side of the pipeline (this is based on the assumption that each wheel keep point contact with the wall). It also shows the rotational direction of the robot when only one wheel rotates while two wheels remain without rotating. When the speed at the point a is V a and speed at b and c are 0 respectively, the robot turns with the angular speed ω a along the direction of the vector connecting the points b and c. With the method of direction choice like this, we just need to start modulating the speeds of the wheels and to turn the robot to the desired direction for steering in the branch actually. The rotation speed should be controlled accordingly as shown in Fig. 19 (b). To turn the robot, in the first, V c at the point c assuming that it is the place closest to the rotational direction z, isset to be 0. Using the CCD camera we already have known the angles θ, θ + 120,andθ between z-axis and wheels, respectively. ω P of the angular speed can be freely selected to let the robot rotate, but V a,and V b have to be decided to select the travel speed of the robot at contact points between each wheel and the inner wall of the pipeline. The following geometrical and vector relation can be used to derive V a,andv b. V a = ω a 3 2 U a (14) V b = ω b 3 2 U b (15)

15 15 ω p = ω a + ω b (16) ω b ω a = ω b ω a cos 120 (17) ω p ω b = ω b ω p cos θ 0 (18) î ω p = ω p (19) Because ω p can be freely selected, only the relation between V a,andv b is necessary such as V b : V a = sin(120 θ 0 ):sin(θ 0 ). (20) As described in this section, only if the robot is guided to the direction to turn, the robot will turn in the branch. It is because the link mechanism of the foldable pantograph type lets each wheel have contact with the inner wall and because the surface of the branch functions such as the guide or the path to keep turning direction desired. In the case of straight drive, when all the speeds of wheels are the same, the robot goes straight ahead though the trajectory is a little different depending on the direction of gravity. The strategy and algorithm for turning in branches is things about the simple choice of steering direction. However, other requirements such as the mechanical construction and the design consideration presented in the section VIII should be kept even if the strategy is simple and effective. VI. Experiments To confirm the effectiveness of the proposed navigation algorithms, several tests have been performed. In this section, the system setup is briefly described and experimental procedures with results are given. A. utline of Experimental Setup + +, EJ H + JH + F K JA H A > = JE + JH A H + J H EL A H + = J= JE E B H = JE 6 A JD A H? = > A 8 E@ A I EC =, H EL E C F M A H + K JA H I EC = J H I -? A H I + +,? = A H = Fig Experimental system setup As shown in Fig. 20, MRISPECT IV is controlled by a personal computer and power is supplied via a tether cable externally. An operator controls the behavior of the robot with a keyboard observing the images from CCD camera gathered through the tether cable. In most cases, the robot moves autonomously and the operator only chooses the direction for moving in elbows and branches with CCD images. The user interface is shown in Fig. 21. All the programs were coded with Borland C++. The left window of 21 is for steering and it helps operator choose the direction for moving in elbows and steering in branches according to the posture of the robot. The right window displays the data and the state information of the robot such as angular velocities of the wheels, which is computed automatically by the speed modulation algorithm. In order to evaluate the feasibility of the developed robot and the proposed algorithm, testbeds were built as shown in Fig. 22. Fig. 22 (a) is for the preliminary test. Fig. 22 (b) is for the advanced navigation experiment

