AI Magazine Volume 23 Number 1 (2002) (© AAAI) Articles RoboCup-2001

The Fifth Robotic Soccer World Championships

Manuela Veloso, Tucker Balch, Peter Stone, Hiroaki Kitano, Fuminori Yamasaki, Ken Endo, Minoru Asada, M. Jamzad, B. S. Sadjad, V. S. Mirrokni, M. Kazemi, H. Chitsaz, A. Heydarnoori, M. T. Hajiaghai, and E. Chiniforooshan

■ RoboCup-2001 was the Fifth International RoboCup has truly been a research-oriented RoboCup Competition and Conference. It was endeavor. Every year, the RoboCup researchers held for the first time in the , follow- analyze the progress of the research and ing RoboCup-2000 in , ; extend the competitions and demonstrations RoboCup-99 in ; RoboCup-98 in ; in the different leagues to create new chal- and RoboCup-97 in . This article discusses in detail each one of the events at RoboCup-2001, lenges. The ultimate goal of RoboCup is to focusing on the competition leagues. reach a point where teams of can suc- cessfully compete with human players. The RoboCup events move toward this goal. This article discusses in detail each one of the events at RoboCup-2001, focusing on the oboCup-2001 was the Fifth International competition leagues. As an overview of the RoboCup Competition and Conference complete RoboCup-2001 (table 1 lists all the (figure 1). It was held for the first time in R teams), and as an introduction to this article, the United States, following RoboCup-2000 in we first provide a short description of the Melbourne, Australia; RoboCup-99 in Stock- RoboCup-2001 events. The general chair of holm; RoboCup-98 in Paris; and RoboCup-97 in Osaka. RoboCup is a research-oriented initia- RoboCup-2001 was Manuela Veloso. The asso- tive that pioneered the field of multirobot ciate chairs in charge of robotic and simulation research of teams starting in 1996. In events, respectively, were Tucker Balch and those days, most of the research was Peter Stone. focused on single-robot issues. RoboCup International symposium: This was a two- opened a new horizon for multirobot research: day international symposium with presenta- Teams of robots need to face other teams of tions of technical papers addressing AI and robots to accomplish specific goals. This chal- robotics research of relevance to RoboCup. lenging objective offers a broad and rich set of Twenty papers and 42 posters were successfully research and development questions, to wit the presented in perception and multiagent behav- construction of mechanically sound and robust iors. The proceedings will be published by robots, real-time effective perception algo- Springer and are edited by program chairs rithms, and dynamic behavior-based approach- Andreas Birk, Silvia Coradeschi, and Satoshi es to support teamwork. Tadokoro.

Copyright © 2002, American Association for Artificial Intelligence. All rights reserved. 0738-4602-2002 / $2.00 SPRING 2002 55 Articles

Figure 1. The RoboCup-2001 Participating People and Robots.

Two simulation leagues: These are the soc- centimeters2 in which robots needed to have cer simulator and the simulation rescue. The full on-board autonomy (table 4). The Sony soccer simulator competition consisted of legged-robot league consisted of teams of three teams of 11 fully distributed software agents. fully autonomous Sony robots. Sixteen teams The framework consists of a server that simu- participated with Sony four-legged robots. The lates the game and changes the world accord- robot rescue competition was jointly held by ing to the actions that the players want to exe- RoboCup and the American Association for cute. The RoboCup Simulation Rescue (AAAI). It was held for competition, with teams of fully distributed the first time as part of RoboCup, and it con- software agents, provided a disaster scenario in sisted of a three-story disaster scenario provid- which teams with different capabilities, for ed by the National Institute of Standards and example, firefighters, police crews, and medical Technology (NIST), where robots navigate teams, needed to conduct search and rescue for through debris to search for victims. victims of a disaster. This event was held for Humanoid robot demonstration: Robo- the first time at RoboCup-2001. Cup-2001, jointly with AAAI, held a demon- RoboCup junior outreach: The RoboCup stration of a humanoid robot. We are planning junior event hosts children 8 to 18 years of age the first humanoid game for RoboCup-2002. interested in robotic soccer. The competitions RoboCup-2001 proved to be a truly significant and demonstrations include two on two soccer contribution to the fields of AI and robotics and robot dancing. and the subareas of multiagent and multirobot Four robot leagues: These leagues are the systems. small-size robot, the middle-size robot, the Sony legged robot, and the robot rescue. The small-size consisted of Robotics Leagues teams of as many as five robotic agents of Robots competed in four leagues at RoboCup- restricted dimensions, approximately 15 cen- 2001: (1) the small-size league, (2) the middle- timeters.3 Off-board vision and computer size league, (3) the Sony legged-robot league, remote control were allowed. The middle-size and (4) robot rescue. The small-size league, robot competition consisted of teams of as chaired by Raul Rojas, involves teams of five many as four robotic agents of restricted robots that play on a field about the size of a dimensions and a surface of approximately 50 table tennis court with an orange golf ball. The

56 AI MAGAZINE Articles

Figure 2. Two Views of the RoboCup-2001 Middle-Size League.

