RoboCup-99: Robot Soccer World Cup III

RoboCup-99, the third Robot World Cup Soccer Games and Conferences, was held in conjunction with IJCAI-99 in Stockholm. Robo-Cup has now clearly demonstrated that it provides a remarkable framework for advanced research in Robotics and Artificial Intelligence. The yearly RoboCup event has included a technical workshop and competitions in different leagues. This chapter presents a comprehensive overview of RoboCup-99 and the scientific and engineering challenges presented to the participating researchers. There were four RoboCup-99 competitions: the simulation league, the small-size robot league, the middle-size robot league, and, for the first time officially, the Sony legged robot league. The champion teams were CMUnited-99 (Carnegie Mellon University, USA) for the simulation league, Sharif CE (Sharif University of Technology, Iran) for the middle-size league, Big Red (Cornell University, USA) for the small-size league, and “Les 3 Mousquetaires” (Laboratoire de Robotique de Paris, France) for the Sony legged robot league. The Scientific Challenge Award was given to three papers on innovative research for the automated statistical analysis of the games, from the University of Southern California (ISI/USC), USA, the Electrotechnical Laboratory (ETL), Japan, and Chubu University, Japan. There will be the first RoboCup European Championship in Amsterdam in May 2000, and the International RoboCup-2000 will take place in Melbourne, Australia, in August 2000.

The CMUnited-99 simulator team became the 1999 RoboCup simulator league champion by winning all 8 of its games, outscoring opponents by a combined score of 110-0. CMUnited-99 builds upon the successful CMUnited-98 implementation, but also improves upon it in many ways. This paper gives a detailed presentation of CMUnited-99’s improvements over CMUnited-98.

In this paper we describe Big Red, the Cornell University Robot Soccer team. The success of our team at the 1999 competition can be mainly attributed to three points: 1) An integrated design approach; students from mechanical engineering, electrical engineering, operations research, and computer science were involved in the project, and a rigorous and systematic design process was utilized. 2) A thorough understanding of the system dynamics, and ensuing control. 3) A high fidelity simulation environment that allowed us to quickly explore artificial intelligence and control strategies well in advance of working prototypes.

Arvand is the name of robots specially designed and constructed by sharif CE team for playing soccer according to RoboCup rules and regulations for the middle size robots. Two different types of robots are made, players and the goal keeper. A player robot consists of three main parts: mechanics (motion mechanism and kicker), hardware (image acquisition, processing unit and control unit) and software (image processing, wireless communication, motion control and decision making). The motion mechanism is based on two drive unit, two steer units and a castor wheel. We designed a special control board which uses two microcontrollers to carry out the software system decisions and transfers them to the robot mechanical parts. The software system written in C++ performs real time image processing and object recognition. Playing algorithms are based on deterministic methods. The goal keeper has a different moving mechanism, a kicker like that of player robots and a fast moving arm. Its other parts are basically the same as player robots. We have constructed 3 player robots and one goal keeper. These robots showed a high performance in Robocup-99: became champion.

This work deals with designing simple behaviors for legged robots to play soccer in a predefined environment: the soccer field. Robots are fully autonomous, they cannot exchange messages between each other. Behaviors are based on information coming from the vision sensor, the CCD camera here which allows robots to detect objects in the scene. In addition to classical vision problems such as lighting conditions and color confusion, legged robots must cope with “bouncing images” due to successive legs hitting ground. This paper presents the work achieved for our second participation. We point out the improvements of algorithms between the two participations, and those we plan to make for the next one.

Multi-agent teamwork is critical in a large number of agent applications, including training, education, virtual enterprises and collective robotics. Tools that can help humans analyze, evaluate, and understand team behaviors are becoming increasingly important as well. We have taken a step towards building such a tool by creating an automated analyst agent called ISAAC for post-hoc, off-line agent-team analysis. ISAAC’s novelty stems from a key design constraint that arises in team analysis: multiple types of models of team behavior are necessary to analyze different granularities of team events, including agent actions, interactions, and global performance. These heterogeneous team models are automatically acquired via machine learning over teams’ external behavior traces, where the specific learning techniques are tailored to the particular model learned. Additionally, ISAAC employs multiple presentation techniques that can aid human understanding of the analyses. This paper presents ISAAC’s general conceptual framework, motivating its design, as well as its concrete application in the domain of RoboCup soccer. In the RoboCup domain, ISAAC was used prior to and during the RoboCup’99 tournament, and was awarded the RoboCup scientific challenge award.

This paper describes analysis results of collaboration among players of RoboCup ’98 simulator teams and on-line adversarial model analysis using LogMonitor. LogMonitor is a tool for analyzing games from logfiles and displaying statistical data such as counts of soccer plays. Evaluation of collaboration in a multi-agent system is closely related with applied domains, which make it difficult to distinguish agent’s universal ability from task oriented programs. In viewing simulation soccer games, play agents’ skills are evaluated from the human soccer standards. This situation is assumed to be similar to collaboration among teammates, that is evaluated from human standards.Adding to the basic actions of the player such as shooting, kicking, etc., a 1-2 pass among teammate agents is used to evaluate teams in collaboration. LogMonitor data shows that 1-2 pass may be useful to evaluate collaboration. Experiments show that adding adversarial information is very useful to make a team more robust.