16 16! " % $ & # ' 5 A A? JA > M H > H =? D - * 9 EI F = E C L E EH A? JE E A > M I > H =? D A I 2 I EJE I B@ H EL E K A 9 D A A I! 7 F F A EH A? JE B+ +,? = A H = " 9 M EI F = E C JD A? JH K H H A JL A? JEA I BM D A A I # 9 M EI F = E C JD A I EH L A? JEA I BM D A A I $ BA > M L E= JE G B> H =? D A I = L EC = JE % L E EH A? JE E A > M H I JA A H E EH A? JE E > H =? D A I & L E C L A? EJ E A > M H I JA A H E EH A? JE E > H =? D A I ', E= C K A M M Fig. 21. User interface in elbows and branches. The locomotion and movement of the robot are easily observed because pipelines are made of transparent plastics with several off-the-shell fittings Various experiments such as steering in branches influenced by gravity can be performed because the components of this testbeds are easily reconfigured. B. Experiments in Elbows Fig. 23 (a) and (b) represent the experimental scenes when the robot moves along the simple elbow and special elbow composed of three welded elbows, respectively. During the experiment power consumption was measured. In the case that speed were not modulated, much power was required since the robot was overloaded due to slippage of wheels. The result says that the navigation by the speed modulation scarcely makes slips of wheels by comparison to no speed modulation. C. Experiments in Branches avigation in branches is largely classified into two cases such as straight drive and turn drive. Fig. 24 shows the scenes of steering tests. The one is simple to operate in any case in comparison with the other. The same speed of each driving wheel always make the robot go straight ahead in branches. The other, however, should accompany careful consideration to the direction of gravity. Turn drive is classified into ten cases according to the turning direction to gravity in branches as shown in Fig. 25. Cases 1 and 2 are the cases that the turn drive is possible without considering of the direction of gravity, while other cases are sorted according to the turning direction with respect to the direction of gravity. The isolation of the robot in branches would occur if turning were impossible in even one case. Figs. 26, 27 and 28 show the experiments of these cases. The tests for cases 3, 7 and 8 often happened to fail because the operator improperly chose the timing for turning. Mistakes by the operator in these cases is because the gravity has an effect on turning in branches in comparison with the other cases. All the experiments, however, finally resulted in success though cases 3, 7 and 8 often happened to fail. The reason of the success is because the turn drive in branches by the speed modulation is accomplished not by the full driving force which can negotiate gravity, but by the triggering force which can generate the driving momentum along the desired direction. VII. Discussion Previously we have presented the algorithms, strategies and experiments for the navigation in elbows and branches by speed modulation. Although they are proved to be satisfactory through various experiments,

17 17 0 H E J= F EF A - > M 8 A H JE? = F EF A = 6 A I J> B HF HA E E = H A F A HE A J * H =? D A? - > M > 6 A I J> B L =? A F A HE A J Fig. 22. Testbeds there are several additional consideration in design for the proposed algorithms works. In this section, four aspects that is most typical among them, are going to be discussed. A. Supplement Devices MRISPECT IV has the supplement devices to assist the turning drive in branches more smoothly and stably though the robot turns in branches with the method of speed modulation. They, which are auxiliary wheels, idle wheels and CD wheel set, not essential but useful for the turning drive in branches under the influence of gravity sometimes. When the robot turns in branches, it may happen that the main wheels don t have contact with the inner wall instantaneously as shown in Fig. 29 according to the posture as shown in Fig. 19. In this case, auxiliary wheels are designed so as to have contact with the wall instead of main wheels, and give the instantaneous drive force to the robot. Idle wheels prevent the casing of the robot from contacting the the inner wall of branches. However, auxiliary wheels and idle wheels have no effect on the drive in elbows and straight pipes because these wheel have never contact with the wall of branches. The usefulness of CCD wheel set has already been mentioned in the section II. In other words, the supplement devices prevent the robot from being isolated in branches and make the differential drive robot be the complete mechanism.

18 18 5 JA F 5 JA F = 8 EA M BJD A H > JE = JH= I F = HA JA > M 5 JA F 5 JA F 5 JA F! >, HEL A F A HB H =? A JA I JE = A > M Fig. 23. avigation in elbows 5 JA F 5 JA F = 5 JH = EC D J@ H EL A > 6 K H E H EL A Fig. 24. avigation in branches B. Design for Differential Drive Type MRISPECT IV cannot be applied to all kind of pipelines because the components such as elbows and branches, reducers, flanges, etc. have their own shapes and sizes. For the design of a robot, the number of driving modules, link mechanism, shapes, size and supplement devices have to be considered synthetically. B.1 umber of Driving Modules Robots of differential drive type should have more than three driving modules in order to negotiate gravity and navigate satisfactorily in three-dimensional space. A robot with four driving modules has its own characteristic features. Usually wheels of only three driving modules among four have contact with the inner wall as illustrated Fig. 30 (b) when a robot moves along an elbow. Also, even if the robot moves in the elbow with the same posture, the three contact points cannot be expected easily, where the same posture is illustrated as ψ of 70. The destination of contact depends upon gravity direction, slip of wheels, etc. when the robot