SPRING 2002 57 Articles

robots are limited in size to at most 18 cen- must be on board the robots themselves. Teams timeters in diameter. One of the key distinc- are composed of as many as four robots. An tions between the small-size league and the orange FIFA (International Soccer Association) other leagues is that teams in the small-size size-5 ball is used. league are allowed to place cameras over the Eighteen teams participated in the middle- field to determine the locations of robotic play- size league at RoboCup-2001, which was ers and the ball. In most cases, teams feed the chaired by Pedro Lima. Three groups of six output of the overhead camera into a central teams each competed in round-robin matches, computer that determines movement com- with the best eight teams proceeding to playoff mands that are transmitted over wireless links games. The top three finishers in the middle- to the robots. However, many researchers are size league were (1) CS FREIBURG, (2) TRACKIES, and interested in the challenge of developing (3) EIGEN. small-size robots with onboard sensing only; The Sony legged robots compete on a 3- the number of teams in this category has been meter by 5-meter carpeted field. Six colored growing each year. landmarks are placed around the field to help The small-size field has evolved substantially the robots determine their location. A small in the last few years. Originally, the field was plastic orange ball is used for scoring. Like the defined as a ping-pong table surrounded by 10- middle-size league, the Sony legged robots are centimeter-tall vertical walls. However, it was limited to on-board sensing (including a color felt that more “finesse” would be achieved in camera). All teams must use identical robots ball handling if the walls were angled; so, in provided by Sony. In 2001, teams were com- 2000 the walls were set at a 45-degree angle and posed of three robots each; in 2002, the teams shortened to 5 centimeters, where the ball is will include four robots. The Sony legged-robot likely to roll out of bounds if it is not handled soccer league has been expanding each year to carefully. Another evolution toward more real- include new teams. RoboCup-2001 included istic play was the addition of “artificial turf” on 16 teams from around the world. the field (actually a short green carpet). The Sony legged league was chaired by The year 2001 marked the first time that Masahiro Fujita. As in the other robot leagues, more teams wanted to attend than could be the competition was conducted in round-robin accommodated at the competition. Space and and playoff stages. For the round robin, teams time limited the organizers to approximately were organized into four groups of four teams. 20 teams in each league. Teams were required Eight teams progressed to the playoffs. The top to submit technical descriptions and video- three finishers in the Sony legged-robot league tapes of their teams to qualify. In the case of were (1) UNSW UNITED’01, (2) CM-PACK’01, and (3) the small-size league, 22 teams were invited, UPENNALIZERS’01. and 20 eventually made the trip to . The year 2001 marked the first year The competition was conducted as follows: RoboCup included a robot rescue event (figure Teams were divided into four groups of five 3). The event was jointly organized by teams each. The composition of the groups RoboCup and AAAI and chaired by Holly Yan- depended on a number of factors, including co. In this competition, robots explored a sim- past performance and country-continent of ulated postearthquake environment for origin. Within each group, a full round-robin trapped or injured human victims. Seven competition was held (each team played every teams participated in this event. No team did other team in the group). At the end of the well enough to place, but two technical awards round-robin phase, the top two teams in each were given. Swarthmore College was given the group were allowed to proceed to the playoffs. Technical Award for Artificial Intelligence for The small-size teams that reached the playoffs Rescue, and Sharif University received the were FU-FIGHTERS, LUCKY STAR II, KU-BOXES,ROGI Technical Award for Advanced Mobility for TEAM, CORNELL BIG RED, 5DPO, ROBOROOS,and the Rescue. We expect this league to grow substan- FIELD RANGERS. Quarter finals, semifinals, and tially in the next few years. The robot rescue finals were held in a single elimination tourna- competition is described in more detail in the ment, with an additional match to determine companion articles in this issue. third place. The top finishers were (1) LUCKY STAR II, (2) FIELD RANGERS, and (3) CORNELL BIG RED. Simulation Leagues Middle-size–league teams play on carpeted fields 5 meters wide by 9 meters long (figure 2). RoboCup-2001 featured the fifth RoboCup soc- The robots are limited to 50 centimeters in cer simulation competition and introduced the diameter. Unlike the small-size league, no first RoboCup rescue simulation competition. external sensing is allowed, and all sensors Both simulation platforms aim to capture

58 AI MAGAZINE Articles

Figure 3. The RoboCup Rescue Robot League. many of the challenges of the robotic leagues, tor (Noda et al. 1998) is an evolving research without requiring participants to build physi- platform that has been used as the basis for cal robots. Like in the real world, simulator successful international competitions and agents must deal with large amounts of uncer- research challenges (Kitano et al. 1997). It is a tainty and both perceptual and actuator noise. fully distributed, multiagent domain with Although the challenges of computer vision both teammates and adversaries. There is hid- and mechanical design are abstracted away, den state, meaning that each agent has only a simulator teams consist of greater numbers of partial world view at any given moment. The agents than do their robotic counterparts and, agents also have noisy sensors and actuators, thus, must address more large-scale multiagent meaning that they do not perceive the world issues. The ability to execute many more test exactly as it is, nor can they affect the world runs in simulation than is possible with real exactly as intended. In addition, the percep- robots also enables a larger range of possible tion and action cycles are asynchronous, pro- approaches to agent control, including learn- hibiting the traditional AI paradigm of using ing-based methods. perceptual input to trigger actions. Communi- cation opportunities are limited, and the Soccer Simulation agents must make their decisions in real time. The soccer simulator competition, chaired These domain characteristics combine to this year by Gal Kaminka, continues to be the make simulated robotic soccer a realistic and most popular RoboCup event from the per- challenging domain. Each year, small changes spective of the number of entrants (figure 4). are made to the simulator both to introduce More than 50 teams met the qualification new research challenges and to “level the requirements, 42 of which actually entered playing field” for new teams. This year, the the competition. The RoboCup soccer simula- biggest changes were the introduction of het-

