Field and Service Robotics - Corke P. and Sukkarieh S.(Eds) Part 13 doc

The given nominal trajectory may be infeasible due to the structural limitation of the robot; so the robot searches for a feasible one within the tolerance.. Only whenthe robot fails to

Fig 8 A nominal trajectory.

side view

door

top view door

Fig 9 Tolerance for segment CD in Fig 8.

We, therefore, develop a novel teaching method for a mobile manipulator whichexists in between the above two approaches In the method, the user teaches the

robot a nominal trajectory of the hand and its tolerance to achieve the current target

task The given nominal trajectory may be infeasible due to the structural limitation

of the robot; so the robot searches for a feasible one within the tolerance Only whenthe robot fails to find any feasible trajectory, it plans a movement of the mobilebase, by using the redundancy provided by the mobile base as another tolerance Theteaching method is well intuitive and does not require much user’s effort because theuse does not have to consider the structural limitation of the robot in teaching Atthe same time, the method does not assume a high recognition and inference ability

of the robot because the given nominal trajectory has much information for motionplanning; the robot does not need to generate a feasible trajectory from scratch.The following subsections explain the teaching method, using the task of openingthe door of a refrigerator as an example

4.1 Nominal Trajectory

A nominal trajectory is the trajectory of the hand pose (position and orientation)

in a 3D object-centered coordinate system To simplify the trajectory teaching, wecurrently set a limitation that a trajectory of hand position is composed of circularand/or straight line segments Fig 8 shows a nominal trajectory for opening a door,composed of straight and circular segments on some horizontal planes; on segment

CD, the robot roughly holds the door, while on segment DE, the robot pushes it at adifferent height The hand orientation is also specified as shown in the figure

4.2 Tolerance

A tolerance indicates acceptable deviations from a nominal trajectory to perform

a task; if the hand exists within the tolerance over the entire trajectory, the task isachievable A user teaches a tolerance without explicitly considering the structurallimitation of the robot Given a nominal trajectory and its tolerance, the robot searchesfor a feasible trajectory

Fig 11 A feasible region.

The user sets a tolerance to each straight or circular trajectory using a coordinatesystem attached to each point on the segment In these coordinate systems, a user

can teach a tolerance of positions relatively intuitively as a kind of the width of

the nominal trajectory Fig 9 shows an example of setting a tolerance for circularsegment CD in Fig 8, which is for opening the door

4.3 Generating Feasible Trajectories

The robot first tries to generate a feasible trajectory within a given tolerance Whenthe robot fails to find a feasible one, it divides the trajectory into sub-trajectoriessuch that each sub-trajectory can be performed without movement of the base; italso plans the movement between performing sub-trajectories

On-line Trajectory Generation The robot sets via points on the trajectory with

a certain interval, and generates a feasible trajectory by iteratively searching forfeasible hand poses for the sequence of via points This trajectory generation is per-formed on-line because the relative position between the robot and the manipulatedobjects may vary from time to time The robot estimates the relative position beforetrajectory generation The previously calculated trajectories are used as guides forefficiently calculating the current trajectory

Fig 10 illustrates how a feasible trajectory is generated; small circles indicatevia points on the given nominal trajectory, two dashed lines indicate the boundary ofthe tolerance, the hatched region indicates the outside of the range of possible handposes A feasible trajectory is generated by searching for a sequence of hand poseswhich are in the tolerance and near to the given via points In the actual trajectorygeneration, the robot searches the six dimensional space of hand pose

During executing the generated trajectory, it is sometimes necessary to estimatethe object position Currently, we manually give the robot a set of necessary sensingoperations for the estimation

Trajectory Division Based on Feasible Regions The division of a trajectory is

done as follows For each via point, the robot calculates a region on the floor inthe object coordinates such that if the mobile base is in the region, there is at leastone feasible hand pose By calculating the intersection of the regions, the robotdetermines the region on the floor where the robot can make the hand follow the

D

E V

X Y

Fig 12 Example feasible regions.

mobile base

6 DOF arm with hand

laser range finder

hand camera

host computer

Fig 13 Our service robot.

entire trajectory Such an intersection is called a feasible region of the task (see Fig.

