Information Technology

Humanoide Fußballroboter

Humanoid Soccer Robots

Dr. Sven Behnke: Albert-Ludwigs-University of Freiburg, Computer Science Institute,Georges-Kohler-Allee, 52, 79110 Freiburg, GermanyTel: +49-761-2037404, Fax: +49-761-2038222, E-Mail: behnke@informatik.uni-freiburg.de

Dr. rer. nat. Sven Behnke studied Computer Science and BusinessAdministration at Martin-Luther-Universitat Halle-Wittenberg (and 1994/95 at University ofHouston), from where he received his MS CS (Dipl.-Inform.) in 1997. Afterwards he moved toFreie Universitat Berlin, where he received his PhD in Computer Science in 2002. During 2003,he worked as a postdoc at the International Computer Science Institute in Berkeley, CA. Since2004, he heads the junior research group Humanoid Robots at Albert-Ludwigs UniversitatFreiburg. His research interests include autonomous intelligent systems and neural informationprocessing.

Keywords: Humanoid Robots, RoboCup, Soccer

Schlagworte: Humanoide Roboter, RoboCup, Fußball

MS-ID: behnke@informatik.uni-freiburg.de 8th May 2005

Heft: X/ (2005)

Abstract

Humanoid robots are enjoying increasing popularity as a research tool. As step towards the long-termgoal of winning against the FIFA world champion, the RoboCup Federation added in 2002 a league forhumanoid robots to its annual soccer competitions. In this paper, the Humanoid League competitionsthat took place so far are reviewed. The different approaches for the design of robot hardware and thesoftware for perception and behavior control of soccer playing humanoid robots are discussed. The paperconcludes with an outlook to the future of the RoboCup Humanoid League.

Zusammenfassung

Humanoide Roboter erfreuen sich zunehmender Beliebtheit als Forschungsgegenstand. Als Schritt inRichtung des langfristigen Ziels gegen den FIFA-Weltmeister zu gewinnen, veranstaltet die RoboCup-Federation seit 2002 auch Fußballwettbewerbe fur humanoide Roboter. Der Artikel blickt auf die Wett-bewerbe der humanoiden Liga zuruck, die bislang stattfanden. Die unterschiedlichen Ansatze des Designsder Roboterhardware und der Software zur Wahrnehmung und Verhaltenssteuerung fur fußballspielendehumanoide Roboter werden diskutiert. Zum Schluss gibt der Artikel einen Ausblick auf die Zukunft derhumanoiden Liga des RoboCup.

2

1 Introduction

Humanoid robots, robots withan anthropomorphic body planand human-like senses, are enjoy-ing increasing popularity as re-search tool. More and moregroups worldwide work on issueslike bipedal locomotion, dexterousmanipulation, audio-visual per-ception, human-robot interaction,adaptive control, and learning,targeted for the application in hu-manoid robots.

These efforts are motivated bythe vision to create a new kindof tool: robots that work in closecooperation with humans in thesame environment that we de-signed to suit our needs.

While highly specialized in-dustrial robots are successfullyemployed in mass production,these new applications require adifferent approach: general pur-pose humanoid robots.

The human body is well suitedfor acting in our everyday environ-ments. Stairs, door handles, tools,and so on are designed to be usedby humans.

The new applications will re-quire social interaction betweenhumans and robots. If a robotis able to analyze and synthesizespeech, eye movements, mimics,gestures, and body language, itwill be capable of intuitive com-munication with humans.

A human-like action repertoirealso facilitates the programmingof the robots by demonstrationand the learning of new skills byimitation.

Last, but not least, humanoidrobots are used as a tool to under-stand human intelligence.

Addressing all of the above as-pects simultaneously exceeds thecurrent state of the art. Today’shumanoid robots display their ca-pabilities in tasks requiring a lim-ited subset of skills.

One of these tasks is play-ing soccer. Kitano and Asada

proposed the RoboCup humanoidchallenge [10] as the millenniumchallenge for advanced robotics.It is to construct by 2050 a team offully autonomous humanoid robotsoccer players that is able to win asoccer game against the winner ofthe most recent FIFA World Cup.

To facilitate research towardsthis long-term goal, the RoboCupFederation organizes since 1997annual robot competitions. Amajor part of these competitionsis RoboCupSoccer, which is con-ducted in five leagues, focusing ondifferent research aspects. Whileteam play and learning are ma-jor topics in the simulation league,the real-robot leagues developedsolutions for robust real-time per-ception, omnidirectional locomo-tion, and ball handling.