This paper uses statistical analysis to demonstrate the progress to date in answering the RoboCup Synthetic Agent Challenge. We analyze the complete set of log data produced by the simulator tournaments of 1997 and 1998, applying techniques such as principal component analysis to identify precisely what has improved, and what requires further work. Since the code that implements our analysis produces its results in real-time, we propose releasing a proxy server that makes statistical analysis available to RoboCup developers. We believe such a server has a crucial role to play in facilitating and evaluating research on the three specific challenge problems of opponent modeling, teamwork and learning. We also suggest that — if RoboCup is to make the most of the efforts of participating researchers — the time is ripe for the institution of a modular team based on a common model.

The quantity of image information is very large, and this is why it is difficult to process them in real time, such as video frame rate. In this paper we demonstrate the developed real-time color detection system for the vision system of RoboCup Small-Size League. The pixel’s color information is converted to HSV color system at first, and judged whether it has the color information of target. The detected information is converted to gray scale NTSC signal to be captured by PC, which is faster than the full color frame grabber.

An innovative technique for segmentation of color images is proposed. The technique implements an approach based on thresholding of the hue histogram and a feed-forward neural network that learns to recognize the hue ranges of meaningful objects. A new function for detecting valleys of the histogram has been devised and tested. A novel blurring algorithm for noise reduction that works effectively when used over hue image has been employed. The reported experimental results show that the technique is reliable and robust even in presence of changing environmental conditions. Extended experimentation has been carried on the framework of the Robot Soccer World Cup Initiative (RoboCup).

This paper describes a practical method for calibrating the geometry and colour information for cameras surveying large rooms. To calibrate the geometry, we use a semi-automatic system to assign real world to pixel coordinates. This information is the input to the Tsai camera calibration method. Our system uses a two stage process in which easily recognizable objects (squares) are used to sort the individual data points and to find missing objects. Fine object features (corners) are used in a second step to determine the object’s real world coordinates. An empirical evaluation of the system shows that the average and maximum errors are sufficiently small for our domain. Objects are recognized through coloured spots. The colour calibration uses six thresholds (Three colour ranges (Red, Green, and Blue) and three colour differences (Red - Green, Red - Blue, Green - Blue)). This paper describes a fast threshold comparison routine.

This paper investigates the use of reinforcement learning in solving the path-tracking problem for car-like robots. The reinforcement learner uses a case-based function approximator, to extend the standard reinforcement learning paradigm to handle continuous states. The learned controller performs comparable to the best traditional control functions in both simulation and also in practical driving.

This paper presents the vision system of the robot soccer team Agilo RoboCuppers — the RoboCup team of the image understanding group (FG BV) at the Technische Universität München.We present a fast and robust color classification method yielding significant regions in the image. The boundaries between adjacent regions are used to localize objects like the ball or other robots on the field. Furthermore for each player the free motion space is determined and its position and orientation on the field is estimated. All this is done completely vision based, without any additional sensors.

This paper describes the mechanical and electrical design, as well as the control strategy, of the FU-Fighters robots, a F180 league team that won the second place at RoboCup’99. It explains how we solved the computer vision and radio communication problems that arose in the course of the project.The paper mainly discusses the hierarchical control architecture used to generate the behavior of individual agents and the team. Our reactive approach is based on the Dual Dynamics framework developed by H. Jäger, in which activation dynamics determines when a behavior is allowed to influence the actuators, and a target dynamics establishes how this is done. We extended the original framework by adding a third module, the perceptual dynamics. Here, the readings of fast changing sensors are aggregated temporarily to form complex, slow changing percepts.We describe the bottom-up design of behaviors and illustrate our approach using examples from the RoboCup domain.

Versatile physical and behavioral features as well as their exploitation through computation-power onboard the robot-players are feasible and necessary goals for the RoboCup small robots league. We substantiate this claim in this paper by classifying different approaches and by discussing their potentials and limitations for research on AI and robotics. Furthermore, we present the most recent results of our approach to these goals in form of the so-called CubeSystem, a kind of construction-kit for robots and other autonomous systems. It is based on a very compact embedded computer, the so-called RoboCube, a set of sensor- and motor-modules, and software support in form of a special operating system and highlevel languages.

We present an approach to shape robots on their sensorial ability. We argue that the interface with the external world may strongly condition the design of a robot, from the mechanical aspects to reasoning and learning. We show the implementation of this philosophy in the RoboCup middle-size player Rullit, shaped on its omnidirectional vision sensor.