11) The robot continuously updates the feasible region, and if its size becomes lessthan a certain threshold, the trajectory is divided at the corresponding via point Fig

12 shows example feasible regions of the trajectory of opening the door shown inFig 8 The entire trajectory is divided into two parts at point V ; two correspondingfeasible regions are generated

5 Prototype System and Experiments

Fig 13 shows our personal service robot The robot is a self-contained mobilemanipulator with various sensors In addition to the above-mentioned functions, therobot needs an ability to move between a user and a refrigerator The robot uses thelaser range finder (LRF) for detecting obstacles and estimating the ego-motion [8]

It uses the LRF and vision for detecting and locating refrigerators and users Fig 14shows snapshots of the operation of fetching a can from a refrigerator to a user

6 Summary

This paper has described our personal service robot The robot has user-friendlyhuman-robot interfaces including interactive object recognition, robust speech recog-nition, and easy teaching of mobile manipulation

Currently the two subsystems, object and speech recognition and teaching ofmobile manipulation, are implemented separately We are now integrating these twosubsystems into one prototype system for more intensive experimental evaluation

Acknowledgment

This research is supported in part by Grant-in-Aid for Scientific Research from istry of Education, Culture, Sports, Science and Technology, and by the KayamoriFoundation of Informational Science Advancement

Min-approach open grasp

Fig 14 Fetch a can from a refrigerator.

References

1 Morpha project, http://www.morpha.de/.

2 R Bischoff Hermes – a humanoid mobile manipulator for service tasks In Proc of

FSR-97, pp 508–515, 1997.

3 R Bischoff and V Graefe Dependable multimodal communication and interaction with

robotic assistants In Proc of ROMAN-2002, pp 300–305, 2002.

4 M Ehrenmann et al Teaching service robots complex tasks: Programming by

demon-stration for workshop and household environments In Proc of FSR-2001, pp 397–402,

2001

5 K Ikeuchi and T Suehiro Toward an assembly plan from observation part i: Task

recognition with polyhedral objects IEEE Trans on Robotics and Automat., Vol 10,

8 J Miura, Y Negishi, and Y Shirai Mobile robot map generation by integrating

omnidi-rectional stereo and laser range finder In Proc of IROS-2002, pp 250–255, 2002.

9 J Pineau et al Towards robotic assistants in nursing homes: Challenges and results

Robotics and Autonomous Systems, Vol 42, No 3-4, pp 271–281, 2003.

10 N Roy et al Towards personal service robots for the elderly In Proc of WIRE-2000,

2000

11 M Takizawa et al A service robot with interactive vision – object recognition using

dialog with user – In Proc of Workshop on Language Understanding and Agents for

Real World Interaction, pp 16-23, 2003.

Shelving Facility

Jackrit Suthakorn1, Sangyoon Lee2, Yu Zhou2, Sayeed Choudhury3,

and Gregory S Chirikjian2

Mahidol University, Bangkok, Thailand

song@jhu.edu or jackrit@trs.ac.th

2

Department of Mechanical Engineering

The Johns Hopkins University, Baltimore, Maryland 21218 USA

slee@konkuk.ac.kr, yuzhou@titan.me.jhu.edu, gregc@jhu.edu

3

Digital Knowledge Center of the Sheridan Libraries

The Johns Hopkins University, Baltimore, Maryland 21218 USA

sayeed@jhu.edu

Abstract This paper describes our continued work of a unique robotics project,

Comprehensive Access to Printed Materials (CAPM), within the context of libraries As libraries provide a growing array of digital library services and resources, they continue to acquire large quantities of printed material This combined pressure of providing electronic and print-based resources and services has led to severe space constraints for many libraries, especially academic research libraries Consequently, many libraries have built or plan to build off-site shelving facilities to accommodate printed materials An autonomous mobile robotic library system has been developed to retrieve items from bookshelves and carry them to scanning stations located in the off-site shelving facility This paper reviews the overall design

of the robot system and control systems, and reports the new improvement in the accuracy of the robot performance; in particular, the pick-up process.