2 RoboCup Huma-

noid League

Motivated by the rapid progress inthe wheeled and four-legged soc-cer leagues, the RoboCup Feder-ation added in 2002 a league forhumanoid robots to the competi-tions. Among the research chal-lenges addressed in this leagueis maintaining dynamic stabilityof the robots while walking, run-ning, and kicking. Another re-search issue is the coordinationof bipedal locomotion and percep-tion. FIRA established a huma-noid league (HuroSot) as well.

Three international competi-tions took place in the RoboCupHumanoid League so far. Becausethe complex humanoid robotswere not ready to play real soccergames, the robots had to demon-strate their capabilities by solvinga number of subtasks.

In the Humanoid Walk theyhad to walk towards a pole,around it and to come back tothe start. Scoring was based onwalking speed and stability. In

the Penalty Kick competition therobots faced each other. Whileone robot tried to score a goal, theother defended. In the Freestylecompetition, the robots had fiveminutes to show a performance toa jury.

Each year, there was also anew technical challenge. In 2004,it consisted of an obstacle walk, apassing task, and balancing acrossa sloped ramp.

The RoboCup HumanoidLeague competition rules [12] re-quire the robots to have a human-like body plan. They must consistof a trunk, two legs, two arms, anda head. The only allowed modeof locomotion is bipedal walking.Initially, external power supply,external computing power, remotecontrol and the use of commercialrobot platforms were discouragedby performance factors. Thesefactors were applied to trial timesand goal counts. Now, the robotsmust be fully autonomous. Noexternal power, computing, orremote control is allowed. Wire-less communication can be usedfor team coordination and gamecontrol only.

Figure 1

Fig. 1 shows some of therobots which took part in theRoboCup Humanoid League com-petitions. In 2002 [1], the Nagararobot was the overall winner. AHonda Asimo prototype of teamHITS Firstep [9] won in 2003 [16].Team Osaka won the 2004 com-petition [11] with the robot Vi-siON [18].

In the remainder of the paper,I will review the hardware andsoftware used by the teams whichparticipated in the RoboCup Hu-manoid League. I will also discussthe development of the rules andfuture research challenges.

3

3 Robot Hardware

As can be seen in Fig. 1, a widevariety of robots participated inthe Humanoid League competi-tions. The robots were grouped toclasses based on their body height.In 2004 three classes were used:H40 (<44cm), H80 (<88cm), andH120 (<180cm). For the 2005competition, the classes are re-duced to KidSize (<60cm) andMediumSize (<180cm).

3.1 Base

Depending on the availability ofcommercial platforms and theirresearch focus, the participatingteams choose to use a commercialrobot base, a construction kit, ora self-developed robot.

For example, the team Sen-chans employed in 2004 twoFujitsu Hoap-2 robots (50cm,7kg, 25DOF, ≈50,000EUR) andthe Vstone Robovie-M construc-tion kit (29cm, 1.9kg, 22DOF,≈3,000EUR). In 2003, the teamHITS Firstep used a prototype ofthe Asimo, developed by Honda(125cm, 50kg, 26DOF, not avail-able commercially). In contrast,the RoboSapien (40cm, 2.5kg,7DOF, ≈100EUR), which was de-veloped for the toy market, wasused by the team NimbRo in 2004.

Examples for self-constructedrobots are Robo-Erectus [19],Rope [14], Tao-Pie-Pie [2], Per-sia [13], Isaac [8], and Alpha [7].These robots have between 7DOFand 22DOF, weigh between 2kgand 30kg and have a size be-tween 30cm and 155cm. Whilethe smaller robots are driven byRC servos, the larger ones aredriven by geared DC motors.

The small servo-driven robotswere more numerous than thelarger robots and generally per-formed better. This mightbe due to the optimization ofweight/torque ratio and controlthat went into RC servos. These

small, lightweight, and powerfulintelligent actuators were devel-oped for model airplanes, cars,and boats.

In contrast, the used DCmotor-gear-controller combina-tions have a lower degree of inte-gration and are less optimized forweight/torque ratio. In addition,the backlash of planetary gearsmakes their control difficult. Onlythe harmonic-drive gears used inexpansive robots like Asimo avoidthis problem.

