ship building with rower

DECEMBER 2000 IEEE Robotics & Automation Magazine 351070-9932/00/$10.00©2000IEEE

Today, industry is being strongly encouraged toimprove quality, productivity, and labor condi-tions. Modern robotic systems are helping to en-hance these features. However, current roboticsystems cannot accomplish some industrial ap-

plications, especially those involving mobility in complexenvironments.

In the area of naval construction, a main stage in the shiperection process is the construction and assembly of huge shipblocks. This work is performed in highly automated work-shops. Another important stage involves fitting the ship blockstogether in the dry-dock after they leave the workshop.Blocks are arc-welded together, and the resulting scenario is socomplex that operators using manual equipment must be

called in. Some welders turn out good quality work, but thejob’s overall quality is quite irregular.

Productivity in the second stage of the process is very low,because operators handling manual equipment need to stopfrequently. The weld length in this stage makes up 4 to 6% ofthe total weld length in a ship; however, it requires about 30to 35% of the total weld manpower. Lastly, the working con-ditions are very bad for the operator, who must weld in closedcells with very little ventilation and very thick fumes. Thesereasons led a team of shipyard companies and researchers todesign and build an automatic welding system for ship erec-tion processes with the main aim of increasing productivity byincreasing total arc time, improving weld quality, and creatingbetter working conditions for operators.

A Four-Legged MobilePlatform IncreasesProductivity and ImprovesQuality and WorkingConditions for IndustrialNaval Applications

by P. GONZALEZ DE SANTOS,M.A. ARMADA,and M.A. JIMENEZ

1994 PhotoDisc Inc.

Walking machine technology seems to be ready to accom-plish industrial applications, and with industrial considerationsin mind, we investigated different machine configurationswhile accounting for the end users’ strict requirements of totalweight, dimensions, payload, and assembly/disassembly capa-bilities. Finally, a four-legged machine able to walk by grasp-ing cell stiffeners was envisaged as the best choice for ourapplication. Thus, this article presents a legged machine (calledROWER) for application in naval construction processes, fo-cusing on the project investigation stage through the mobileplatform’s configuration and design rather than presenting thewhole project, in which three other companies were in-

volved. The legged machine moves through the cell by grasp-ing the top and bottom reinforcements rather than walking onthe bottom plane. This improves the machine’s stiffness andassures its stability. The article describes the walking machine’smain structure, leg configuration, and grasping mechanismand how they cope with the industrial working scenario.Lastly, the features and functionality of the ROWER walkingmachine are reported.

ROWER System OverviewThe overall system consists of a commercial welding systemhandled by a commercial manipulator. These subsystems are

carried on a mobile platform that provides mobility in theworking area. A stereo vision system finds the starting andending points of the welding seam. All four subsystems are re-mote-controlled by a computer supervised by an operator lo-cated off the work site. The supervisor computer also containsa database of the geometric description of each working cell.

The main task of the mobile platform is to provide full mo-tion to a welding manipulator in a very complex scenario: thedouble hull of a ship. This scenario consists of cells fitted withstiffeners on the top, bottom, and side walls to reinforce theship’s structure.

The complexity of the working scenario and the system mo-bility requirements made a walking machinethe best choice for this application. Leggedlocomotion has been extensively investi-gated in the last two decades, during whichtime many walking machines have beenbuilt as testbeds for research purposes. Mostof them are laboratory prototypes, but thereare machines that have been tested in out-door environments and under extreme con-

ditions. Some of these prototypes have been mainly focused onspace [1] and submarine tasks [2], where machine reliability is ofparamount importance for the success of the mission. Such ap-plications do not permit either an easy removal of some parts orrecovery of the machine after a subsystem breakdown. The on-going applications of walking/climbing machines in industrialprocesses [3, 4] as well as special applications on natural terrain[5, 6] demand reliability, speed, and accuracy, as well as main-tainability.