As libraries provide a growing array of digital library services and resources, theycontinue to acquire large quantities of printed material This combined pressure ofproviding electronic and print-based resources and services has led to severe spaceconstraints for many libraries, especially academic research libraries Consequently,many libraries have built or plan to build off-site shelving facilities to accommodateprinted materials However, given that these locations are not usually within walkingdistance of the main library, access to these materials, specifically the ability tobrowse, is greatly reduced Libraries with such facilities offer extensive physicaldelivery options from these facilities, sometimes offering multiple deliveries per day.Even with such delivery options, the ability to browse in real-time remains absent.The goal of the CAPM Project is to build a robotic, on-demand and batch scanningsystem that will allow for real-time browsing of printed materials through a webinterface We envisage the system will work as follows: an end user will identify that

a monograph is located in an off-site facility The user will engage the CAPM systemthat, in turn, will initiate a robot that will retrieve the requested item The robot willdeliver this item to another robotic system that will open the item and turn the pages automatically By using existing scanners, optical character recognition (OCR)software, and indexing software developed by the Digital Knowledge Center (aresearch and development unit of the Sheridan Libraries at Johns Hopkins), the

S Yuta et al (Eds.): Field and Service Robotics, STAR 24, pp 437–446, 2006.

© Springer-Verlag Berlin Heidelberg 2006

CAPM system will not only allow for browsing of images of text, but also for searching and analyzing of full-text generated from the images.

The details of the mechanical structure, the navigation system, control andsoftware, simulations, experiments and results of the robot were previously described

in [1] This paper focuses on improvement in the accuracy of entire deliveryprocedure of the robot system, in particular, the pick-up of bookcases, while thefuture work, will be concentrated on developing the robotic system to complete theremaining processes

Since the CAPM robot is designed to work in an off-site shelving facility thatbelong to the Johns Hopkins University, several assumptions in the design are madebased on the actual environment of this facility All the paths in the facility areassumed to be smooth and flat Each book is assumed to be stored in a special case,which has a pair of wing-like handles for engaging with a passive gripper It isassumed that each item is stored in a specifically designed case and arranged side-by-side with a small in-between gap Finally, a barcode is attached to each case

Currently at the Moravia Park shelving facility, after receiving a request, a library officer at the facility will drive a portable personnel lift to retrieve the requested item

to its location, and then bring it to a waiting area for the next scheduled transportationThen, a batch of requested items is transferred to the main library In the same manner,our robot will be initially parked at the docking station until an item is requested Therobot is equipped with a database system of book locations and a global map of theoff-site shelving facility After receiving a request, the robot will autonomously runalong a known path to the book location and retrieve the requested item from the shelf.Then, the robot will carry the item back to the scanning station and then return to thedocking station

This is not the first time a robot has been built to perform a specific servicefunction In 1995, Hansson introduced an industrial robot in a Swedish library [2].Safaric [3] presented an example of a telerobot controlled via Internet [4] Byrdintroduced a successful service robot used to survey and inspect drums containinglow-level radioactive waste stored in warehouses at Department of Energy facilities [5]

The CAPM system differs from other existing systems in the following ways.First, the system retrieves individual items, as opposed to boxes of items, such as thesystem at the California State University at Northridge [6] Second, the CAPMsystem does not assume an existing or fixed shelving and space arrangement Thisflexibility will allow it to work in many diverse environments Third, the CAPMretrieval robot is an autonomous system Fourth, the economic analysis by acollaborating research group in the Department of Economics at Johns Hopkins

University has verified that a relatively inexpensive robotic system is cost-effective,

especially in comparison to potential benefits Finally, the page-turning system, to bebuilt in the future, will accommodate a wide variety of paper types and materials

In subsequent sections of this paper, we report the design, control systems,experiments and results of an autonomous robotic library system for an off-siteshelving facility Sections 2 and 3 briefly review the robot design, the robot controlsystems and software and navigation system (detailed descriptions can be found in [1,6].) Section 4 explains improvement in the accuracy of entire delivery procedure ofthe robot system We then report experiments and results in Section 5

2 Hardware of the Robotic System

2.1 Mechanical Structure

This section presents designs and descriptions of two major components of the CAPMlibrary robot: the manipulator arm and the locomotion device

2.1.1 Manipulator arm system

In order to retrieve books from bookshelves and carry them to the scanning station, aspecific manipulator arm system was designed Since each bookshelf is 10-foot-high,

a vertical translation system (VTS) was used to move the robot manipulator todifferent altitudes The VTS is a sliding rod with an electric motor for driving a lead-screwed rod An enhanced commercial 6-DOF robot manipulator, the F3 made byCRS Robotics, Inc., is affixed to a platform which is a part of the vertical translationsystem (See Figure 1.)