SPRING 2002 59 Articles

RoboCup-2001 Scientific Challenge Award Energy-Efficient Walking for a Low-Cost Humanoid Robot, PINO

walking motions are decided by the rela- tionship between a gravity potential effect and structural parameters of the robots. Thus, there is no control over walking Link4 Link4 behaviors such as speed and dynamic l4 change in step size. θ4 In the biped walking method, we started leg body with the hypothesis that the walker can Link2

change the walking speed without chang- l2 ˘2 Link2 ing the step length if the moment of inertia of the swing leg at the hip joint has ade- l3 l1 Link3 Link3 Link1 quately been changed. We designed a con- θ3 trol method using the moment of inertia of θ the swing leg at the hip joint. The method 1 was applied to the torso of the PINO model in computational simulations and con- firmed that the method enables stable Figure A. PINO. walking with limited torque. Figure B. Planar Four-Link Model of the Robot. Left: Whole view. Right: Mechanism. A cycle of biped walking can be subdi- vided into several phases: (1) two-leg sup- planar walker. In this case, the inverted The RoboCup humanoid league, which is porting, (2) one-leg supporting, and (3) pendulum represents the supporting leg, scheduled to start in 2002, is one of the landing. Both legs are grounded in the two- and the two DOF pendulum represents the most attractive research targets. We believe leg supporting phase and landing phase, swing leg. The inverted pendulum model is that the success of the humanoid league is whereas only one leg is grounded in the the most energy-efficient model of the sup- critical for the future of RoboCup and will one-leg supporting phase. In conventional porting leg. have major implications in robotics biped walking algorithms, knees are always Figure B shows the four-link model with research and industry. Building humanoid bent so that motors are continuously high- torso. This model consists of link1, link2, robots that compete at RoboCup requires ly loaded. This approach is very different link3, and link4; link1 has a joint with the sophistication in various aspects, including from normal human walking postures. It ground. We define every joint angle θ1, θ2, mechanical design, control, and high-level should be noted that most of the current θ3, θ4 as an absolute angle of link1, link2, cognition. control methods for humanoid walking are link3, and link4, respectively. We assume PINO is a low-cost humanoid platform designed independently of the structural that every joint has a viscosity coefficient composed of low-torque servomotors and properties of the robot hardware. In gener- of 0.01 [N ⋅ m ⋅ s/ rad] and that the knee low-precision mechanical structures. It has al, these control methods require extreme- joint also has a knee stopper. Each link has been developed as a humanoid platform ly large torque to realize desired walking uniformly distributed mass m1, m2, m3, and that can widely be used by many RoboCup patterns. Although knees are bent when m4, respectively. Table A shows the link researchers in the world. Figure A shows walking on uneven terrain or major parameters of the four-link model that are the whole view and the mechanical archi- weights are loaded, the legs are stretched obtained from the real PINO. tecture of PINO. straight when walking on a flat floor. This m [kg] 0.718 l [m] 0.2785 It is intentionally designed to have low- posture can easily be modeled by inverted 1 1 m [kg] 0.274 l [m] 0.1060 torque motors and low-precision mechani- pendulum, which is known to be energy 2 2 m3 [kg] 0.444 l3 [m] 0.1725 cal structures because such motors and efficient. In addition, movement of the tor- m [kg] 3.100 l [m] 0.4515 mechanical structures significantly reduce so affects the overall moment of inertia 4 4 production cost. Although many huma- and, thus, affects energy efficiency. Our noid robots use high-performance motor goal is to mimic human walking posture to Table A. Link Parameters. systems to attain stable walking, such minimize energy through a combination of Given the control method to verify motor systems tend to be expensive. an inverted pendulum controlled by a these hypotheses (Yamasaki et al. 2001), Motors that are affordable for many swing leg and feedback control of torso researchers have only limited torque and parameter spaces were searched to identify movement. an optimal parameter set. Optimal solu- accuracy. Development of a method that The basic idea behind the low-energy allows biped walking using low-cost com- tions were found for three cases: (1) torso walking method is to consider legs of movement is controlled by feedback from ponents would have a major impact on the humanoid robots, during the one-leg sup- research community as well as industry. In body and leg movement, (2) torso is fixed porting phase, as a combination of an vertically, and (3) the three-link model the past, many researchers have studied a inverted pendulum model and a two– simple planar walker without any control without torso is compared with the four- degree-of-freedom (DOF) pendulum mod- link model with torso. torque (McGeer 1990). In such methods, el, assuming the structure of PINO to be a

erogeneous players and the introduction of a lator for the first time this year in version 7.0 of standardized coach language. the simulator.1 In previous versions, teams Heterogeneous Players Heterogeneous could consist of players with different behav- players were introduced to the RoboCup simu- iors, but their physical parameters, such as size,

60 AI MAGAZINE Articles

Left: Figure C. Result of the Foot Gait of Case 1. Middle: Figure D. Result of the Foot Gait of Case 2. Right: Figure E. Result of the Foot Gait of Case 3.