Process OutlineAs mentioned above, the ship erection process consists ofthree main activities:

◆ Block construction in the workshop.◆ Block transportation to the dry-dock or slip-way.◆ Block assembly in the dry-dock or slip-way.The first activity consists in the fabrication of big blocks of

the ship, as shown in Fig. 1. This process is easily automated,and its productivity is relatively high. The third activity con-sists of connecting two consecutive blocks by welding to-gether all the longitudinal reinforcements and all the verticalbulkheads. For environmental safety, most ships, especiallytankers and bulk carriers, are built with a double bottom anddouble hull so the cargo will not spill out if the hull isbreached. This double structure forms cells all over the ship’shull (see Fig. 1). A typical tanker bottom cell (horizontal) canbe up to 10 m long, 4 m wide, and 3 m high. The weldingseams in these cells are mainly located on the walls, bottom,and stiffeners on the section of the junction plane as well as inthe ceiling and bottom of a perpendicular plane. Therefore,the welding process can be defined as the welding together ofelementary cells, and this can be considered as the workingscenario for the mobile platform.

DECEMBER 2000IEEE Robotics & Automation Magazine36

Manhole

Hull

Stiffener

Tanker

Horizontal Cell

Figure 1. A ship block in the dry-dock.

The requirements of mobility in a horizontalplane full of stiffeners make the scenariovery hard for a wheeled or trackedvehicle to negotiate.

Summarizing, the scenario consists of a cell whose measure-ments may vary from about 7 to 10 m long, 2 to 4 m wide, and1.5 to 3 m high. Longitudinal stiffeners form part of the ceilingand bottom of the cell to reinforce the hull structure. Stiffenerheight ranges from about 0.2 m to 0.7 m. The mobile platformmust bear the welding system along the longitudinal andtransversal axes of any cell in the defined range and help the ma-nipulator reach the ceiling and bottom as well. To allow work-ers inside, the cells have a manhole measuring about 0.8 × 0.6 m(See Fig. 1). This feature is very important in the design of theoverall system, because the volume of each subsystem must beconstrained to the manhole’s dimensions. Another constraint isthat the weight of each subsystem (manipulator, welding sys-tem, mobile platform, vision system, etc.) must weigh less than50 kg due to labor regulations on systems carried by operators.This is the reason why the mobile platform must be dividedinto different parts and assembled inside the cell. For our pro-ject’s purposes, the reference cell is 10 × 4 × 3 m, the top stiffen-ers are 0.635 m tall, and the bottom stiffeners are 0.575 m tall. Inthe top and the bottom of the cell, the stiffeners stand 0.88 mapart. The requirements of mobility in a horizontal plane full ofstiffeners make the scenario very hard for a wheeled or trackedvehicle to negotiate. Table 1 summarizes the main require-ments for the walking machine.

Legged locomotion seems to be more adequate for mobil-ity in this kind of environment than traditional vehicles. Forthis particular application, different leg configurations wereenvisaged as likely candidates for walking on the floor of thecell. The pantographic configuration was rejected because theankle sweeps through a volume when propelling the body for-ward/backward or up/down [7, 8]. This can cause the ankleto collide with a stiffener, and that must be avoided. This fea-ture restricts the foot location area on the bottom of the celltoo much. An orthogonal leg was also considered and finallyselected. Its main advantage is that its vertical link can be setvery close to the stiffeners without sweeping through any vol-umes when moving the body. Thus, the ankle will not crashinto the stiffeners. In this process, the geometry of the cell, de-fined in the supervisor database, is known a priori to an accu-racy of about ±20 mm. For welding, the manipulator must beplaced most accurately by the mobile platform. Therefore, themobile platform must be accurate enough to accomplish thistask. This makes accurate positioning a new feature never be-fore required in walking machines up to this accurate stage.

Structure of the ROWER Walking MachineMain StructureThe structure of the body depends on the payload layout andthe functionality of the manipulator. In the present case, themanipulator can work in both the up and the upside-down po-sitions. However, it cannot work properly if it is wall-mounted.This feature led us to devise a manipulator mounted on a turn-table able to rotate about the longitudinal axis of a U-shapedframe structure body (see Fig. 2). Nevertheless, the problemsconcerning the manipulator and related welding equipment lie

outside the scope of this article. Thus, a parallelepiped bodystructure will be assumed in the rest of this article. The bodystructure is made of welded aluminum plates. Every plate has anumber of windows; by stripping off needless material, this de-sign makes for a lighter-weight structure.