We built and installed a passive gripper to the end-effector of the robotmanipulator The gripper is used to passively grasp the bookcase The structures ofthe gripper and bookcases were designed to fit to each other A barcode scanner isinstalled inside the gripper in order to recognize and ensure the precision of picking arequested item

2.1.2 Locomotion device

The locomotion device is responsible for the gross motion of the robot We havemodified a commercial servo-controlled mobile robot platform, the Labmate made byHelpmate Inc An aluminum-alloy cart is built and attached to the Labmate mobileplatform This cart is used to store the robot manipulator controller and the powersource while the Labmate mobile platform is used as the base of the manipulator armsystem A ranging sensor system was installed on the mobile platform to collaborateand improve the navigation system All electronic devices used to control the verticaltranslation system and sensor systems were installed on the mobile platform Because

of the installation of a power source onboard, the robot does not require an externalpower line while working Figure 1 shows the overall mechanical structure of thelibrary robot

In our work, due to limits of sensor performance, 8 sensors are used: 4 sonarsensors and 4 infrared sensors One sonar and one infrared sensor are paired together

to get each of the 4 sensor readings needed This is done because of the distancemeasuring limits which each have The Polaroid 6500 sonar sensor has a range of 15-

1067 cm while the Sharp GP2D02 infrared sensor has a range of 10-80 cm It can beobserved that by using these two sensors combined, we can achieve a range of 10-

1067 cm of reliable distance measurement Each sensor is controlled and interfaced

to the main computer via a micro-controller (BASIC Stamp II)

All the processes and activities of the system are controlled by an onboard IntelPentium II laptop The control systems of the library robot consist of several sub-controllers: the control system of the VTS, the control system of the robot

manipulator, the control system of the mobile platform, the high-level control system

of the library robot, and the control software

Fig.1 The Robotic Library System

3.2 Control of the Robot Manipulator

The six-axis F3 robot system manufactured by CRS Robotics Inc is used as thecontrol system of the robot manipulator Articulated joints provide the F3 arm withsix degrees of freedom, and absolute encoders mounted on the motor shaft in eachjoint provide positional feedback to the controller The F3 robot arm uses theCartesian coordinate system

Control programs were written in the C++ language and downloaded to thecontroller Control programs use the ActiveRobot interface developed by CRSrobotics and include two object classes of the ActiveRobot interface: one provides theapplication with the main interface to the robot system, and the other enables theapplication to create and modify robot locations The input variables to the controlprograms are the speed and the position and orientation of the final location of theend-effector The controller provides the computation of inverse kinematics

Sonar/IR Sensor System for Robot Navigation

Locomotion Device

Robot Manipulator

IR Sensors for Manipulator Motion Planning

Gripper Vertical

Translation Device

3.3 Control of the Mobile Platform

The locomotive device, or the mobile platform Labmate, has the drive systemmicroprocessor that the user's host computer communicates with through an RS-232serial port The host computer always initiates communications between the host computer and the Labmate The Labmate uses a Cartesian coordinate system forposition control The coordinate system is a global reference that is initialized at power up and reset Odometry (dead reckoning) is the practice of calculating positionfrom wheel displacements The Labmate control system depends on encodersmounted on each wheel to keep track of position Control programs are basically composed of commands that direct the Labmate to a particular location We assumehere that the entire map of the workspace is stored in the form of a look-up table inthe memory of the Labmate If a destination is given, the Labmate computes thedirection from the current location by referring to the look-up table To compensatefor the errors, we used two kinds of sensors: ultrasonic ranging sensors and infraredsensors For more successful navigation of an autonomous vehicle over extendeddistances, references to the external world at regular intervals are necessary Figure 2 illustrates the diagram of the library robot controls

3.4 Control Systems of the Library Robot

We call the main control system of the library robot the “high-level control system”.This control system consists of a Pentium II 233 MHz computer notebook and a serialport splitter hub The computer notebook functions as the central processing unit ofthe library robot, and it communicates to every subsystem through the serial ports

3.5 Control Software

The control software is designed based on the idea of event driven programming.Principally, the main control program controls the mobile platform, the verticaltranslation device, and the arm through serial ports When the main control programbegins to run, it initializes the serial ports of the computer at first and starts the event listeners for all the serial ports Then the main control program leaves the control tothe listeners It is actually these event listeners that control the movement of theplatform, the vertical translation device, and the arm Basically, each event listener will monitor the status of one serial port Once the status of that port changes, thelistener will judge what kind of event happens and execute a corresponding function

We use the word ‘lift’ interchangeably with ‘the vertical translation device’