19 19 C + = I A + = I A + = I A! + = I A " + = I A # + = I A $ + = I A % + = I A & + = I A ' + = I A Fig. 25. Steering Cases according to the branch posture to gravity direction C 5 JA F 5 JA F 5 JA F! = + = I A C 5 JA F 5 JA F 5 JA F! > + = I A Fig. 26. Turn tests of the case 1 and 2 in branches passes from straight pipe to elbow The case which all wheel can have contact with the inner wall is when ψ is 45, 135, 225 and 315 as shown in Fig. 30 (c). It means that the number of driving modules which can give driving force to the robot is only three in elbows. Also, it means that the other one wheel slips or idles when the position of wheels are far from, P, Q and R. Therefore, the robot with four driving modules can not be more effective than that with three modules. In the case of branches, the turn drive largely depends on generating the driving course. Driving force would increase if many modules had contacts with the wall but the force is not the main requirement to turn in branches. Fig. 31 is the figure to explain this feature. In the posture of robot such as Fig. 31 (a), only one driving module is driven through the algorithm of turning in branches and two driving modules are in Fig. 31 (b). Turning drive of the other with the more driving force is more difficult than the one in the test. The

20 20 5 JA F 5 JA F 5 JA F! = + = I A! C C 5 JA F 5 JA F 5 JA F! > + = I A " C 5 JA F 5 JA F 5 JA F!? + = I A # C 5 JA F 5 JA F 5 JA + = I A $ Fig. 27. Turn tests of the cases 3 6 in branches reason is that the one generates the more suitable turning course than the other. Thus, many driving modules is not so effective as expected. However, the robot of four driving modules still opens to further discussion because the experiment about this case has not been performed thoroughly. It is still difficult to predict the turning process in branches before doing experiments. B.2 Various Link Mechanism Driving module of MRISEPCT IV has two main wheels. one is front main wheel and the other is em rear main wheel. Linkage locomotion of them are independent each other. However, front main wheels of three driving modules are constraint and so are rear main wheels. It is because the linkages joint three driving modules. If three driving modules are not jointed by linkages, then they are unfolded and folded respectively. This mechanism may have advantage to complement the defective locomotion which all four wheels have not contact with the walls of elbows as mentioned in Fig. 30. However, this mechanism causes that the central axis of the robot does not coincide with that of the pipeline because of gravity. Also, it can cause a new problem in the branch navigation as shown in Fig. 32 (a). ne of driving modules is freely unfolded toward turn drive space without regard to the link locomotion of the other driving modules, and the separated driving module cannot but conflict with the straight drive in branches. Fig. 32 (b) shows the design consideration about the wheel set. MRISPECT IV has two wheel set which are front wheel set and rear wheel set. If a robot has the only one wheel set with actuation capability, it should be constructed by another same type robot and connected by universal joint each other[8] because

21 21 5 JA F 5 JA F 5 JA F! = + = I A % C C 5 JA F 5 JA F 5 JA F! > + = I A & C 5 JA F 5 JA F 5 JA F!? + = I A ' 5 JA F 5 JA F 5 JA + = I A C Fig. 28. Turn tests of the case 7 10 in branches the only one set loses the balance easily in even straight pipelines and the recover of the balance is almost impossible by itself. Therefore, the proper combination control of two robot and universal joint should be required. This mechanism has the advantage in the elbow navigation. The velocity control for the differential drive in elbows is somewhat easy because it can be applied to the simplified case presented in section IV, and can use Eq. (4). Also, the robot negotiate sharped bent elbow as illustrated in Fig. 32 (c) because it is composed of flexible body with a universal joint. However, the flexibility causes the front robot unit to fall into turn drive space in branches, and the straight drive happens to be impossible. Though the universal joint can have the active steering capability in order to complement this defect, the mechanism is an articulated steering type beside differential drive type. B.3 Shape of a Branch and Size of a Robot Fig. 33 (a) depicts a branch consisting of two straight pipes crossing at right angles to each other. The fact is proved through experiments that it is easy to drive straightly and possible to turn in this branch regardless of gravity. It is because the branch behaves like a guide to the robot as mentioned in the previous sections. It is remarkable that the robot turns such a shapely bent branch as illustrated in Fig. 33 (a) more easily than the smoothly curved branch. The reason is that the one has the suitable inner wall to let the robot contact in comparison with the other. In other words, the one has the structure which is less affected by gravity than the other. It, however, seems to be difficult that the robot is designed to avoid the interference between the body of itself and the shape of this branch. After all, the important thing for turning in branches is the design