DECEMBER 2000 IEEE Robotics & Automation Magazine 37

Figure 2. Body of the ROWER walking machine.

Table 1. Main Walking Machine Requirements

Requirement Value Comments

Dimension ofworking cells

Length: 7 to 10 mWidth: 2 to 4 mHeight: 1.5 to 3 m

Type of stiffenersBulb, T-stiffenerand L-stiffener

Stiffener height 0.2 to 0.7 m

Number ofconsecutive stiffenersto be welded withoutmoving foot positions

2

In a given machineclasping positionthere are two topand two bottomstiffeners to bewelded (see weldingseam in Fig. 4)

Maximum weight ofevery part

50 kg

Labor regulationstates that anoperator can carry upto 25 kg. Twooperators areintended to be incharge of carryingthe parts.

Machineassembly/disassemblytime

15 minutes

The assembly/disassembly taskconsists ofdismounting allsubsystems, passingthem to the nextworking cell andreassembling the fullsystem. This fulloperation is requiredin two hours.

Payload 130 kg

Includes weldingmanipulator, weldingtools, stereo visionsystem, manipulatorturntable, etc.

Although a six-legged machine provides very good stabil-ity, especially for industrial applications, a four-legged mobileplatform was chosen in order to minimize the machine’s totalweight, which is a top requirement in this application. Also,using four legs drastically reduces the final price, which is nor-mally an important industrial requirement. This initial selec-tion was considered the ideal solution for the final machine’sstructure, since there are no machine stability problems, ascommented below. The leg and the machine structures areshown in Fig. 3 and Fig. 4, respectively, and discussed in theparagraphs below. Figure 4 also shows the working position of

the machine with respect to the seam pathand surrounding stiffeners [9, 10].

Leg Structure and Grasping DeviceAs mentioned before, the orthogonal legwas considered best for this application.Thus, the leg consists of a vertical link actu-

ated by a prismatic joint and two more degrees of freedom toproduce horizontal motion. The different combinations ofjoint types considered—for the first two joints—included therotation-translation joint (as in the Ambler walking machine[1]) and the rotation-rotation joint (SCARA-like robot). Thelatter was finally selected because it provides greater speedthan the other, since electric rotary joints are normally fasterthan electric prismatic joints.

An additional problem with the chosen leg structure wasthe required size of the vertical link, which had to be 2 m highin order to place the manipulator close enough to the ceilingof the cell. This long link and the heavy payload the machinemust carry (about 130 kg) make the vertical link very prone toflexion and vibration, especially when the body is located at itshighest position. This feature also jeopardizes the machine’sstatic stability. To overcome this problem, an additional struc-ture was proposed to walk by grasping the stiffeners. Thewalker structure is basically the same, but a very simple grasp-ing mechanism was added to every vertical link. This graspingmechanism consists of two grasping feet and a prismatic actua-tor. The grasping foot consists of a disk and a peg that can ro-tate passively around a vertical axis as shown in Fig. 3. A radialsymmetrical device is necessary to avoid having to re-orientthe device before it grasps the stiffener. These feet are placed atthe top and bottom of the vertical link. The prismatic actuatoris located at the top of the vertical link, and its function is topress the top foot against the top stiffener as shown in Fig. 4.Also, this device can be retracted to let a leg pass between thetop and bottom stiffeners. This simple foot design features theadvantage of being able to clasp stiffeners of different shapes,such as bulbs, T-stiffeners, and L-stiffeners, which are themost common stiffener shapes in ship construction.

Figure 3 identifies the different elements in a leg. The legwas designed to use commercial subsystems as much as possi-ble. The SCARA structure of the leg basically consists of twohorizontal links made of commercial aluminum profiles,which are actuated by dc motor and Harmonic-Drive reduc-ers (100 W) to eliminate the backlash. The vertical actuator isa dc motor (250 W) that rotates the lead-screw of a commer-cial linear table. The extensible upper link consists of a com-mercial electrical cylinder. Some parts of the transmissionmechanisms and the leg attachments to the body are made ofsteel or special 7075 aluminum to lighten the weight as muchas possible.