Figures C, D, and E show the foot trajec- tory for each case.. Table. . B shows. the initial θ θ θ θ angular velocity 1, 2, 3, 4 time to touch V. alue Case 1 Case 2 Case 3 down t and energy consumption. From θ 2 .1 [rad/sec] 1.736 0.962 3.374 table B, t of the four-link model with torso θ 2 .2 [rad/sec] 1.692 0.223 1.384 is longer than that of the three-link model θ .3 [rad/sec] 0.000 0.000 0.000 without torso t , and energy consumption θ 2 4 [rad/sec] 1.309 — — of case 1 is smaller than that of case 2, t2 [sec] 0.319 0.406 0.296 although every angular speed is larger. Energy consumption [J] 0.064 0.109 0.025 From these results, we can verify that the walking motion with appropriate swings of the torso enables the robot to walk with Table B. Results of Three Cases. lower energy consumption. We chose the moment of inertia of the swing leg at the hip joint, and we applied can be altered when whole body motion is Acknowledgments τ ϕ appropriately used. This is an important feedback torque leg = –kleg to the hip joint. The dynamic simulation was supported by As a result, in the lower-limb model of PINO, insight toward achieving practical human- Masaki Ogino. The authors thank him and the maximum torque required was reduced oid robots for low-cost production as well members of the Asada Laboratory at Osaka to the range of approximately 0.2 [N ⋅ m] as high-end humanoid seeking for University. (at k = 0.13) to 0.35 [N ⋅ m] (at k = 0.22). ultra–high performance using whole-body This enables the low-cost humanoid PINO to movement. perform reasonably stable biped walking. OPENPINO — Fuminori Yamasaki Further, in the four-link model with tor- All technical information on PINO is now Ken Endo so, it was verified that appropriate swings available under GNU General Public Minoru Asada of the torso enable the robot to walk with License and GNU Free Document License Hioraki Kitano lower energy consumption, as low as 0.064 as OPENPINO (exterior design and trade- [J]. marks are not subjects of GNU license). It is In this study, we observed the interest- intended to be an entry-level research plat- ing relationship between the control para- form for possible collective efforts to fur- meters and the walking behaviors, but ther develop humanoid robots for addi- understanding the details of the mecha- tional research. Authors expect the LINUX- nism that realize such behaviors is our like community is building around OPEN- future work. This study demonstrates that PINO. the energy efficiency of humanoid walking

speed, and stamina, were all identical. This six randomly generated players. At start-up, year, teams could choose from among players teams were configured with all default players. with different physical characteristics. In par- However, the autonomous online coach could ticular, in any given game, each team was able substitute in the randomly generated players to select from identical pools of players, includ- for any player other than the goalie. The only ing the default player type from years past and restriction is that each random player type

SPRING 2002 61 Articles

Figure 4. The RoboCup-2001 Simulation Leagues.

could be assigned to, at most, three teammates. players, with some observers commenting on The random players are generated by the their speedy players that they positioned on simulator at start-up time by adjusting five the sides of the field. parameters, each representing a trade-off in Standardized Coach Language Past player abilities: (1) maximum speed versus sta- RoboCup simulator competitions have allowed mina recovery, (2) speed verus turning ability, teams to use an omniscient autonomous coach (3) acceleration versus size, (4) leg length ver- agent. This coach is able to see the entire field sus kick accuracy, and (5) stamina versus max- and communicate with players only when the imum acceleration. These parameterizations play is stopped (for example, after a goal or for were chosen with the goal of creating interest- a free kick). Typically, each team developed its ing research issues regarding heterogeneous own communication protocol between the teams without creating a large disadvantage for players and the coach. teams that chose to use only default players. At This year, a standardized coach language was the outset, it was not known whether using introduced with the goal of allowing a coach heterogeneous players would be advantageous. from one team to interact with the players Experimentation leading to the competition from another. The standardized language has a established that using heterogeneous players specific syntax and intended semantics. Teams could provide an advantage of at least 1.4 goals had an incentive to use this language because a game over using only the default players messages encoded in the standardized lan- (Stone 2002). guage could be communicated even when the Indeed, at least one of the top-performing ball was in play (although with some delay and teams in the competition (UvA Trilearn from frequency limit to prevent coaches from the —fourth place) “micromanaging” their players). took good advantage of the heterogeneous One offshoot of introducing a standardized

62 AI MAGAZINE Articles

Figure 5. A RoboCup Junior Game. language was that an auxiliary competition rescue simulation competition, which was could be introduced: a coach competition. In chaired by Satoshi Tadokoro. The basis of this competition, entrants provided only a RoboCup rescue is a disaster rescue scenario in coach that was paired with a previously which different types of rescue agents—fire- unknown team that is able to understand the fighters, police workers, and ambulance work- standardized coach language. Entrants were ers—attempt to minimize the damage done to judged based on how well this unknown team civilians and buildings after an earthquake. could perform against a common opponent The setting was a portion of Kobe, , the when coached by the entrant’s coach program. site of a recent devastating earthquake. Results For the second year in a row, a first- The simulator included models that cause time entrant won the RoboCup simulator com- buildings to collapse, streets to be blocked, fires petition: tsinghuaeolus from Tsinghua Univer- to spread, and traffic conditions to be affected sity in . They beat the brainstormers based on seismic intensity maps. Each partici- from Karlsruhe University in by a pant had to create rescue agents for each of the score of 1–0, scoring the lone goal of the game three types as well as one control center for each during the third overtime period. The winners type of agent (that is, a fire station, a police sta- of the inaugural coach competition were the tion, and a rescue center). The agents sense the ChaMeleons from Carnegie Mellon University world imperfectly and must react to dynamical- and Sharif-Arvand from Sharif University of ly changing conditions by moving around the Technology in . world, rescuing agents, and putting out fires according to their unique capabilities. Commu- Rescue Simulation nication among agents of different types is RoboCup-2001 hosted the inaugural RoboCup restricted to going through the control centers.