Weight reduction is not onlyimportant for the actuators, butfor all other robot parts aswell. Consequently, most teamsused lightweight materials likealuminum or reinforced compos-ites to construct the robot skele-tons. Furthermore, the construc-tion tried to optimize the stiff-ness/weight ratio. One impor-tant feature of many robots wasthe addition of a second bearingpoint for servo-driven hinge joints,which improves stiffness. Anothercommon feature was the carving-out of material not needed for sta-bility, in order to reduce weight.

While most robots used an ex-ternal skeleton that also servedas cover, the VisiON robot andRoboSapien used a plastic coverin addition to an internal skele-ton. The cover protects delicateparts, like electronic boards, andimproves the robot appearance.As the degree of physical robot in-teraction will increase, covers willbe more important in the future.

3.2 Sensors

In order to perceive themselvesand their environment, the robotsneed sensors. For proprioception,joint angle sensors, such as po-tentiometers and encoders wereused. Accelerometers and gyroswere used by some robots to es-timate attitude. Some robots alsohad force sensors on their feet.

To perceive their color-coded

environment, the robots wereequipped with color cameras,placed on top of the trunk. Dif-ferent approaches were used tocover a wide field-of-view. Someteams used moving cameras (e.g.Rope), some teams used wide-angle lenses (e.g. Senchans), andthe VisiON robot used an omni-directional mirror. In addition tocameras, some teams also used IRdistance sensors (e.g. attached tothe feet).

3.3 Computing

To interface the sensors, processtheir readings, make behavior de-cisions, and control the motors,some electronics and computingpower is needed.

To interface the sensors andthe actuators, microcontroller-based electronics boards wereused. They include specializedhardware modules for A/D con-version and pulse generation.

In addition to the microcon-trollers, many robots also in-cluded a more powerful computer.Many of these were based onthe PC/104 standard for indus-trial computers. Few teams usedoff-the-shelf Pocket PCs (up to400MHz XScale [3]) or small PCs(up to 1.7GHz Pentium M) to con-trol their robots. The more pow-erfully processors allow for real-time image processing. As the de-gree of autonomy of the robotsincreases, the importance of on-board computing power will in-crease.

So far, some robots relied onexternal computing power to pro-cess video transmitted via an ana-logue wireless link. Many robotsincluded a digital wireless link(WLAN or Bluetooth) to trans-mit debug information and to al-low for remote control.

Almost all of the robots werepowered by rechargeable batter-ies. Here, the capacity/weightratio and the ability to deliver

4

high currents are important pa-rameters. It seems that Lithium-polymer batteries are currentlythe best choice for the power sup-ply. In order to save weight, mostteams used small batteries thatlasted only for a few minutes ofoperation.

4 Software

Making a humanoid robot playsoccer is a non-trivial tasks thatrequires software for perceptionand behavior control.

4.1 Perception

On the perception side, two maintasks can be distinguished: pro-prioception and computer vision.

For the servo-based robotsmost of the proprioception is doneusing potentiometers that mea-sure the joint angles. In somerobots, the potentiometer volt-ages are also read by a microcon-troller, in order to keep track ofthe actual joint angles. Of course,when implementing control forDC-motor based joints, sensorsfor position, speed or torque areread by the microcontrollers, andpreprocessed.

Similarly, force sensors, ac-celerometers, gyroscopes, and dis-tance sensors are read by themicrocontrollers and preprocessedthere. One particularly interest-ing preprocessing step is the fu-sion of accelerometer readings andmeasured rotational speeds in or-der to estimate the attitude ofa robot. This estimate can beimportant for walking on slopedsurfaces. Relevant for feedback-based balance is also the aggre-gation of multiple force sensorsplaced in a foot sole to computethe center of pressure (COP).

Many robots also monitor thebattery voltage in order to signalthe need for a fresh battery.

For the perception of the en-vironment, the images capturedby the onboard color camerasmust be analyzed. As in otherRoboCupSoccer leagues, the com-puter vision heavily relies on colorsegmentation. Individual pixelsare classified to belong to color-classes (e.g. orange for the ball,green for the field, blue and yel-low for the goals). The classifiedpixels are aggregated to blobs andrelative coordinates (angle anddistance) are estimated for the keyobjects.

Most teams used some sort ofactive perception, such as turningthe robot towards the object of in-terest, moving the camera or tilt-ing the upper body in order to seethe ball.

Visual perception in the hu-manoid league is more diffi-cult than in other RoboCupSoc-cer leagues, because the cameramoves significantly. Due to inac-curacies in posture estimation andodometry, the camera pose is notknown well.