All mobile robots require some form of motion control in order to exhibit interesting autonomous behaviors. This is even more essential for multi-robot, highly-dynamic environments, such as robotic soccer. This paper presents the motion control system used by CMUnited-98, the small-size league champion at RoboCup-98. The team consists of five robots that aim at achieving specific goals while navigating in a limited space shared with the five other opponent robots. We introduce our motion control algorithm, which allows a general differential-driven robot to accurately reach a target point with a desired orientation in an environment with multiple moving obstacles. We describe how the features of our motion controller help to build interesting and robust behaviors. We also briefly compare our system to other motion control techniques and include descriptions and illustrations of the performance of our fully-implemented motion control algorithm.

Dual Dynamics (DD) is a mathematical model of a behavior control system for mobile autonomous robots. Behaviors are specified through differential equations, forming a global dynamical system made of behavior subsystems which interact in a number of ways. DD models can be directly compiled into executable code. The article (i) explains the model, (ii) sketches the Dual Dynamics Designer (DDD) environment that we use for the design, simulation, implementation and documentation, and (iii) illustrates our approach with the example of kicking a moving ball into a goal.

An early stage in image understanding using colour involves recognizing the colour of target objects by looking at individual pixels. However, even when, to the human eye, the colours in the image are distinct, it is a challenge for machine vision to reliably recognize the whole object from colour alone, due to variations in lighting and other environmental issues. In this paper, we investigate the use of decision trees as a basis for recognizing colour. We also investigate the use of colour space transforms as a way of eliminating variations due to lighting.

This paper presents Saracinescu, the goalkeeper robot of the Italian team that was used at the Robocup ’98 Paris championship. The machine features an original omni-directional vision system whose performance, enhanced by a simple but effective movement strategy, proved to be very smart and led to good results during the tournament. The paper describes the vision algorithms in detail, and discusses some issues that are still being developed and/or refined. An overview of the other components of the machine (mechanical structure and ball-kicking mechanism, computing architecture, auxiliary software routines for initial positioning, etc.) is also included.

The main ideas behind the implementation of the IALP RoboCup team are discussed: an agent architecture made of a hierarchy of behaviors, which can be combined to obtain different roles; a memory model which relies of the absolute positions of objects. The team is programmed using ECL, a Common Lisp implementation designed for being embeddable within C based applications. The research goal that we are pursuing with IALP is twofold: (1) we want to show the flexibility and effectiveness of our agent architecture in the RoboCup domain and (2) we want to test ECL in a real time application.

In this paper, the autonomous agent architecture used to implement the RoboCup simulator league UFSC-Team is presented. This architecture consists of three concurrent processes that encapsulate different inference engines. These take decisions in three different levels, called reactive, instinctive and cognitive. This architecture is an evolution of the concurrent architecture for cognitive multi-agents, used in the implementation of the UFSC-Team’98 that has participated in the RoboCup’98. The present implementation was designed to solve some agent synchronization and real-time response problems presented by the old architecture, due mainly to its centralized decision approach.

This paper describes the method employed to track and identify each robot during a Robocup match. Also, the playing ball is tracked with almost no extra processing effort. To track the robots it is necessary the use of adequate markers so that not only the position is extracted but also the heading. We discuss the difficulties associated with this problem, various possible approaches and justify our solution. The identification is performed thanks to a minimalist bar code placed in each robot. The bar code solves the problem of resolving some ambiguities that can arise in certain configurations. The procedure described can be executed in real time as it was shown in Paris in RoboCup-98.

Reinforcement learning has proven to be a set of successful techniques for finding optimal policies on uncertain and/or dynamic domains, such as the RoboCup. One of the problems on using such techniques appears with large state and action spaces, as it is the case of input information coming from the Robosoccer simulator. In this paper, we describe a new mechanism for solving the states generalization problem in reinforcement learning algorithms. This clustering mechanism is based on the vector quantization technique for signal analog-to-digital conversion and compression, and on the Generalized Lloyd Algorithm for the design of vector quantizers. Furthermore, we present the VQQL model, that integrates Q-Learning as reinforcement learning technique and vector quantization as state generalization technique. We show some results on applying this model to learning the interception task skill for Robosoccer agents.

Self-localization is important in almost all robotic tasks. For playing an aesthetic and effective game of robotic soccer, self-localization is a necessary prerequisite. When we designed our robotic soccer team for RoboCup’98, it turned out that all existing approaches did not meet our requirements of being fast, accurate, and robust. For this reason, we developed a new method, which is presented and analyzed in this paper. We additionally present experimental evidence that our method outperforms other methods in the RoboCup environment.

Knowing the position and orientation of a mobile robot situated in an environment is a critical element for effectively accomplishing complex tasks requiring autonomous navigation. Techniques for robot self-localization have been extensively studied in the past, but an effective general solution does not exist, and it is often necessary to integrate different methods in order to improve the overall result.In this paper we present a self-localization method that is based on the Hough Transform for matching a geometric reference map with a representation of range information acquired by the robot’s sensors. The technique is adequate for indoor office-like environments, and specifically for those environments that can be represented by a set of segments. We have implemented and successfully tested this method in the RoboCup environment and we consider this a good benchmark for its use in office-like environments populated with unknown and moving obstacles (e.g. persons moving around).