SPRING 2002 63 Articles

RoboCup-2001 Engineering Challenge Award Fast Object Detection in Middle-Size RoboCup

Fast and reliable analysis of image data is one of the key points in per- formance. To make a soccer robot act fast enough in a dynamically changing envi- ronment, we will reduce the number of sensors as much as possible and design fast software for object detection and reliable decision making. Therefore, in RoboCup, we think it is worth getting fast and almost correct results rather than slow and exact. To achieve this goal, we propose three ideas: (1) a new color model, (2) object detection by checking image color on a set of jump points in the perspective view of the robot front camera, and (3) a fast Left: Figure A. Position of Jump Points in a Perspective View of the Robot. method for detecting edge points on Right: Figure B. An Illustration of Ball Segmentation by a Surrounding Rectangle. straight lines. The other details of our robot (that is, its mechanical design, hardware 1 shows an image of the RoboCup middle- objects, such as robots, the yellow goal, control, and software) is given in Jamzad et size soccer field with certain points on it. and the blue goal. For each object, we al. (2000). The points that are displayed in perspective return its descriptive data, such as color, to the robot’s front camera are called jump size, and the coordinate of its lower-left points. They have equal spacing on each and upper-right corner of its surrounding A New Color Model perspective horizontal line. Their vertical rectangle and a point Q on the middle of its We propose a new color model named HSY spacing is related to the RoboCup soccer lower side (figure 2). Point Q is used to find (the H is from CIELab, S from HSI, and Y field size. The actual spacing between jump the distance and angle of the robot from from Y IQ color models [Sangwine and points is set in such a way that at least five this object. Horne 1998]). The reason for this selection jump points are located on a bounding box is that the component Y in Y IQ converts a surrounding the ball (which is the smallest color image into a monochrome one. object on the soccer field), no matter how Straight-Line Detection far or how close the ball is. By checking the Therefore, comparing with I in HSI, which During the match, there are many cases image color only at these jump points, is the average of R, G, and B, Y gives a bet- when the robot needs to find its distance there is a high probability that we could ter mean for measuring the pixel intensity. from walls. In addition, the goal keeper at find the ball. In our system, we obtained The component S in HSI is a good measure all times needs to know its distance from satisfactory results with 1200 jump points. for color saturation. Finally, the parameter walls and also from white straight lines in To search for the ball, we scan the jump H in CIELab is defined as follows: front of the goal. Because the traditional –1 points from the lower right point toward H = tan b*/a* edge-detection methods (Gonzalez and the upper left corner. At each jump point, where a* denotes relative redness-green- Woods 1993) are very time consuming in the HSY equivalent of the RGB values is ness, and b* shows yellowness-blueness real-time situation, we propose a very sim- obtained from a lookup table. Because we (Sangwine and Horne 1998). H is a good ple and fast method to find the edge points have defined a range of HSY for each of the measure for detecting regions matching a on straight lines as follows: standard colors in RoboCup, we can easily given color (Gong and Sakauchi 1995), As seen in figure 3, to locate points on assign a color code to this HSY value. If a which is exactly the case in RoboCup the border of the wall and the field, we jump point is red, then it is located on the where we have large regions with a single select a few points on top of the image ball. Because this jump point can be any color. (these points are on the wall) and assume a point on the ball, from this jump point, we drop of water is released at each point. If no can move toward right, left, up, and down, object is on its way, the water will drop on checking each pixel for its color. As long as Object Detection in the field, right on the border with the wall. the color of the pixel being checked is red, To implement this idea, from a start point Perspective View we are within the ball area. This search w we move downward until reaching a stops in each direction when we reach a i In the real world, we see everything in per- green point f . All candidate border edge border point that is a nonred pixel. In one i spective: Objects far away from us are seen points are passed to Haugh transform (Gon- scan of all jump points in a frame, in addi- small, and those closer up are seen larger. zalez and Woods 1993) to find the straight- tion to a red ball, we can find all other This view is true for cameras as well. Figure line equation that best fits these points.

64 AI MAGAZINE Articles

not able to respond in real-time speed. To wall overcome this processing speed problem, wn w3 w2 w1 we preferred to have a nonexact, but reli- able, solution to the vision problem. Fast object detection was achieved by checking the color of pixels at jump points and defining a rectangular shape bounding box around each detected object. To sim- f1 plify the calculations, the distance and f2 angle of an object from the robot is esti- mated to be that of this rectangle. field Although we obtained satisfactory f3 fn results from our method in real soccer robot competitions, we believe the combi- robot nation of a CCD camera in front and an omnidirectional viewing system on top of the robot can give a more reliable perfor- mance, especially for localization.

— M. Jamzad B. S. Sadjad V. S. Mirrokni Figure C. An Illustration of a Robot View. M. Kazemi Straight lines w f show the pass of water dropped from the top of the wall. i i H. Chitsaz Conclusion response from the robot vision system for A. Heydarnoori, fast decision making. Although the tradi- M. T. Hajiaghai In a dynamically changing environment tional methods of image processing for seg- E. Chiniforooshan such as RoboCup, where most objects are mentation, edge detection, and object find- moving around most of the time, we need ings are very accurate,with the processing near–real-time (25 frames a second) capabilities of PCs today, these methods are

RoboCup rescue has many things in com- RoboCup-2000, this year, RoboCup Junior mon with RoboCup soccer. It is a fully distrib- hosted 24 teams totaling nearly 100 partici- uted, multiagent domain with hidden state, pants from the local Washington area and oth- noisy sensors and actuators, and limited com- er U.S. states as well as Australia, Germany, and munication opportunities. RoboCup rescue the . introduces the challenges of scaling up to RoboCup Junior 2001, chaired by Elizabeth many more agents and coordinating multiple Sklar, included three challenges: (1) soccer, (2) hierarchically organized teams. rescue, and (3) dance. These categories are In the competition, competing agents oper- designed to introduce different areas within ate simultaneously in independent copies of the field of robotics, such as operation within the world. That is, they don’t compete against static versus dynamic environments, coordina- each other directly but, rather, compare their tion in multiplayer scenarios, and planning performance under similar circumstances. The with incomplete information. The challenges scoring metric is such that human lives saved is also emphasize both competitive and collabo- the most important measure, with injuries and rative aspects for the teams, both within the building damage serving to break ties. games and throughout preparation. In particu- Seven teams competed in the 2001 RoboCup lar, the dance challenge allows students to rescue competition. The winning team was bring creativity in terms of art and music to an YABAI from the University of Electro-Commu- event that traditionally focuses on engineer- nications in Japan. ing. Extensive feedback and analysis has been RoboCup Junior made through interviewing students and men- tors who have participated in RoboCup junior RoboCup junior (figure 5) aims to develop edu- events. This research by committee members cational methods and materials using robotics and collaborators is ongoing and involves eval- emanating from the RoboCup soccer theme. uation of the effectiveness of educational team Following on the success of activities held at robotics.