22 22 + J=? J B= K EE= H M D A A + J=? J BM D A A Fig. 29. Function of auxiliary wheels 3 2 =!, L EA M " # 3 % 3 2 % 2 = > + J=? J BB K HM D A A I? + J=? J BJD HA A M D A A I 2 Fig. 30. Feature of four driving modules of the robot accommodating itself to the shape of branches. The size of a robot determines whether turning in branches is possible by speed modulation or by articulated steering joint. For example, when the length of the robot is a little long as shown in Fig. 33 (b), the robot cannot turn in branches by speed modulation though the robot has the proper size for moving in elbows [18]. When front wheel set of the robot is placed in a branch and rear wheel set has contact with the inner side of a straight pipe, rear wheel set is confined absolutely to the straight pipe space. Rear wheel set stops from steering though the robot tries to turn. Thus, to turn in the branch, rear wheel set should pass over the line u u from which the area of the branch is. The robot should start steering locomotion before front wheel set reach the line v v. Iffront wheel set passes over the line v v and the robot tries turning, then separation and isolation will occur. However, the robot can turn in branch until front wheel set reach the line w w if the body of the robot except wheels does not have contact with the wall. This means that supplement device such as CCD wheel set extends turning boundary to the line w w. Therefore, the length of the robot should be shorter than 1.75D. n the other hand, the robot could turn easily but could not drive straightly because it would be isolated in turn drive space if the length of the robot is shorter than the diameter D of pipe. C. Extensiblity In general, an existing inpipe inspection system has mainly articulated vehicles such as trains [4], [5], [8], [11], [15], [18], [19]. The system is composed of two driving vehicles for traction which are normally placed

23 23 + K H L A 4 6 = 6 K H E C B H EL E K A ) H? 2 + K H L A 4 6 > 6 K H E C? K H I A I > = > > 6 K H E C H EL E K A I Fig. 31. Turning trajectory according to the posture of a robot L E EH A? JE L E EH A? JE 5 A F = H = JE = 5 A F = H= JE B= M D A A > I E C B> = =? A 7 EL A H I = E J L E EH A? JE L E EH A? JE 5 A F = H = JE? A C JE= JE B= I D = HF > A JA > 5 A F = H= JE B= H > J Fig. 32. Various link mechanisms in the front and the rear of system, and the other vehicles such as control vehicle, DT vehicle as ultrasonic device, flexible joints connecting each vehicle, etc. MRISPECT IV may be used as a driving vehicle from the viewpoint of the overall system. Therefore, extensibility is needed to play an adequate role as one part of the whole system. If the driving vehicle itself is able to adjust to all fitting of pipelines, the articulated system including the driving vehicle could do that. This extensibility depends on the followings. 1)The driving vehicles of the system should have more traction force enough to negotiate gravity in vertical pipelines and branches as the number of other vehicles increases. 2) In branches, it should be considered how the joints affect the turning drive because the turning course by the driving vehicle connecting the other vehicles with the joints is different from that by the single vehicle. 3)Total length of the driving vehicle including the joints should be satisfied with the size constraint for turning in branches. We considered these problems in the beginning, based on the previous studies which is obtained during the development of MRISPECT II and III [18], [19]. The concept and the configuration of the inpipe robot using the speed modulation of driving modules is simple because it does not need additional steering vehicles or modules or steering joints. Thus it can be very attractive though practical design is a little complicated.