Virtual RoboCup is a real-time 3D visualization tool for 2D simulated soccer games as played in the RoboCup simulation league. Players are modeled as anthropmorphic figures and animated step-keepingly with the underlying 2D simulation. Important aspects of player animation concern the generation of natural 3D player movements and realistic player-ball interactions during kicks. A key contribution of Virtual RoboCup is its novel approach to task-level animation in which task-level directives for 3D animation of anthropomorphic characters are generated via on-line classification of fast paced 2D simulation data. As further contribution, we investigated to what extend human observers perceptually process the level of detail in Virtual RobCup animations. A psychological experiment was designed to test the effectiveness of 3D body animation. Although observers failed to notice differences in animation detail, clear effects of character animation on perceived skill were found. The experiment confirms that is is very well justified to spend valuable computational resources on naturalness and richness of detail in realtime character animation.

Increasingly, agent teams are used in realistic and complex multiagent environments. In such environments, dynamic and complex changes in the environment require appropriate adaptation of the teamwork (collaboration) among team-members. As RoboCup proposes to provide multi-agent researchers with a standard test-bed for evaluation of methodologies, it is only natural to use it for investigating this essential capability. During the RoboCup-98 workshop and competition a unique event took place: a comparative evaluation of the teamwork adaptation capabilities of 13 of the top competing teams. An evaluation attempt of this scale is a novel undertaking, and presents many novel challenges to researchers in the multi-agent community. This preliminary report describes the data-collection session, the experimental protocol, and some of the preliminary results from analysis of the data. Rather than proposing solutions and well understood results, it seeks to highlight key challenges in evaluation of multi-agent research in general, and of teamwork in particular..

Many AI professionals consider RoboCup small robots league competition as an ideal platform for testing distributed artificial intelligence techniques. Among these techniques are Multi-Agent systems (MAS), which advocate collective intelligence by focusing on autonomy of agents and their intercommunication. Multi-Agent Systems have been used by computer scientists and software engineers in several disciplines such as Internet and Industry [16]. For the robotics community Multi-Agent Systems was a paradigm shift from the classical centralized approach in building intelligent machines. By the late 80’s MAS were used in several multi robot systems ranging from cellular robots (Fukuda et al) [1] to a team of trash-collecting robots ( Arkin et al) [2]. This paper describes a distributed approach in implementing the Multi-Agent system architecture of a robotic soccer team, Temasek POlytechnic Team (TPOT).

To meet the timing requirements set by the RoboCup soccer server simulator, this paper proposes a multi-threaded approach to simulated soccer agents for the RoboCup competition. At its higher level each agent works at three distinct phases: sensing, thinking and acting. Instead of the traditional single threaded approaches, POSIX threads have been used here to break down these phases and implement them concurrently. The details of how this parallel implementation can significantly improve the agent’s responsiveness and its overall performance are described. Implementation results show that a multithreaded approach clearly outperforms a single-threaded one in terms of efficiency, responsiveness and scalability. The proposed approach will be very efficient in multi-processor systems.

A three-level functional architecture for a team of mobile robots is described in detail, including the definition of the role assigned to each level, the main concepts involved, and the corresponding implementation for each individual robot. The architecture is oriented towards teams of fully autonomous cooperative robots, able to carry out different types of cooperative tasks. Complexity is reduced by the decomposition of team strategies into individual behaviors, which in turn are composed of primitive tasks. Relationships among robots of the team are modeled upon the joint intentions framework. An application to Robotic Soccer and some of its preliminary results are presented.

The Real Magicol soccer players team for Robot Cup 99 is based on a BOC architecture extension which combines reactive and deliberative reasoning by the distribution of the knowledge system into modules called behaviors. The hardware remains the same as the 1998 Robot Cup one, our research bearing on the cooperative architecture based on agent concept.

Soccer Server has been used as the official simulator for RoboCup Simulation League last three years. Based on this experience, I investigate the feature of Soccer Server and figure out the issue to building such kind of open simulator. Then, I propose a new design of simulator that will provide more flexible version up, easiness of maintenance, and wide application.

This paper presents ETHNOS-IV - a real-time programming environment for the design of a system composed of different robots, devices and external supervising or control stations. ETHNOS is being used for different service robotics applications and it is has also been used successfully used in RoboCup in the Italian ART robot team during the Stockholm 99 competition. It provides support from three main point of views which will be addressed in detail: inter-robot and intra-robot communication, real-time task scheduling, and software engineering, platform independence and code-reuse. Experimental results will also be presented.

This paper describes the hardware and software architecture of the Attempto RoboCup-99 team. We first present the design of our heavily modified commercial robotic base, the robot sensors and onboard computer. Then the robot control architecture which realizes a hybrid control, consisting of a reactive behavior based component and a planner component for more complex tasks is introduced. Also the problems we currently are working on are presented, as there are a fast and reliable self localization algorithm and a robust behavior based reactive component for the hybrid control system.

Research in dynamical physical agents, consensus of proper physical decisions among physical agents, and an example of passing is shown. The interest is to introduce introspection of the dynamical behavior of each physical body so that every agent has better knowledge. This has to lead to better passes.