SPRING 2002 65 Articles

Figure 6. PINO,the Humanoid Robot.

Humanoid Exhibition RoboCup-2001).4 Special thanks to all the par- ticipating teams (see sidebar), without whom The humanoid exhibition, chaired by RoboCup would not exist. Thanks also to Eliz- Dominique Duhaut, had only one partici- abeth Sklar for input on the RoboCup junior pant—the PINO team—and received major atten- section. tion. PINO is a small-size (70 centimeters in height) humanoid robot that can walk and fol- Notes low the ball (figure 6). It was developed by ERA- 1. Mao Chen, Ehsan Foroughi, Fredrik Heintz, Spiros TO Kitano Symbiotic Systems Project, which is Kapetanakis, Kostas Kostiadis, Johan Kummeneje, a five-year government-founded project in Itsuki Noda, Oliver Obst, Patrick Riley, Timo Steffens, Japan. A paper describing biped walking control YiWang, and Xiang Yin. Users manual: RoboCup soc- PINO for won this year’s Scientific Challenge cer server manual for SOCCER SERVER 7.07 and later. Award for development of an energy minimum Available at http://sourceforge.net/projects/sserver/. walking method (see the sidebar). An interest- 2. All technical information on PINO is now available ing feature of PINO is that it was designed to be under GNU General public licensing as OpenPINO open platform for humanoid research and only platform PHR-001 (www.openpino.org/). 2 uses low-cost off-the-shelf components. 3. For more information, visit www.. org. 4. See www.robocup.org/games/01Seattle/315.html) Conclusion as well as the RoboCup executive committee (www.cs.cmu.edu/~robocup2001/robocup-federa- RoboCup-2001 continued the ongoing, grow- tion.html ing research initiative that is RoboCup. RoboCup-2002 will take place in June 2002 in References 3 , Japan, and Pusan, South . Gong, Y., and Sakauchi, M. 1995. Detection of Regions Matching Specified Chromatic Features. Acknowledgments Computer Vision and Image Understanding 61(2): We thank the full organizing committee of 163–269.