An important property of event driven programming is that the execution order ofthe functions is not fixed, it depends on the need to execute This property is suitablefor the sensor driven system of the library robot In total four event listeners arecreated They monitor the status of the platform, the lift, the arm, and the sensors respectively Figure 3 shows the software structure of the robot

To assist the motion planning of the library robot, the complete working process ofthe library robot was simulated using 3DSMAX Based on the simulation, a completepath was generated, and experiments were executed to test and adjust the performance

of the robot To simplify the implementation, a map-based scheme was employed to

control the mobile platform A fixed global coordinate system is defined with itsorigin at the docking spot The positions of the intermediate stops and the destinationwere defined in the global system An optimal path was chosen to connect the currentstop and the next stop The major problem appearing in the experiments is positioningerror It was found that the positioning error was closely related to the moving speed

of the platform If the speed was low, the motor may lose some steps because of thecertain heavy overall load If the speed was high, the platform may deviate from thedesired path at the turning corners because of the inertia After a few adjustments ofthe speed setting, the positioning was improved considerably

Fig.2 Diagram of the library robot controls.

Fig.3 Software structure of the library robot.

To enhance the accuracy of pick-up process in the robot manipulator controls, weintegrated an infrared sensing system to the manipulator The infrared sensing systemconsists of two infrared sensors (Sharp GP2D02), the sensor controller, and aninput/output serial COM port to communicate with the high-level controller The twoinfrared sensors were attached to the fingertips of the end-effecter, and are controlled

by a circuit controller made of a BASIC Stamp II microchip and other required

circuitry components Figure 4 shows a picture of the IR sensors installed on thegripper The high-level controller communicates with the infrared sensing systemthrough the I/O serial com port The high-level controller receives the ranginginformation from the sensor and converts this information to an updated data file.

In the pick-up process, once the library robot reaches the position in front of thedesired bookshelf, the robot (which already has the desired book's coordinates) startsthe pick-up process However, to eliminate the error that may occur by slightly miss-positioning the book, the book-position-scanning process was added

Fig.4 IR sensors installed on the gripper.

Because the manipulator picks up a book in the same manner as a forklift, thepositioning of the book is critical Figure 5 shows the motion planning of the roboticmanipulator in the pick-up process The book-position-scanning process is a process

to correct the book-positioning errors This allows the book to be placed in a specificranging position instead of only at an exact position The control architecture of thebook-position-scanning process is illustrated in Figure 6

The book-position-scanning process normally begins with the scanning a certainrange of the bookshelf by the IR sensors The robot uses the known book positionstored in its database as the reference position The manipulator begins scanning fromleft to right in the horizontal direction relative to the reference position The scanningpath starts at the position located at 40 mm to the left of the reference position, andthen moves to the position located at 40 mm to the right of the reference position,stopping at every 5 mm increment This scanning creates 17 positions along the path.All of these positions are located in the same X-Z plane, which means that only the X-coordinate values vary At each stop, the median of distance data is calculated andstored in the computer After the scanning is finished, a set of median distance values

is used to find the exact range of the bookcase

We developed a computer algorithm for the process and obtained remarkableimprovement in accuracy from tests The algorithm was implemented in C++ tocontrol the scanning path of the robot manipulator The code picks up the datainformation from infrared sensing system and analyzes these data to generate anactual book location at the end Once the actual book location is generated, the robot updates the pick-up book position and executes picking-up the book

Fig.5 Motion planning of the robotic manipulator in the pick-up process.

Fig.6 Control architecture of the manipulator’s sensing system.

Process

Experiments on the robot manipulator were conducted to determine the accuracy ofthe pick-up process performed by the manipulator and its feedback control system.The experimental set-up, results, and error analysis are described below

To verify the accuracy of the pick-up process, we set up the experiment by placing the library robot in the perpendicular direction to the bookshelf (see Figure 7.)The robotic manipulator and its feedback control are designed to be able to scan andpick up a bookcase in a certain range of -40 mm to +40 mm from the reference point,where this range is called “active range.” The active range is located on a bookshelf

by marking the far-left, reference, and far-right positions, along with a set of finemeasurements in increments of 5 mm

In each experiment, a bookcase was randomly placed in the active range Werecorded the number of successful and failed trials in attempting to pick up abookcase at each position We performed the experiments three times at each position,

Wing-Liked Handle

without order of positioning Figure 7 shows the experiment set-up of the roboticmanipulator tests, and in the following part the experimental results will be described.There are no failed trials (out of 51 trials) in this experiment Figure 8 shows thestep-by-step picking up process of the experiment The robot manipulator started at itsinitial position and begins scanning from far-left to far-right positions After thealgorithm determines and finds the actual book position, the robot manipulatorexecuted the picking-up process.