This paper describes a softbot agent capable of learning to choose its actions, in order to achieve its goal when facing an opponent in a dynamic environment. The agent uses rewards gathered from the environment to assess and improve the quality of its own behavior. A multilayer perceptron neural network is assessed regarding its adequacy as a value function approximator for state-action pairs in the robotic soccer domain.

Creating complex agents for simulation environments has long been the exclusive realm of AI experts. However it is far more desirable that experts in the particular application domain, rather than AI experts, are empowered to specify agent behavior. In this paper an approach is presented that allows domain experts to specify the high-level team strategies of agents for RoboCup. The domain experts’ specifications are compiled into behavior based agents.The 1999 RoboCup World Cup provided an interesting basis for evaluation of the approach. We found that for RoboCup it is not necessary to allow a user to change low level aspects of the agents’ behavior in order for them to create a range of different, interesting teams. We also found that the modular nature of behavior based architectures make them an ideal target architecture for compiling enduser specifications.

The annual robot soccer competition (RoboCup) provides an excellent opportunity for research in distributed robotic systems. A robotic soccer team demands integrated robots that are autonomous, efficient, cooperative, and intelligent. In this paper, we introduce the concept of Purposeful Behavior, to tackle the problem of achieving reactive and coordinated behavior in a team of autonomous robots. We are building a new control framework for autonomous robots to reason about goals and actions, react to unexpected situations, learn from humans and experience, and collaborate with teammates. Building such robots may require techniques that are different from those employed in separate research disciplines. We describe our experience in building these soccer robots and highlights problems and solutions that are unique to such multi-agent robotic systems in general. These problems include a framework for multi-agent programming, agent modeling and architecture, evaluation of multi-agent systems, and decentralized skill composition.

We developed an Autonomous Information Indication System for the RoboCup simulation league. This delivers and displays a three-dimensional view of the game to an audience using low-speed networks such as the Internet. Moreover, the audience has the ability to select a favorite shot from four different ones that are positioned on the field. Recenltly, our system performed succesfully at the RoboCup Japan Open 99. This paper outlines the feasibility and effectiveness of our system based on our evaluation of various experiments.

In this paper, we present a multi-layered architecture for spatial agents. The focus is laid on the declarativity of the approach, which makes agent scripts expressive and well understandable. They can be realized as (constraint) logic programs. The logical description language is able to express actions or plans for one and more autonomous and cooperating agents for the RoboCup (Simulator League). The system architecture hosts constraint technology for qualitative spatial reasoning, but quantitative data is taken into account, too.The basic (hardware) layer processes the agent’s sensor information. An interface transfers this low-level data into a logical representation. It provides facilities to access the preprocessed data and supplies several basic skills. The second layer performs (qualitative) spatial reasoning. On top of this, the third layer enables more complex skills such as passing, offside-detection etc. At last, the fourth layer establishes acting as a team both by emergent and explicit cooperation. Logic and deduction provide a clean means to specify and also to implement teamwork behavior.

RoboCup was introduced as a challenge area at IJCAI-97. We have been actively pursuing research in this area and have participated in the RoboCup competitions, winning the RoboCup-98 and RoboCup-99 simulator competitions. In this paper, we report on the main technical issues that we encountered and addressed in direct response to the learning and teamwork challenges stated in the IJCAI-97 challenge paper. We describe “layered learning” in which off-line and online, individual and collaborative, learned robotic soccer behaviors are combined hierarchically. We achieve effective teamwork through a team member agent architecture that encompasses a “flexible teamwork structure.” Agents are capable of decomposing the task space into flexible roles and can switch roles while acting. We report detailed empirical results verifying the effectiveness of the learned behaviors and the components of the team member agent architecture.

In recent years, many researchers in AI and Robotics pay attention to RoboCup, because robotic soccer games needs various techniques in AI and Robotics, such as navigation, behavior generation, localization and environment recognition. Localization is one of the important issues for RoboCup. In this paper, we propose a method of robot’s localization by integrating vision and modeling of the environment. The environment model that realizes the robotic soccer filed in the computer can produce an image of robot’s view at any location. In the environment model, the system can search and appropriate location of which view image is similar to the view image by the real robot. Our robot can estimate location from goal’s height and aspect ratio on the camera image. We search the most suitable position with hill-climbing algorithm from the estimated location. We programmed this method, and tested validity. The error range is reduced from lm∼50cm by robot’s estimation from 40cm∼20cm by this method. This method is superior to the other methods using dead reckoning or range sensor with map because it does not depend on the field size on precision, and does not need walls as landmark.

An extended value function is discussed in the context of multiple behavior coordination, especially in a dynamically changing multiagent environment. Unlike the traditional weighted sum of several reward functions, we define a vectorized value function which evaluates the current action strategy by introducing a discounted matrix to integrate several reward functions. Owing to the extension of the value function, the learning robot can estimate the future multiple rewards from the environment appropriately not suffering from the weighting problem. The proposed method is applied to a simplified soccer game. Computer simulations are shown and a discussion is given.