4.2 Behavior Control

Based on the information ex-tracted from the sensors, therobots must decide how to act.Usually, this decision is done ondifferent abstraction levels.

The lowest of these levels is theindividual joint, e.g. the left knee.The servos used by most robotsinclude electronics that generatesPWM signals for the motor in or-der to quickly reach the target an-gle, which is given as a pulse train.While the motor-control of theseintelligent actuators is usually notchanged by the teams, such con-trol must be implemented on mi-crocontrollers when using combi-nations of DC motors, joint sen-sors, and motor drivers.

The next abstraction level,used by only some of the robots,is the level of a body part, e.g. theleft leg. Here, multiple joints are

controlled in a coordinated way, inorder to reach a target leg length,a leg angle, or a foot-plate - hip-plate angle. The use of these moreabstract actuators simplifies thegeneration of gait patterns.

On the level of the entirerobot, trajectories are gener-ated for body-parts or individ-ual joints. Many teams im-plemented parameterizable mo-tion primitives, like walk forward,turn, lie down, and kick. Here, itis important to pay attention tothe stability of the robot. Dueto the large feet of most robots,feed-forward control is frequentlysufficient to ensure the balanceof the robots. Stable trajectoriesare generated off-line and chainedduring operation. While initiallythe motion primitives were trig-gered manually, the rules requirenow fully autonomous behavior.

The use of preprogrammedmotion sequences is also discour-aged by rule changes that favoradaptive behavior. For instance,in the 2004 Obstacle Walk the po-sition of two obstacles was notknown in advance. In 2005, theball position for penalty kicks isonly roughly defined.

Because chaining of the mo-tion primitives requires interme-diate stops, some teams imple-mented online-trajectory genera-tion that can be used for om-nidirectional walking. Here, atarget vector defines speed anddirection of the robot motionas well as rotation around thevertical axis. Such a versatilegait is extremely helpful whenapproaching a ball and whenavoiding obstacles, without loos-ing the robot’s heading direc-tion. This has been demonstratedin the wheeled and four-leggedRoboCupSoccer leagues, whereomnidirectional drives and omni-directional gaits are used by manyteams.

5

Figure 2

One of highlights of RoboCup2004 was the goalie behavior ofTeam Osaka. As can be seenin Fig. 2, their VisiON robotjumped to the ground to defendagainst a shot. Afterwards, it wasable to get up again. Anotherhighlight of last year’s competi-tion was the passing demonstra-tion given by two Hoap-2 robotsof team Senchans [15], shown inFig. 3. They used learned visuo-motor mappings to implement theskills of trapping, approaching to,and kicking a ball.

Figure 3

In preparation for the com-petition to take place in Osaka(July 2005), the German Open(April 2005) included for the firsttime activities for humanoid soc-cer robots (see Fig. 4).

Penalty kick demonstrationswere shown by the DarmstadtDribblers (with Mr. DD and aKondo KHR-1) and team NimbRo(with Toni [6] and Kondo [4]).The NimbRo robots were able toapproach the ball using an om-nidirectional walk. The robotsslowed down in the vicinity of theball. When the ball was in front oftheir kicking foot and they facedthe goal, they performed a strongkick. If the ball did not cross thegoal line, they approached it againand kicked a second time.

Figure 4

The other humanoid leagueactivity at German Open 2005was soccer games between Brain-stormers Osnabruck and teamNimbRo. Both teams used upto four augmented RoboSapienrobots. These were the firsthumanoid robot soccer games.When playing soccer with mul-tiple robots per team, coordina-tion of them becomes relevant.Many approaches developed inother RoboCupSoccer leagues canbe applied to the humanoid leagueas well.

For example, team NimbRo [5]used an off-the field computer tofuse ball observations of individ-ual robots to a global ball esti-mate. This was communicated tothe robots to be used in case theball was outside their field of view.The external computer also com-puted team behavior, which as-signed roles like goalie, primaryattacker, and secondary attackerto the individual players. Thiscomputer was as well used forgame control (start, stop, andkickoff) and to log all importantvariables.

Ball fusion and team behaviorsrelied on localization. To simplifythe localization task, six uniquelyidentifiable color markers wereplaced around the field as land-marks. Team NimbRo used a 3DMarkov grid to integrate observa-tions of markers and goals as wellas robot motion commands overtime. Computer vision, proba-bilistic localization, behavior con-trol, and wireless communicationwere implemented on a PocketPC [4] that replaced the originalhead of RoboSapien.