An autonomous robotic library system was built as a prototype The robot design,control systems, simulations, experiments and results were presented Animplementation using IR sensors and a new algorithm to enhance the accuracy of anoperation process, book pick-up process, was described, and reported its successfulexperimental results

While these points outline the specific benefits and qualities of CAPM, it is important to note a ultimate goal of this project The CAPM Project will introducerobotics into the library and, perhaps more importantly, digital library context As robotics have provided great impact and utility within manufacturing and,increasingly, computer-assisted surgery, it is possible that similar gains will beachieved in developing digital libraries Already, some cultural heritage faculty andlibrarians have identified CAPM as the best option for digitizing the vast amounts ofknowledge retained in print format Additionally, through batch scanning, CAPMwill produce automatically and systematically preserved copies of printed materials.Future work includes the improvement of the robot navigation system, and thedevelopment of the page-turning robotic system for the next stage of the CAPMproject

Fig.7 Picture shows the experiment set up for testing the robotic manipulator.

Fig.8 Picture shows step-by-step in pick-up experiment: 1) Manipulator starts at the process's

initial position, 2) Manipulator starts the scanning process from the far-left position, 3) Manipulator finishes the scanning process at the far-right position, 4) Manipulator begins picking up the bookcase 5) During the pick-up process, and 6) Manipulator successfully picked

up the bookcase.

Reference

Library System for an Off-Site Shelving Facility,” Proceedings of the IEEE International

Conference on Robotics and Automation (ICRA) 2002, Washington DC, May 2002.

pp.16-18, 1995.

Proceedings of the IEEE International Symposium on Industrial Electronics 1999, Vol 1,

1999 pp 298-303.

Remote Systems, Monterey, CA, Feb 1995.

5. http://library.csun.edu/

Mechanical Engineering, The Johns Hopkins University, April 2003.

S Yuta et al (Eds.): Field and Service Robotics, STAR 24, pp 447–456, 2006.

© Springer-Verlag Berlin Heidelberg 2006

International Contest for Cleaning Robots: Fun Event

or a First Step towards Benchmarking Service Robots

Erwin Prassler1, Martin H¨agele2, and Roland Siegwart3

1 Gesellschaft f¨ur Produktionssysteme

Nobelstr 12, 70569 Stuttgart, Germany

prassler gps-stuttgart.de

2 Fraunhofer Institute for Manufacturing Engineering and Automation

Nobelstr 12, 70569 Stuttgart, Germany

haegele ipa.fhg.de

3 Ecole Polytechnique Federale de Lausanne, Autonomous Systems Lab

CH-1015 Lausanne, Switzerland

roland.siegwart epfl.ch

Abstract In this paper we report on the First International Contest for Cleaning Robots,

which took place jointly with IROS 2002 in Lausanne, Switzerland The event had twoprimary objectives As an educational event with a significant fun factor it was supposed toattract the brainpower and activate the creativeness of students and young researchers for

an application of service robotics, which has a significant economic potential The cost forcommercial cleaning services is estimated at around US$ 50 billion per year only in Europe

A fair contest, of course, required that all contestants had equal race conditions, This in turnrequired to have a well-define set up, which could be reproduced for every contest team andfor any single run With that, the second major objective of the event, which was to define abenchmark for robotic cleaning, was a natural byproduct of organizing a fair contest

1 Introduction

Are the times, when one had to spend tedious hours cleaning the kitchen or thebathroom or the mess in the children’s bedroom finally over? It sounds like adream, but there seems to be some hope at the horizon In the beginning of October

2002, teams of students and young researchers from all over the world gathered

in Lausanne in Switzerland at the beautiful lake of Geneva to compete in the firstworld championship for cleaning robots The championship was held at the EcolePolytechnique Federale de Lausanne (EPFL) jointly with the 2002 InternationalConference on Intelligence Robots and Systems IROS 2002 one of the largest annualrobotics conferences world wide In the beginning it was intended by the organizers

as a trial to attract the brain power of students and young researcher to an application

of robotics technology which seemed to be far less funny and entertaining thansoccer playing robots but has a significant economic potential: cleaning robots Themarket for cleaning services in Europe alone is estimated to be US$ 50 billion peryear

The organizers were everything but sure that what they called the “Robo-cupeffect” could be transferred to an application such as cleaning and that a cleaningrobot contest could attract only roughly as many students as a contest of soccer

playing robots Their worries were unjustified The contest became a big success.Fifteen teams from 10 countries worldwide participated in the contest The eventreceived overwhelming media-coverage and attracted not only public and academicaudience but also a significant number of industrial representatives.