This report discusses two major views on BDI deliberation for autonomous agents. The first view is a rather conceptual one, presenting general BDI design principles, namely heuristic options, decomposed reasoning and layered planning, which enable BDI deliberation in realtime domains. The second view is focused on the practical application of the design principles in RoboCup Simulation League. This application not only evaluates the usefulness in deliberation but also the usefulness in rapid cooperative implementation. We compare this new approach, which has been used in the Vice World Champion team AT Humboldt 98, to the old approach of AT Humboldt 97, and we outline our ideas for further improvements, which are still under work.

Our agent team AT Humboldt 99 (AT stands for “Agent Team”) was developed as extension of our former team AT Humboldt 98, which became vice champion at RoboCup-98. We started to extend it by improved skills, new options and a larger planning horizon, respectively. So the most features of our current team were already part of AT Humboldt 98 which has been briefly described in [3] and extensive described in [5]. A description of our first soccer team AT Humboldt 97, which became world champion at RoboCup-97, can be found in [1].

This paper describes a framework for formalising tactical reasoning in dynamic multi-agent systems, populated by synthetic (software) agents. The proposed framework is based on a hierarchy of synthetic agent architectures and is expressive enough to capture a subset of desirable properties from both the situated automata and subsumption-style architectures, while retaining the rigour and clarity of logic-based possible worlds semantics. This framework is successfully realised in the RoboCup Simulation League domain. Not only did it provide a solid design approach to object-orientation, but it also enabled incremental implementation and testing of software agents and their modules. In particular, the framework allowed us to correlate enhancements in the agent architecture with tangible improvements in team performance. Cyberoos98 was 3rd place winner of the Pacific Rim series at PRICAI-98. Cyberoos99 finished in the top 18 of the RoboCup-99.

The major purpose of this research is to study cooperative planning for multiagent systems in time-critical environment. The RoboCup simulator league is the most interesting target for our research.

Team Erika’s main focus is on the facilatation of the design of agent behavior. The behavior code is generated by a graph editor which process transition diagram like graph. Since the concept is represented visually, high design efficency can be achieved. Besides, people other than computer scientist can design the behavior easily without understanding the underlying stucture.

This paper describes the Essex Wizards team participated in the RoboCup’99 simulator league. It is mainly concentrated on a multi-threaded implementation of simulated soccer agents to achieve real-time performance. Simulated robot agents work at three distinct phases: sensing, thinking and acting. POSIX threads are adopted to implement them concurrently. The issues of decision-making and co-operation are also addressed

The emphasis of FCFoo was mainly on building a library for developers of RoboCup teams, designed especially for educational use. After the competition the library was more or less totally rewritten and finally published as part of the Master Thesis of Fredrik Heintz [4].

This document describes the software architecture of the Footux-99 team (simulation league). It is now well known that purely reactive (resp. cognitive) agents are out of date. An agent must be able to respond reactively when necessary, but it should have a general behaviour guideline, strategy. The most classical approach consists in using a hybrid architecture.The architecture we are introducing in this article is a hybrid one. It combines vertical and horizontal hybrid approachs where each layer is based on a subsumption architecture.The aim of our approach is to study the possibility to obtain a cooperative behavior within a multi-agents system without using centralized control, and thus to observe the emergence of potential relations between an agent and the society to which it belongs.

The Gongeroos’99 team involves agents built within the broad framework defined by the BDI agent architecture [3] with novel features involving the application of notions from team-oriented programming [2] and multi-hop ad-hoc communication networks [1] from the area of mobile computing. Gongeroos’99 achieved a 9th place ranking in RoboCup-99’s software simulation league.

The development of the Headless Chickens III emphasized a high level team specification environment, called the Strategy Editor, that was intended for use by endusers, rather than computer programmers[2]. Using the strategy editor consisted of placing players on a image of the ground and indicating the direction(s) the player should kick and/or dribble when they get the ball. Different player formations and passing/dribbling patterns could be specified for different game situations. The designer could also specify the style of play for each of the players, e.g. defensive or inclined to shoot or dribble.

IALP is a team for the simulation league of the RoboCup initiative [4]. The team is programmed using ECL, a public domain implementation of Common Lisp [1].

In order to realize flexible strategic planning in multi-agent systems that are working in dynamic environment, it is necessary to provide a mechanism to integrate hierarchical planning (include team planning) and reactive behavior. The main issues of this integration are: how to switch the context of plan In the dynamic environment, it is important how to terminate making and executing a plan when the environment changes so that the plan is not useful any more.how to organize multiple planning It is better that agents in a complex environment can have ability to making multiple planning, because such agents may have multiple goals in the same time. For example, in the case of soccer, while agents have an obvious goal “win the game (or score goals)”, also the agent should have another instinctive goals, that is “not to miss their position”, “follow the rule”, and so on, in the same time. The similar requirement will be happen when agents try to make a consensus by communication during they were acting something. So, it will make the problem simple that the agent has parallel planning process, that is action planning and communication planning.