Grasping is accomplished by all four legs grasping the stiff-eners on opposite sides, as in Fig, 4. With this final machineconfiguration, there are no stability problems because thegrasp is firm; however, a stiffener detection problem arises.


Disk

Peg Rotation Axis

Rigid Joint

Expansible orAdaptable Link

Passive Joint: ThePeg-and-Disk System

Can Freely RotateAround the

Expansible LinkGrasping Device

Extensible Link

Linear Table DC Motor 1

BodyDC Motor 2

C1

C2

Horizontal Link 1

Horizontal Link 2

DC Motor of Linear Table

Adaptable Link

ExtensibleActuator

DC Motor

Grasping Device

Figure 3. ROWER leg.

Shipbuilding procedures do not guaranteethe accuracy required for our purpose; hence,the use of sensors to detect stiffenerswith precision is essential.

That means it is necessary either to know very precisely theposition of the stiffener where a foot has to be located, or toknow the stiffener’s approximate position and to detect it ac-curately using some kind of sensor. As mentioned before, thecell’s geometry is known a priori, but cell construction accu-racy is not good enough for walking in a blind mode, and stiff-ener detection sensors proved absolutely necessary.

Assembling/Disassembling the MachineThe maximum operator-carried weight and manhole size re-quirements are satisfied by all project subsystems except themobile platform. Thus, the mobile platform must be takenapart so it can fit through the manhole into the cell. The easi-est solution seems to be to break the mobile platform downinto smaller modules. That means building the machine inmodules or series of standardized components that work to-gether in the whole system. The first approach is to divide themachine into body and legs, making it necessary to handle fivemain pieces. The body has been defined to satisfy weight anddimension requirements (see Fig. 2). The leg design satisfiesthe dimension requirements easily, but not the weight re-quirements, so one more leg breakdown is required. The finalsolution is to divide the leg into two parts: vertical links andhorizontal links; hence, the walking platform is divided intonine different parts.

The mobile platform is intended to be mounted/dis-mounted in different cells; therefore, an easy mechanism forassembling/disassembling is also required. Two aspects haveto be considered. The first is how to break the structure intodifferent pieces. The second is how to break up thepower/signal system distributed along the leg structure. Theseconsiderations have been taken into account in the design ofthe ROWER walking machine, which is configured to bequickly assembled/disassembled for operational reasons. Thesolution is simply to provide bolts and guides for mechanicalassembly and connectors for electrical connections at points C1

(four bolts and two connectors) and C2 (five bolts and oneconnector), as indicated in Fig. 3. These easy-to-handle ele-ments let a pair of operators affix a leg to the body in less than 3min. A similar time is required to dismount a leg.

Leg and Body DimensionsTo select the dimensions of every main piece, it was necessaryto consider not only the minimum and maximum dimensionsof the cell but also the mobility requirements of the mobileplatform’s body when the feet are clasped to the stiffeners. Themobile platform’s movement through the cell is basically veryslow. Hence, in order to optimize process time, the manipula-tor (i.e., the center of the body) requires maximum displace-ment to weld as much as possible without moving the feet.This body movement was stated as ±0.4 m along the longitu-dinal axis of the cell, ±0.2 m along the transversal axis of thecell, and about 0.5 m along the vertical axis of the cell. Themobile platform’s workspace is measured when the right andleft legs of the mobile platform are grasping alternating stiffen-

ers (see Fig. 4). To find the mobile platform’s workspace, thelength of the horizontal links has been computed in simulationas about 0.630 m. The stroke of the extensible link has beenselected as 0.5 m. Thus, the mobile platform can be adapted towork in many different cells. The prismatic joint in the verti-cal link provides a half-meter run, as required by cell specifica-tions. The dimensions of the body were found using the


1 2 3 4 5 6

z y

x

A A B B

Initial Positionat Stiffener A

Leg Trajectory fromStiffener A to B

Final Positionat Stiffener B

Figure 5. Leg motion sequence for releasing one stiffener and clasp-ing the next.

Welding Seam

Stiffener

Body

1760 mm

Welding Seam

Leg

2130

mm

Figure 4. Structure of the ROWER walking machine.

dimensions of the manhole and the layout of the subsystemson board the body. Finally, the body was defined as a 0.56 ×1.2 × 0.25 m parallelepiped. Table 2 summarizes the dimen-sions of the ROWER mobile platform. The leg elements inthis table are defined in Fig. 3.