24 24 1 JA H BA H A? A L E EH A? JE K * H =? D L M 5 JH = EC D J F EF A K H % #, = 1 JA HBA HA? A > A JM A A = H > = > H=? D > 5 E A B= H > J, L M " # Fig. 33. Shape of a branch and Size of a robot VIII. Conclusion Few of previous robots for inpipe navigation seem to have the differential driven method that is relatively popular in plain 2D circumstances. The reason is that it is very difficult to estimate or interpret how the robot can turn in a special fittings such as branches or elbows. MRISPECT IV navigates almost perfectly all fitting of gas pipelines without regard to the effect of gravity, posture and direction. It is being renewed and applied for test in field conditions. Though the excellency of the robot has been depicted from the viewpoint of simple construction and effective navigation, what is stressed is possibility and applicability rather than such excellency. To verify these abilities, new algorithm for navigation in the special and constrained threedimensional space such as elbows, branches are proposed through geometrical analysis, 3D modeling and various experiments. In addition, technical points were discussed from various angles in order to be considered in development of inpipe robots. We hope that these studies will be referred in other developments of inpipe robots. References [1] J. kamoto Jr, J. C. Adamowski, M. S. G. Tsuzuki, F. Buiochi, C. S. Camerini, Autonomous System For il Pipelines Inspection, Mechatronics, Vol. 9, pp , [2] T. kada and T. Kanade, A Three Wheeled Self Adjusting Vehicle in a Pipe, FERRET 1, Int. J. of Robotics Research, Vol. 6,o. 4, pp , [3] T. kada and T. Sanemori, MGRER: A Vehicle Study and Realization for In Pipe Inspection Tasks, IEEE J. of Robotics and Automation, Vol. 3, o. 6, pp , [4] S. Hirose, H. hno, T. Mitsui, and K. Suyama, Design of In pipe Inspection Vehicles for φ25,φ50,φ150 pipes, Proc. of IEEE Int. Conf. on Robotics and Automation, pp , [5] K.-U.Scholl, V. Kepplin, K. Berns, and R. Dillmann, An Articulate Service Robot for Autonomous Sewer Inspection Tasks, Proc. of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IRS 99), Vol. 2, pp , [6] K. Suzumori, K. Hori, and T. Miyagawa, A direct-drive pneumatic stepping motor for robots: designs for pipe-inspection microrobots and for human-care robots, IEEE. Int. Conf. on Robotics and Automation(ICRA), Vol.4, pp , [7] T. Tsubouchi, S. Takaki, Y. Kawaguchi, and S. Yuta, A straight pipe observation from the inside by laser spot array and a TV camera IEEE/RSJ Int. Conf. on Intelligent Robots and Systems(IRS), Vol.1, pp , [8] M. Mhramatsu,. amiki, U. Koyama, and Y. Suga, Autonomous mobile robot in pipe for piping operations, IEEE/RSJ Int. Conf. on Intelligent Robots and Systems(IRS) Vol. 3, pp , [9] H. T. Roman, B. A. Pellegrino, W. R. Sigrist, Pipe crawling inspection robots: an overview IEEE Trans of Energy Conversion, pp , [10] W. eubauer, A Spider Like Robot that Climbs Vertically in Ducts or Pipes, IEEE/RSJ Int. Conf. on Intelligent Robots and Systems(IRS), pp , [11] T. Fukuda, H. Hosokai, M. Uemura, Rubber gas actuator driven by hydrogen storage alloy for in-pipe inspection mobile robot with flexible structure, IEEE Int. Conf. on Robotics and Automation(ICRA), Vol. 3, pp , [12] C. Anthierens, A. Ciftci, and M. Betemps, Design of an electro pneumatic micro robot for in-pipe inspection, IEEE Int. Symposium on Industrial Electronics(ISIE), Vol. 2, pp , [13] I. Hayashi,. Iwatsuki, S. Iwashina, The running characteristics of a screw-principle microrobot in a small bent pipe Proc. of Int. Symposium on Micro Machine and Human Science(MHS), pp , [14] S. Iwashina, I. Hayashi,. Iwatsuki, and K. akamura, Development of in-pipe operation micro robots Proc. of Int. Symposium on Micro Machine and Human Science(MHS), pp , 1994