5 Conclusions and

Outlook

Despite impressive achievementsof some teams, the overall perfor-mance of the soccer playing hu-manoids is still far from perfect.Basic soccer skills, such as ro-bust dynamic walking and kickingwithout loosing balance are notpossessed by all robots.

Even the best robots some-times show instability while walk-ing, fail to kick the ball or de-fend against shots not taken.Consequently, further research isneeded. Within the Huma-noid League, the performance ofsmaller, servo-driven robots ingeneral exceeded the performanceof larger robots. The only con-vincing larger robot so far was theHonda Asimo prototype, out ofreach for almost all researchers.

On the other hand, the avail-ability of low-cost robot bases,like RoboSapien, and constructionkits, like Kondo KHR-1, makesit possible to enter the huma-noid league competitions withoutthe need for huge resources. Ofcourse, the performance of suchstandardized platforms might notbe sufficient to win the competi-tions.

As the performance of the hu-manoid soccer robots improves,the rules are changed to makethe task harder. For instance,the allowed foot size decreases,which makes maintaining balancemore difficult. Improving robust-ness and speed of dynamic bipedalwalking will remain one of the ma-jor research issues in the Huma-noid League. To correct for dis-turbances, it will be necessary toincorporate more feedback. Firstattempts have been made, e.g. byBaltes, McGrath, and John An-derson [2], who use measured an-gular velocities to detect the dan-ger of falling and modify the gaitpattern to prevent the fall.

6

The Humanoid League is mov-ing away from isolated tasks per-formed by individual robots to-wards fully autonomous robotsthat cooperate in a team. For Os-aka 2005, 2 vs. 2 soccer gamesare planned in the KidSize class.Playing soccer games will posenew challenges for the robots.They have to interact withoutdamaging each other. The robotsshould be able to maintain bal-ance, even when pushed by an-other robot. When a fall cannotbe avoided, the robots must mini-mize the impact using a protectivepose. Afterwards, they should beable to stand up again. Someof these issues have been inves-tigated in the Robo-One compe-tition [17], where remotely con-trolled humanoid robots engage inmartial arts.

Playing soccer games alsoraises the bar for perception. Asin other leagues, the robots mustlocalize on the field and perceiveother players. Finally, the tworobots of a team must be coordi-nated.

The task of humanoid socceris a complex one and the devel-opment just started. As demon-strated by other RoboCupSoccerleagues, I expect quick progressin robot hardware and softwarefor perception and behavior con-trol. One of the biggest challengeswill be the integration of subsys-tems. While it is not that difficultto build a vision system or imple-ment walking, it is hard to inte-grate all components necessary toplay soccer onboard a humanoidrobot, with both high reliabilityand secure recovery procedures inthe case of a subsystem failure. Toachieve this, the use of new mate-rials, intelligent actuators, minia-turized sensors, and mobile com-puting power will be necessary.

References

[1] M. Asada, O. Obst,D. Polani, B. Browning,A. Bonarini, M. Fujita,T. Christaller, T. Taka-hashi, S. Tadokoro, E. Sklar,and G.A. Kaminka. Anoverview of RoboCup-2002Fukuoka/Busan. AI Maga-zine, 24(2):21–40, 2003.

[2] J. Baltes, S. McGrath, andJ. Anderson. The useof gyroscope feedback inthe control of the walk-ing gaits for a small huma-noid robot. In D. Nardi,M. Riedmiller, C. Sammut,and J. Santos-Victor, editors,RoboCup 2004: Robot Soc-cer World Cup VIII, volume3276 of LNCS, pages 628–635. Springer, 2005.

[3] S. Behnke, T. Langner,J. Muller, H. Neub, andM. Schreiber. NimbRo RS:A low-cost autonomous hu-manoid robot for multi-agentresearch. In Proceedings ofWorkshop on Methods andTechnology for EmpiricalEvaluation of Multi-AgentSystems and Multi-robotTeams (MTEE) at the 27thGerman Conference on Arti-ficial Intelligence (KI2004),Ulm, Germany, 2004.

[4] S. Behnke, J. Muller, andM. Schreiber. Using hand-held computers to controlhumanoid robots. In Pro-ceedings of 1st InternationalConference on DextrousAutonomous Robots andHumanoids (darh2005),Yverdon-les-Bains, Switzer-land, to appear May 2005.