Control SystemHardware/SoftwareThe control hardware is envisaged as a traditional robot controlsystem. It is located and distributed on board the body of themachine. The main controller is a PC-based computer thatworks as a stand-alone system. This control-ler is connected via a serial line to the super-visor computer. The joint controllers arespecially tailored cards based on the LM628microcontroller, which provides a program-mable PID control filter and a trajectorytracking command to follow joint trajecto-ries, as specified by the host computer.Drivers, also tailored for this application, are based on thePWM technique. The mobile platform is powered through atether containing dc power supply, communication with thesupervisor controller, and welding system power supply. Themobile platform pulls the tether during motion. It is attached tothe mobile platform in such a way that it does not interfere withthe area swept by the feet.

The control software is written in C++ language, and itruns on a real-time multitasking system for the DOS operatingsystem. This system is a library to be linked with the applica-tion program. It offers a number of functions for managingtasks, semaphores, mailboxes, interrupts, and so on. The mainprogram may be divided into the following modules: hostcommunications, gait generator, supervisor system, controllercommunications, and sensor system. This last module ismainly in charge of level sensors and positioning sensors.

Gait GenerationThe gait generator is one of the main modules of the control-ler. It takes care of the leg and body sequence of motions topropel the mobile platform forward, backward, and sideways.When the mobile platform is clasped to stiffeners, no static sta-bility problems appear. The body only moves when all fourlegs are clasped to stiffeners. The translation of the mobileplatform from one pose to the next is achieved by moving legsone at a time. One leg clasping the stiffeners (see 1 in Fig. 5) isreleased in the recovery of the expansible link. Then, the ver-tical link is lifted and placed in the middle of the top-bottomstiffener clearance (see 2 in Fig. 5). In this pose, the leg travelsat high speed to a location close to the next stiffener. In this lo-cation, the extensible link is extended and the vertical link islowered until the leg cannot pass through the top-bottomstiffener clearance (see 4 in Fig. 5). Now the speed is de-creased, and the leg moves forward until the next stiffener isdetected (see 5 in Fig. 5). At this stage, the vertical link is low-ered until the foot contacts the bottom stiffener, and the ex-

tensible link is expanded until the grasping process is com-pleted (see 6 in Fig. 5). With this leg motion algorithm, thefull motion of the mobile platform is achieved by first movingin sequence the two front legs, then propelling the body for-ward, and finally repeating the leg motion procedure with therear legs. The movement of the mobile platform parallel to thestiffeners is accomplished in a similar manner. Also, this struc-ture allows the body to rotate through about 40 degrees with-out moving the feet. This motion is sometimes required forthe manipulator to get to hard-to-reach areas. Full machinerotation can be accomplished by moving the body and legs in

sequence. For a 90-degree turn, the machine must performeight leg motions and three body rotations. Hence, the ma-chine exhibits high mobility to accomplish the required weld-ing tasks.

Stiffener DetectionIt was mentioned before that the ROWER walking machinecould move over the stiffener because the cell geometry is apriori given. However, shipbuilding procedures do not guar-antee the accuracy required for our purpose; hence, the use ofsensors to detect stiffeners with precision is essential.

Some problems arise at this point. The first one is that theorientation of the foot and the leg ankle changes as the legreaches the stiffener (see Fig. 5); therefore, it is not possible tolocate a simple range or contact sensor on the foot/ankle ofthe leg to detect the stiffener. Many simple sensors in a circularconfiguration are ruled out because of price and accommoda-tion. Another problem is the installation of electrical cables ina freely rotary foot and the installation of cables from connec-tor C2 to the foot (see Fig. 3). Cables on the bottom foot haveno special problems, but cables on the top foot need a floatinghose to allow the extensible link to change its length, and thatwill jeopardize functionality.

All these problems can be overcome by using a set of posi-tioning sensors already installed in the machine: the jointencoders. The idea is to monitor the positioning errors of ev-ery encoder on a leg to detect the foot/stiffener contact. Theadvantages of doing so are as follows:

◆ No additional sensors are needed. The encoders are al-ready installed because they are necessary for joint posi-tioning control tasks.