The contest, however, was not only a remarkable event from a technology keting point of view There was a second very important aspect tied with the event.The set-up of the contest held in Lausanne was a first step towards a benchmark forservice robots At large the task and the conditions under which it had to be solvedwere clearly defined by an independent entity in such a way that it can be fully repro-duced at any time by anybody In Lausanne the final evaluation of the performance ofthe competing system for various reasons was still left to the subjective assessment

mar-of jurors introducing some uncertainty in the performance measurement However,

it will be only a cost issue, to replace the human jurors by a technical measuringsystem such guaranteeing a performance measure and comparison, which is free ofany subjective assessment

Of course, it may not be straightforward to transfer the set-up of contest inLausanne to an application other than cleaning At the same time is evident, however,that it will be extremely difficult for any service robot, let it be a cleaning robot orany other robot assisting the human, to mature to a state ready for production and

to find its way into the market without clearly defined performance measures andfigures Having a Formula 1 in service robotics does not only mirror the state ofthe available technology It may also significantly reduce the duration of innovationcycles and the time to market of this new generation of robots Benchmarks willnot only push the engineering part of the development of those systems They willalso push the development of robust and sound algorithms A lesson, which could

be learnt from the contest in Lausanne, was that even seemingly trivial everydaytask always leaves enough room not only for ad-hoc heuristics but also for soundalgorithmic solutions

In the following section we will first describe the set-up of the first internationalchampionship for cleaning robots This includes details such as the set-up of thecontest area, the contest rules or the measurement and scoring procedure Thissection is followed by an overview of the teams which competed in the contesttogether with an short overview of their systems and the employed cleaning strategy

In Section 3, we further present the results of the contest including the qualificationand the semi-finals In the final section we subject the event in Lausanne a criticalreview We will discuss some problems, which could hardly be avoided in such afirst event but may be circumvented in the future These problems are of particularinterest if the contest is to be further developed towards a true benchmark for cleaningrobots

2 The Set-Up of the Contest

The contest was arranged in of three sub-contest:

• a contest robotic floor cleaning in a “regular” living room

• a contest robotic window cleaning

• an idea contest “household robot”

Due to the lack of space we will focus in this paper on the floor cleaning sub-contest.This was the event which attracted the most attention amongst the teams as well asamongst the audience and press.

2.1 Overall Structure of the Contest

The contest for robotic floor cleaning took place in two adjacent identical contestareas each 5 × 5 square meters in size and furnished as a typical living room with

a black linoleum surface floor The two adjacent contest areas are shown in Fig 1.Wall-like barriers approx 80 in high surrounded these areas to prevent the audiencefrom entering and the robots from escaping the contest areas The model layout ofthe contest area was published about two weeks before the contest

Fig 1 Contest areas for robotic floor cleaning

The contest was organized very

much like a soccer championship In

total, 12 teams from seven countries

accepted the challenge and ran for the

floor cleaning contest These 12

com-peting teams were divided into four

groups of equal size, each group

con-sisting of three teams In the

qualifi-cations, the best team was determined

in each of the four groups This was

done in a pair wise direct

competi-tion Within the groups each team had

to compete with the two other teams,

which means that each group had to

participate in two runs This resulted

in three runs per group and 12 runs in

total in the qualifications

Two competing teams run their

cleaning robots in parallel in the two

adjacent contest areas This schedule

enabled the audience and jury to

ob-serve the performance of the robots

in direct comparison and also

signifi-cantly contributed to the entertainment value of the contest A single run was limited

to 10 minutes During this period, the two robots had to operate autonomously inthe contest areas and clean an area as large as possible The robots were placed andswitched on upon a signal by a member of the jury

According to the rules published before the contest, the teams were not permitted

to enter the contest area during a run, otherwise they would be disqualified Forpragmatic reasons, this rule could not be sustained Too many robots got stuck incorners or underneath furniture So it was decided by the jury and the organizers toallow interventions but to punish each intervention with a loss of 33% of the scores