66 AI MAGAZINE Articles

Soccer-Simulation League WAHOO WUNDERKIND FC, University of Virginia, CMHAMMERHEADS’01, Carnegie Mellon University, USA, David Evans USA, Tucker Balch 11MONKEYS3, Keio University, Japan, Keisuke WRIGHTEAGLE2001, University of Science and COPS STUTTGART, , Ger- Suzuki Technology of China, P. R. China, Chen XiaoP- many, Reinhard Lafrenz ANDERLECHT, IRIDIA-ULB, Belgium, Luc Berre- ing CS FREIBURG, , Germany, waerts YOWAI2001, University of Electro-Communica- Bernhard Nebel AT HUMBOLDT 2001, Humboldt University Berlin, tions, Japan, Koji Nakayama EIGEN, Keo University, Japan, Kazuo Yoshida Germany, Joscha Bach ZENG01, Fukui University, Japan, Takuya Morishi- FUN2MAS, Politecnico di Milano, , Andrea A-TEAM, Tokyo Institute of Technology, Japan, ta Bonarini Hidehisa Akiyama, FUSION, Fukuoka University, Japan, Matsuoka ATTUNITED-2001, AT& T Labs-Research, USA, Rescue-Simulation League Takeshi Peter Stone GMD-ROBOTS, GMD-AIS, Germany, Ansgar Breden- BLUTZLUCK, University of Leuven, Belgium, Josse ARIAN, University of Technology, Iran, Habibi feld Colpaert Jafar Sharif ISOCROB, Instituto de Sistemas e Robtica, Portu- CHAMELEONS’01, Carnegie Mellon University, GEMINI-R, Tokyo Institute of Technology, Japan, gal, Pedro Lima USA, Paul Carpenter Masayuki Ohta JAYBOTS, Johns Hopkins University, USA, Darius CROCAROOS, University of Queensland, Australia, JAISTR, Japan Advanced Institute of Science and Burschka Mark Venz Technology, Japan, Shinoda Kosuke MINHO, , , António CYBEROOS2001, CSIRO, Australia, Mikhail NITRESCUE, Institute of Technology, Ribeiro Prokopenko Japan, Taku Sakushima ROBOSIX,University Pierre and Marie Curie, DIRTY DOZEN, Institute for Semantic Information RESCUE-ISI-JAIST, University of Southern California, , Ryad Benosman Processing, Germany, Timo Steffens USA, Ranjit Nair SHARIF CE, Sharif University of Technology, Iran, DRWEB (POLYTECH), State Technical University, RMIT ON FIRE, RMIT University Australia, Lin Mansour Jamzad , Sergey Akhapkin Padgham SPQR, University “La Sapienza,” Italy, Luca Iocchi ESSEX WIZARDS, University of Essex, Huosheng Hu YABAI, University of Electro-Communications, THE ULM SPARROWS, University of Ulm, Germany, FC PORTUGAL 2000, Universidades do Porto e Japan, Takeshi Morimoto Gerhard Kraetzschmar Aveiro, Portugal, Luis Paulo Reis TRACKIES, , Japan, Yasutake Taka- FC PORTUGAL 2001, Universidade de Aveiro, Portu- Small-Size Robot League hashi gal, Nuno Lau FCTRIPLETTA, Keio University, Japan, Norihiro 4 STOOGES, University of Auckland, , Kawasaki Jacky Baltes Sony Legged-Robot League FUZZYFOO, Linkopings Universitet, , 5DPO, University of Porto, Portugal, Paulo Costa ARAIIBO, University of Tokyo, Japan, Tamio Arai Mikael Brannstrom CM-DRAGONS’01, Carnegie Mellon University, USA, ASURA, Fukuoka Institute of Technology, Japan, GEMINI,Tokyo Institute of Technology, Japan, Brett Browning Takushi Tanaka Masayuki Ohta CORNELL BIG RED, Cornell University, USA, Raffael- BABYTIGERS 2001, Osaka University, Japan, Minoru GIRONA VIE, University of Girona, , lo D’Andrea Asada Muñoz FIELD RANGERS, Polytechnic, Singapore, CERBERUS, Bogazici University, , Levent HARMONY, Hokkaido University, Japan, Hidenori Hong Lian Sng Akin, and Technical Univ. of Sofia, Bulgaria, Kawamura FU FIGHTERS, Universitt Berlin, Germany, Sven Anton Topalov HELLI-RESPINA 2001, Allameh Helli High School, Behnke Freie CMPACK’01, Carnegie Mellon University, USA, Iran, Ahmad Morshedian FU-FIGHTERS-OMNI, Universitt Berlin, Germany, Manuela Veloso JAPANESE INFRASTRUCTURE TEAM, Future University- Raul Rojas Freie ESSEX ROVERS, University of Essex, UK, Huosheng Hakodate, Japan, Hitoshi Matsubara HWH-CATS,College of Technology and Commerce, Hu KARLSRUHE BRAINSTORMERS, Universitaet Karlsruhe, , R.O.C., Kuo-Yang Tu Hwa Hsia GERMAN TEAM, Humboldt University Berlin, Ger- Germany, Martin Riedmiller KU-BOXES2001, Kinki University, Japan, Harukazu many, Hans- Dieter Burkhard LAZARUS, Dalhousie University, , Anthony Igarashi LES 3 MOUSQUETAIRES, LRP, France, Pierre Blazevic Yuen LUCKY STAR II, Singapore Polytechnic, Singapore, MCGILL REDDOGS, McGill University, Canada, LIVING SYSTEMS, Living Systems, Germany, Klaus Ng Beng Kiat Jeremy Cooperstock Dorer OMNI, Osaka University, Japan, Yasuhiro Masutani ROBOMUTTS++, The University of Melbourne, Aus- LUCKY LUBECK, University of Lubeck, Germany, OWARIBITO, Chubu University, Japan, Tomoichi tralia, Nick Barnes Daniel Polani Takahashi SPQR-LEGGED, University “La Sapienza,” Italy, MAINZ ROLLING BRAINS, Johannes Gutenberg Univer- ROBOROOS 2001, University of Queensland, Aus- Daniele Nardi sity, Germany, Felix Flentge tralia, Gordon Wyeth TEAM SWEDEN, Orebro University, Sweden, Alessan- NITSTONES, Nagoya Institute of Technology, ROBOSIX UPMC-CFA, University Pierre and Marie dro Saffiotti Japan, Nobuhiro Ito Curie, France, Ryad Benosman UNSW UNITED, University of New South Wales, OULU 2001, University of Oulu, , Jarkko ROGI TEAM, University of Girona, Spain, Bianca Australia, Claude Sammut Kemppainen Innocenti Badano UPENNALIZERS, University of Pennsylvania, USA, PASARGAD, AmirKabir University of Technology, ROOBOTS, The University of Melbourne, Australia, Jim Ostrowski Iran, Ali Ajdari Rad Andrew Peel USTC WRIGHT EAGLE, University of Science and RMIT GOANNAS, RMIT, Australia, Dylan Mawhinney SHARIF CESR, Sharif University of Technology, Iran, Technology of China, P. R. China, Xiaoping ROBOLOG 2001, University of Koblenz, Germany, Mohammad T. Manzuri Chen Frieder Stolzenburg TEAM CANUCK, University of Alberta, Canada, UW HUSKIES, University of Washington, USA, SALOO, AIST/JST, Japan, Itsuki Noda Hong Zhang Dieter Fox SBCE, Shahid Beheshti University, Iran, Eslam TPOTS,Temasek Engineering School, Singapore, Nazemi Nadir Ould Khessal Robot Rescue League SHARIF-ARVAND, Sharif University of Technology, UW HUSKIES, University of Washington, USA, Dinh Iran, Jafar Habibi Bowman EDINBURGH, , UK, Daniel TEAM SIMCANUCK, University of Alberta, Canada, VIPERROOS, University of Queensland, Australia, Farinha Marc Perron Mark Chang SHARIF-CE, Sharif University of Technology, Iran, TSINGHUAEOLUS, Tsinghua University, P. R. China, Amir Jahangir Shi Li Middle-Size Robot League SWARTHMORE, Swarthmore College, USA, Gil Jones TUT-GROOVE, Toyohashi University of Technology, UTAH, Utah State University, USA, Dan Stormont AGILO ROBOCUPPERS, Munich University of Tech- Japan, Watariuchi Satoki MINNESOTA, University of Minnesota, USA, Paul nology, Germany, Michael Beetz UTUTD, University of , Iran, Amin Bagheri Rybski ARTISTI VENETI, University, Italy, Enrico Pag- UVA TRILEARN 2001, Universiteit van Amsterdam, FLORIDA, University of South Florida, USA, Robin ello The , Remco de Boer Murphy CLOCKWORK ORANGE, University of Technology, The VIRTUAL WERDER, University of , Germany, Netherlands, Pieter Jonker Delft Ubbo Visser

Table 1. RoboCup 2001 Teams.