◆ No extra electrical cables are needed. Encoders are ade-quately and securely connected to the controller.

◆ The method and the algorithms are useful for detectingstiffeners in the approach phase and securing the grasp-ing as well; i.e., to detect the stiffener in the downward


Important improvements are made in weldingquality (much higher with a robotic system

than with a human operator), fewer hazards,and better operator working conditions.

motion of the vertical link and the upward motion of theextensible link. This saves two contact sensors or forcesensors as well.

To illustrate this method, an experiment has been con-ducted that consists of moving a leg against a stiffener as indi-cated in Fig. 5 (see leg positions 4 and 5). The initial footposition is (x, y) = (890, 1160) mm, and the final commandedfoot position is (x, y) = (890, 960) mm, where this y compo-nent is supposed to further the stiffener position. The foot iscommanded to follow a straight line parallel to the x-axis ofthe robot’s body reference frame (perpendicular to the stiffen-ers) and the foot position and joint positioning errors are re-corded at a given frequency. The controller analyzes the jointpositioning errors and stops foot motion when it detects an er-ror greater than a given threshold. In this stiffener detectionprocess only the joints of the horizontal links are involved.Figure 6 shows the joint positioning errors when a leg hits thestiffener at low speed. The joint positioning errors are small atfirst, but they grow quickly when the leg contacts the stiffener.

Figure 7 shows foot position along the defined trajectory.The x component does not change (foot motion follows astraight line parallel to the x-axis of the body) except at theend of the trajectory. At this point the leg exerts force againstthe stiffener, and the x component of the footreaccommodates because of the flexion and backlash in thewalking machine structure. The y component of the foot tra-jectory shows how the leg accelerates, then it moves at a con-stant speed, and finally it decelerates rapidly when the stiffeneris detected. The positioning error still grows, but the final po-sition does not change. At this point it is assumed that the stiff-ener has been detected, at footy = 961 mm, and the procedureof locating the lower foot on the stiffener is performed in asimilar way. Finally, the detection of the upper foot against thetop stiffener is accomplished following the same method.

Figure 5 illustrates how a leg grasps the stiffeners when theyare perfectly aligned. Normally, stiffeners are neither fullyaligned nor is the distance between top and bottom stiffenersconstant. In the first case the system still works because the legis just stopped when one of the two feet contacts a stiffener(top or bottom). In this situation the peg of the other foot willbe a few centimeters away from the stiffener, but the disk ofthe foot can contact the stiffener and exert the required forcedue to it has a big radius of about 10 cm. It is easy to under-stand from Fig. 5(a) that a leg can grasp the stiffeners even ifone of them is away from the plane of the other less than theradius of the foot disk.

The second problem is easier to overcome because the ex-tensible link is moved until the foot contacts the stiffener.Therefore, if the distance between top and bottom stiffeners isless than nominal, then the grasp will be achieved earlier thanexpected. On the other hand, if the distance is bigger thannominal, then the grasp will be performed latter than ex-pected, but it will be accomplished because the extensible ac-tuator has a maximum extension that makes the leg biggerthan the top-bottom distance. Of course, if either foot of theextensible actuator reaches the commanded final positionwithout detecting any stiffener, then the system will bestopped and the operator will be warned.

First Prototype and Final ResultsFigure 8 shows a picture of the ROWER walking machine ina double-bottom cell at the Fincantieri shipyard facilities, It-aly. The machine is carrying the manipulator, welding tools,and accessories. The leg, with the dimensions defined in Table2, can develop a speed of 0.15 to 0.4 m per second dependingon the position of the trajectory in the leg workspace. Thespeed achieved by its vertical link is about 0.1 m per secondthroughout its workspace.

With this real leg, the task of moving it from one stiffenerto the next is accomplished in about 5.5 s, while foot graspingand foot release take about 5 and 7 s, respectively. Thus themovement from one welding pose to the next is achieved inabout 80 s. By “welding pose” we mean the pose of the ma-chine on alternating stiffeners, as shown in Fig. 8. These speed


0 1 2 3 4 5 6−500

0

500

1000

1500

2000

2500

3000

Joint 2

Joint 1

Time (seconds)

Join

tPos

ition

ing

Err

or(c

ount

s)

Figure 6. Joint positioning errors during the stiffener detection pro-cess.