The following paper describes the design principles of decision making in the Karlruhe Brainstormers team that participated in the RoboCup Simulator League in Stockholm 1999. It is based on two basic ingredigents: the priority - probability - quality (PPQ) concept is a hybrid rule-based/ learning approach for tactical decisons, whereas the definition of goal-orientented moves allows to apply neural network based reinforcement learning techniques on the lower level.

Its on-line coach agent characterizes Kasuga-bito III. Kasuga-bito III is composed of LogMonitor [LogM]and Kasuga-bito II, which was runner-up in JapanOpen’98 and the champion in the RoboCup Pacific Rim Series ’98. The LogMonitor advises their position to players. The positioning strategy is authenticated by analysis of the logfiles of evaluations at RoboCup’98. Our player agents are advised by the on-line coach and changes their formation.

In this paper we describe our team, KU-Sakura2, which is to participate in the simulation league of RoboCup-99 Stockholm. KU-Sakura2 is characterized by soccer agents that make tactical plays and passes using communication between players.

The main interest of our research concerns motivation action control and goal management of agents (magma). Action Control of the magmaFreiburg team is based on extended behavior networks, which add situation-dependent motivational influences to the agent, extend original behavior networks to exploit information from continuous domains and allow concurrent execution of behaviors. Advantages of the original networks, such as reactivity, planning capabilities, consideration of multiple goals and its cheap calculations are maintained.

Our agent team is the result of a development which had to take place under tight time limitations. The total development time available was slightly less than three months where over most of the time the team developers could invest no more than a few hours per week. The code was developed from scratch to improve over the design and quality of last year’s code. Thus one of the challenges was to keep a smooth development line and to avoid dead ends in the development, as well as to maintain a development environment in which a larger number of developers could work productively.

Since the offside rule was adopted in RoboCup-98, many teams without teamwork ability got offside penalty many times in their matches. Those teams who have dribble skill won, because most of other teams have not effcient defence strategy.

Oulu99 team was formed by students of University of Oulu, Finland as a part of student’s Software Project course. Entire software was designed and written from scratch, even though there was source code available from previous Oulu teams. as a result, Oulu99 finished on 13th place in Robocup’99.

Pardis, was one of the entries in RoboCup-99, simulation league. It had a optimistic timing in communication with the server. And lost most of the cycles in the real league, because of relying on the enough network bandwidth. So unfortunately it had chance to be only in the first round robin. It used an experimental model, consisting of finite set of categories for each player. Each softbot in Pardis team, was a player acting as designed in a specific category. The coach had the ability to map each player in the opponent team with one of the same categories. It dynamically changed the characteristics (category) of the facing teammate to be effective against the analyzed opponent player. Although in the real league, there was no chance to see the use of the coach and it was never activated. The players read their behavioral configuration once at the start of the game and kept playing that way.

PaSo-Team is a Multi-Agent system for playing soccer game in the Simulation League of the RoboCup competition. This paper describes the ideas and the technical structure of PaSo-Team’99, that played at RoboCup-99, in Stockholm during IJCAI’99. The main goal of the 1999 project was about the integration of a reactive model with some kind of high-level reasoning. Obstacle avoidance and motion reasoning are encapsulated at the behavior level. They use a proper world model (built from the sensed data), that focus on the relevant objects. The choice&evaluate problem is performed through an utility function over a proper coding of the prototypical game arrangements.

Our team - PSI was developed at Program Systems Institute of Russian Academy of Science. This paper is a short description of the dynamical refinement planning method that we use to construct our software agents.

The RoboCup scenario yields a variety of fields of research. The main goal of the RoboLog project, undertaken at the University of Koblenz in Germany, is the specification and implementation of flexible agents in a declarative manner. The agents should be able to deal with the real-time requirements but also be capable of more complex behavior, including explicit teamwork. To this end, we develop a declarative multi-agent script language for the specification of collective actions or intended plans that are applicable in certain situations. The agents should be able to recognize such situations by means of qualitative spatial reasoning, possibly supported by communication.

This paper describes the research in a new Rogi Team conceived by simulation in Java. It is a development of ideas for rational agents that cooperate and use revision of exchanged information and consensus techniques.

The Ulm Sparrows RoboCup team was initiated in early 1998. Among the goals of the team effort are to investigate methods for skill learning, adaptive spatial modeling, and emergent multiagent cooperation [1]. We develop both a middle-size robot league and a simulation league team. Based mostly on equipment and technology available in our robot lab, we implemented a first version of both teams for RoboCup-98 in order to gain practical experience in a major tournament. Based on the these experiences, we made significant progress in our team effort in several areas: we designed new robot hardware, extended our vision processing capabilities and implemented a revised and more complete version of our soccer agent software architecture. In particular, we added Monte Carlo localization techniques to our robots, enhanced environment modeling, and started to apply reinforcement learning techniques to improve basic playing skills.