SPRING 2002 67 Articles

Gonzalez, R. C., and Woods, R. E. 1993. Digital Image oped teams of robotic soccer agents that have been Processing. Reading, Mass.: Addison-Wesley. RoboCup world champions in three different leagues, Jamzad, M.; Foroughnassiraei, A.; Chiniforooshan, namely, simulation (1998, 1999), CMU-built small- E.; Ghorbani, R.; Kazemi, M.; Chitsaz, H.; Mobasser, wheeled robots (1997, 1998), and Sony four-legged F.; and Sadjed, S. B. 2000. Middle-Sized Robots: dog robots (1998). Veloso is vice president of the ARVAND. In RoboCup-99: Robot Soccer World Cup II, RoboCup International Federation. She was awarded 74–84. Stockholm: Springer. a National Science Foundation Career Award in 1995 and the Allen Newell Medal for Excellence in Research Kitano, H., and Asada, M. 1998. RoboCup Humanoid in 1997. Her e-mail address is [email protected]. Challenge: That’s One Small Step for a Robot, One Giant Leap for Mankind. In Proceedings of the Inter- Tucker Balch is an assistant pro- national Conference on Intelligent Robots and Sys- fessor of computing at the Georgia tems. Washington, D.C.: IEEE Computer Society. Institute of Technology. He has been involved in RoboCup since Kitano, H.; Tambe, M.; Stone, P.; Veloso, M.; Corade- 1997. Balch competed in robotic schi, S.; Osawa, E.; Matsubara, H.; Noda, I.; and Asa- and simulation leagues, chaired da, M. 1997. The RoboCup Synthetic Agent Chal- the organization of the small-size lenge ‘97. In Proceedings of the Fifteenth league in 2000, and served as asso- International Joint Conference on Artificial Intelli- ciate chair for robotic events for gence, 24–29. Menlo Park, Calif.: International Joint RoboCup-2001. Balch is also a trustee of the Conferences on Artificial Intelligence. RoboCup Federation. Balch’s research focuses on McGeer, T. 1990. Passive Dynamic Walking. Interna- coordination, communication, and sensing for mul- tional Journal of Robotics Research 9(2): 62–82. tiagent systems. He has published over 60 technical Noda, I.; Matsubara, H.; Hiraki, K.; and Frank, I. articles in AI and robotics. His book, Robot Teams, 1998. SOCCER SERVER:A Tool for Research on Multia- edited with Lynne Parker, will be published in 2002. gent Systems. Applied Artificial Intelligence 12:233– His e-mail address is [email protected]. 250. Peter Stone is a senior technical Sangwine, S. J., and Horne, R. E. N. 1998. The Colour staff member in the Artificial Intel- Image-Processing Handbook. New York: Chapman and ligence Principles Research De- Hall. partment at AT&T Labs Research. Stone, P. 2002. ATTUnited-2001: Using Heteroge- He received his Ph.D. in 1998 and neous Players. In RoboCup-2001: Robot Soccer World his M.S. in 1995 from Carnegie Cup V, eds. A. Birk, S. Coradeschi, and S. Tadokoro. Mellon University, both in com- Berlin: Springer Verlag. Forthcoming. puter science. He received his B.S. Yamasaki, F.; Endo, K.; Asada, M.; and Kitano, H. in mathematics from the Universi- 2001. A Control Method for Humanoid Biped Walk- ty of Chicago in 1993. Stone’s research interests ing with Limited Torque. Paper presented at the Fifth include planning and machine learning, particularly International Workshop on RoboCup, 7–10 August, in multiagent systems. His doctoral thesis research Seattle, Washington. contributed a flexible multiagent team structure and Yamasaki, F.; Matsui, T.; Miyashita, T.; and Kitano, H. multiagent machine learning techniques for teams operating in real-time noisy environments in the 2000. PINO, The Humanoid That Walks. In Proceed- presence of both teammates and adversaries. He is ings of the First IEEE-RAS International Conference currently continuing his investigation of multiagent on Humanoid Robots. Washington, D.C.: IEEE Com- learning at AT&T Labs. His e-mail address is puter Society. [email protected]. Yamasaki, F.; Miyashita, T.; Matsui, T.; and Kitano, H. Hiroaki Kitano is a senior re- 2000. PINO, the Humanoid: A Basic Architecture. Paper presented at the Fourth International Work- searcher at Sony Computer Sci- shop on RoboCup, 31 August–1 September, Mel- ence Laboratories, Inc.; director of bourne, Australia. ERATOKitano Symbiotic Systems Project, Japan Science and Tech- nology Corporation, a govern- Manuela Veloso is associate profes- ment organization for basic sci- sor of computer science at Carnegie ence; and president of The Mellon University (CMU). She RoboCup Federation. Kitano was a received her Ph.D. in computer sci- visiting researcher at Carnegie Mellon University ence from CMU in 1992. She from 1988 to 1993 and received a Ph.D. in computer received a B.S. in electrical engi- science from Kyoto University in 1991. Kitano neering in 1980 and an M.Sc. in received The Computers and Thought Award from electrical and computer engineer- IJCAI in 1993 and the Prix Ars Electronica in 2000. ing in 1984 from the Instituto His e-mail address is [email protected]. Superior Tecnico in . Veloso’s long-term research goal is the effective construction of teams of intelligent agents where cognition, perception, and action are combined to address planning, execution, and learning tasks, in particular, in uncertain, dynam- ic, and adversarial environments. Veloso has devel-

68 AI MAGAZINE