0 1 2 3 4 5 6Time (seconds)

X

Y

850

900

950

1000

1050

1100

1150

1200

Foo

tPos

ition

(mm

)

Figure 7. Foot position along the defined trajectory.

figures let the machine travel along the cell at an average speedof about 13 mm/s, which seems slow. The machine’s speed isnot significant, however, if compared to the speed of thewelding process achieved by a human operator. Note that fora single foot pose (the body of the machine can move ±0.4 malong the y-axis and ±0.2 me along the x-axis) there are 4 to 5m of multipass welds that an operator would take a long timeto perform. Furthermore, the arc-welding time of a roboticsystem can be half the manual arc-welding time. In addition tothese features, other important improvements are made inwelding quality (much higher with a robotic system than witha human operator), fewer hazards, and better operator work-ing conditions. These are the main improvements achieved bythis automatic welding system.

Accuracy is paramount in such a big machine, so special at-tention was paid to the design of links, mechanical transmis-sions, and gears. As a result, the machine displays positioningerrors of less than ±10 mm, which is good enough to performmotion all over the working cell. The misalignments betweentheoretical and actual walking machine positions are correctedwith the stereo vision system handled by the manipulator andused to detect the starting and ending welding seam pointswith precision. The main final features of the ROWER walk-ing machine are summarized in Table 3.

Summary and ConclusionsWalking machines have been developed for years as testbedsfor legged locomotion research. Now the technology is ma-ture enough to accomplish industrial applications. This articleproposed a novel legged machine to provide a mobile weldingsystem in a complex industrial environment, the double bot-tom of a ship, during the ship erection process. The machineuses the characteristics of the ship’s geometry to increase itsown stiffness and to guarantee walking stability. This articledescribed how the machine’s structure and dimensions havebeen tailored to accomplish tasks in the working scenario. Al-gorithms for gait generation and stiffener detection without

using additional sensors were also outlined. The machine hasbeen tested in a shipyard under real working conditions, andthe functionality of its working principle has been positivelyvalidated. By increasing the welding arc time, this system re-duces the total working time by one-third compared with thework done by human operators.

AcknowledgmentsThis work is part of the project “On Board AutomaticWelding System for Ship Erection” supported by EuropeanCommission DG XII under Brite-Euram contractBRE2-CT94-0925 and developed in collaboration withTecnomare S.p.A (Italy), Fincantieri S.p.A. (Italy) and AESA(Spain). Part of the research was funded by CICYT (Spain)under Grant TAP94-1549-CE. The authors would like tothank the many people who have helped with the configura-tion, design, construction, and programming of the machine,especially J. Reviejo and J. Tabera from the IAI-CSIC.

KeywordsWalking machines, legged locomotion, naval construction,shipbuilding.


Table 3. Features of the ROWERWalking Machine

Body mobility

X: ±0.2 m

Y: ±0.4 m

Z: ±0.25 m

Body speed 0.5 m/s

Leg speed 0.7 m/s

Foot grasping time 5 s

Foot releasing time 7 s

Machine average speed 13 mm/s

Figure 8. The ROWER walking machine.

Table 2. Dimension, Weight, and Payload of theROWER Walking Machine

Body

Height 0.25 m

Length 1.2 m

Width 0.56 m

Weight 50 kg

Leg

Vertical Links:AdaptableDisplacement oflinear tableDisplacement ofextensible link

0.535 m

0.5 m

0.5 m

Horizontal Links 0.63 m

Weight 65 kg

Mobile PlatformTotal Weight 310 kg

Payload 130 kg

References[1] J. Bares et al., ”Ambler: A walking robot for autonomous planetary explo-

ration,” IEEE Computer, vol. 22, no. 6, pp. 18-26, June 1989.[2] Y. Ishino et al., “Walking robot for underwater construction,” in Proc. Int.