[5] S. Behnke, J. Muller, andM. Schreiber. Playing soc-cer with RoboSapien. InProceedings of 2005 RoboCupInternational Symposium, toappear July 2005.

[6] S. Behnke, J. Muller, andM. Schreiber. Toni: A soccerplaying humanoid robot. InProceedings of 2005 RoboCupInternational Symposium, toappear July 2005.

[7] M. Bennewitz, F. Faber,D. Joho, M. Schreiber, andS. Behnke. Enabling a huma-noid robot to interact withmultiple persons. In Pro-ceedings of 1st InternationalConference on DextrousAutonomous Robots andHumanoids (darh2005),Yverdon-les-Bains, Switzer-land, to appear May 2005.

[8] F. Davide, M. Enrico,M. Fabio, M. Lino,M. Giuseppe, N. Diego,R. Giuseppe, and T. Al-berto. The Isaac project:development of an au-tonomous biped robot.In Team descriptions forRoboCup 2004 HumanoidLeague, 2004.

[9] S. Imura and N. Maeda.Description of HITS dreamteam ’Firstep’ of Honda In-ternational Technical School(HITS). In Team descrip-tions for RoboCup 2003 Hu-manoid League, 2003.

[10] H. Kitano and M. Asada.The RoboCup humanoidchallenge as the millen-nium challenge for advancedrobotics. Advanced Robotics,13(8):723–737, 2000.

[11] P. Lima and L. Custodio.RoboCup 2004 overview. InD. Nardi, M. Riedmiller,C. Sammut, and J. Santos-Victor, editors, RoboCup2004: Robot Soccer WorldCup VIII, volume 3276 ofLNCS, pages 1–17. Springer,2005.

[12] N.M. Mayer. Humanoid KidSize League and MediumSize League rules and setup.

7

http://er04.ams.eng.osaka-u.ac.jp/humanoid webpage/humanoid.pdf.

[13] H. Nezamabadi-pour,M. Iranmanesh, M. Alavi,M.N. Mahani, S.R. Ataee,H.H. Rezaei, M. Arabnezhad,and P. Hosseini. Persia Bteam description. In Teamdescriptions for RoboCup2005 Humanoid League,2005.

[14] H.H. Nghia, S.W. Yong, C.C.Meng, and H.G. Soon. TeamROPE. In Team descriptionsfor RoboCup 2004 HumanoidLeague, 2004.

[15] M. Ogino, M. Kikuchi,J. Ooga, M. Aono, andM. Asada. Optic flow basedskill learning for a humanoidto trap, approach to, andpass a ball. In D. Nardi,M. Riedmiller, C. Sammut,and J. Santos-Victor, edi-tors, RoboCup 2004: RobotSoccer World Cup VIII,volume 3276 of LNCS, pages323–334. Springer, 2005.

[16] E. Pagello, E. Menegatti,A. Bredenfel, P. Costa,T. Christaller, A. Jacoff,D. Polani, M. Riedmiller,A. Saffiotti, E. Sklar, and

T. Tomoichi. RoboCup-2003:new scientific and technicaladvances. AI Magazine,25(2):81–98, 2004.

[17] ROBO-ONE. Biped RobotEntertainment. http://www.robo-one.com.

[18] Team Osaka. VisiON. http://www.sansokan.jp/robot/info/vision en.html.

[19] C. Zhou and P.K. Yue. Robo-Erectus: A low-cost au-tonomous humanoid soccerrobot. Advanced Robotics,18(7):717–720, 2004.

8

Fukuoka 2002:

Priscilla Nagara Hoap-1 Robo-Erectus Footprints

Padova 2003:

HITS Firstep Senchans Robo-Erectus Foot-Prints Isaac Tao-Pie-Pie

Lisbon 2004:

Alpha of Darmstadt Rope, Senchans A, Senchans B, Persia,Team NimbRo Dribblers VisiON, NimbRo RS, VisiON

Figure 1: Some of the robots which took part in the RoboCup Humanoid League competitions.

VisiON vs. Robo-Erectus Jumping Down Getting Up

Figure 2: RoboCup 2004 Penalty Kick final.

Passing Receiving

Figure 3: RoboCup 2004 passing demonstration by Senchans.

9

RoboSapien soccer game: Brainstormers vs. NimbRo

Mr. DD (Darmstadt Dribblers) vs. Toni (NimbRo)

Kondo KHR-1 robots of Darmstadt Dribblers and NimbRo

Figure 4: RoboCup German Open 2005 Humanoid League demonstrations.

10