Conf. on Advanced Robotics, 1983.[3] J.C. Grieco, M. Prieto, M.A. Armada, and P. Gonzalez de Santos, “A

six-legged climbing robot for high payloads,” in Proc. 1998 IEEE Int.Conf. Control Applications, Trieste, Italy, September 1998, pp. 446-450.

[4] J.S. Byrd and K.R. DeVries, “A six-legged telerobot for nuclear applica-tion development,” Int. J. Robot. Res., vol. 9, no. 2, pp. 43-52, Apr.1990.

[5] J. Bares and D. Wettergreen, “Dante II: Technical description, results, andlesson learned,” Int. J. Robot. Res., vol. 18, no. 7, July 1999, pp. 621-649.

[6] A. Halme and K. Hartikainen, “Designing the control system of an ad-vanced six-legged machine,” in J.O. Gray and D.G. Caldwell (Eds.), Ad-vanced Robotics & Intelligent Machines. London: IEE, 1996, pp. 177-190.

[7] S. Hirose, “A study of design and control of a quadruped walking vehicle,”Int. J. Robot. Res., vol. 3, no. 2, pp. 113-133, Summer 1984.

[8] D.R. Pugh et al., “Technical description of the adaptive suspension vehi-cle,” Int. J. Robot. Res., vol. 9, no. 2, pp. 24-42, April 1990.

[9] P. Gonzalez de Santos, M.A. Armada, and M.A. Jimenez, “Walking ma-chines: Initial testbeds, first industrial applications and new research,”Computing and Control Eng. J., vol. 8, no. 5, pp: 233-237, 1997.

[10] P. Gonzalez de Santos, M.A. Armada, and M.A. Jimenez, “An industrialwalking machine for naval construction,” in Proc. IEEE Int. Conf. Robot-ics and Automation, Albuquerque, NM, April 1997.

Pablo Gonzalez de Santos is a research scientist at theSpanish Council for Scientific Research (CSIC). He receivedhis Ph.D. degree from the University of Valladolid (Spain).Since 1981, he has been involved actively in the design anddevelopment of industrial robots and also in special roboticsystems. His work during the last ten years has focused onwalking machines. He worked on the AMBLER project as avisiting scientist at the Robotics Institute of Carnegie MellonUniversity. Since then, he has been leading the developmentof several walking robots such as the RIMHO robot designedfor intervention on hazardous environments, the ROWERwalking machine developed for welding tasks in ship erectionprocesses, and the SILO4 walking robot intended for educa-tional and basic research purposes. He has also participated inthe development of other legged robots such as the REST

climbing robot and the TRACMINER. Dr Gonzalez deSantos is now studying how to apply walking machines to thefield of humanitarian demining.

Manuel A. Armada received a Ph.D. in physics fromValladolid University (Spain) in 1979. He has been involvedsince 1976 in research activities related to automatic control(singular perturbations and aggregation applied to bilinear sys-tems; adaptive and nonlinear control; multivariable systems inthe frequency domain; digital control). He has worked onmore than 40 RTD projects and has published over 150 pa-pers (including contributions to several books, monographs,journals, international congresses, and workshops). He is cur-rently the head of the Automatic Control Department at theInstituto de Automatica Industrial (IAI-CSIC), with his mainresearch direction concentrated in robot design and control,with special emphasis in new application fields such as flexiblerobots and walking and climbing machines.

Maria A. Jiménez is currently a tenured scientist at theSpanish Council for Scientific Research (CSIC). She receivedher B.S. in physics and Ph.D. degrees from the University ofCantabria (Spain) in 1986 and 1994, respectively. In 1987, shejoined the Department of Automatic Control at IAI-CSIC towork in several projects related to the design and developmentof special robotic systems. Her work during the last ten yearshas been focused on walking machines. So, her Ph.D. thesiswas focused on the generation of wave gaits and adaptability toirregular terrain. She was the local project manager of thePalaiomation Project for building a walking dinosaur. Also,she has been involved in several projects related to walkingmachines for naval applications. Her research interests includeconfiguration, simulation, and implementation of autono-mous walkers.

Address for Correspondence: P. Gonzalez de Santos, IndustrialAutomation Institute-CSIC, La Poveda, 28500 Arganda delRey, Madrid, Spain. E-mail: [email protected].


ship building with rower

Documents