The aim of developing UBU is to subject a series of tools and procedures for agent decision support to a dynamic real-time domain. These tools and procedures have previously been tested in various other domains, e.g., intelligent buildings [2] and social simulations [6]. The harsh time constraints of RoboCup requires true bounded rationality, however, as well as the development of anytime algorithms not called for in less constrained domains (cf. [3]). Artificial decision makers are in the AI and agent communities usually associated with planning and rational (as in utility maximising) behaviour. We have instead argued for the coupling of the reactive layer directly to decision support. A main hypothesis is that in dynamic domains (such as RoboCup), time for updating plans is insufficient. Basically depending on the size requirements of agents, and on the communication facilities available to the agents, we have placed decision support either in the agents, or externally. In the former case, deliberation is made in a decision module. In the latter case, a kind of external calculator which we have named pronouncer provides rational action alternatives. The input to the pronouncer is decision trees or influence diagrams. The structure and size of these models are kept small, to guarantee fast evaluation (cf. [7]). The pronouncer can be made into an agent too, e.g., by using a wrapper. The coach function is particularly interesting in this context, since it is “free” and since it could hold the pronouncer code. An important problem here is the uncertainty and space constraints on the communication with the coach. The concept of norms as constraints on agent actions has also been investigated [1]. A team in which each boundedly rational player maximises its individual expected utility does not yield the best possible team: Group constraints on actions must be taken into account (see, e.g., [4]). Norms is our way of letting the coalitions that an agent is part of play a part in the deliberation of the agent.

I want to do various researches by using RoboCup Seccor Server in my laboratory. However, the history of my laboratory in RoboCup is shallow and only half a year has passed since the research began. The research theme at which we aimed at first was ”Real time, Distribution, and Cooperation” and it was the approaches from three sides. The outline of the team is described in this paper around World Modeling which should be called the result of ”Real time and Distribution” thought to have reached at a standard level by present.

This paper discusses the design of the team Zeng99. The goal of team Zeng99 is to show a performance of Hierarchical Fuzzy Intelligent Control system in the field of multi agent problems. It worked well at RoboCup99 competition, even with little error in an invoking clients. It also allow independent/cooperative development client by client.

This paper discusses some important features, which make the All Botz, the University of RoboCup team, a very unique team. In particular, the use of cheap hardware and the design of the video server.

In this paper we describe Big Red, the Cornell University Robot Soccer team. The success of our team at the 1999 competition can be mainly attributed to three points: 1.An integrated design approach; students from mechanical engineering, electrical engineering, operations research, and computer science were involved in the project, and a rigorous and systematic design process[2] was utilized.2.A thorough understanding of the system dynamics, and ensuing control.3.A high fidelity simulation environment that allowed us to quickly explore AI and control strategies well in advance of working prototypes.

One of the necessary steps in entering a small-size RoboCup team is the actual construction of the robots. We have successfully built robots for RoboCup-97 and RoboCup-98, leading to two champion teams, namely CMUnited-97 [2] and CMUnited-98 [1].

This paper describes the 5dpo team. The paper will be divided into three main sections, corresponding to three main blocks: the Global Level, the Local Level and the Interface Level. These Levels, their subsystems and some implementation details will be described next.

This paper describes the team FU-Fighters that won the second place in the RoboCup’99 F180-league competition.The paper presents the mechanical and electrical design of our robots, including a kicking device. We also explain the hierarchical control architecture we used to generate the behavior of individual agents and the team. This reactive approach is mainly based on the Dual Dynamics concept developed by H. Jäger. In addition we describe, how the problems of vision and radio communication have been addressed.

Figure 1 describes the overview of our team’s system. The concept of our team is to employ simple, inexpensive robots and to control them by high-speed and actual vision feedback. The special hardware for color detection is developed and employed for our vision system to exract coordinates of ball and markers. Global strategy for the team is realized by rule based logic on the host computer.

OWARI-BITO team consists of 5 small robots, each of which is sized in 10 cm wide, 10 cm deep and 17 cm high (except an antenna). The purposes of the research project are the study on the cooperation among robots, the advanced local vision system and the robust communication environment, in addition to be able to win the game.

This paper describes the main features of the new Rogi Team and some research applied focused on dynamics of physical agents. It explains the vision system, the control system and the robots, so that the research on dynamical physical agents could be performed. It presents part of the research done in physical agents, especially consensus of properly physical decisions among physical agents, and an example applied to passing.

TPOTS is a Team of small size robots designed using a hybrid control architecture distributed among the robots and the host computer. The major characteristic of the RoboCup soccer competition is the dynamic nature of the environment in which robots operate. The only static object in the competition field is the field itself. Team and opponent robots as well as the ball can be placed anywhere in the field, be it a purposeful strategic positioning, a missed action or a forced displacement. This has led many researchers to shift from the traditional model-based top down control [1,2] to a reactive behavior based approach [3,4,5,6,7]. Robots need not waste a huge amount of resources building maps and generating paths that might prove useless at the time of action. Instead robots are supposed to react to the actual changes in the environment in a simple stimulus-response manner [8]. However due to the size limitations imposed by the RoboCup small robots league (15cm diameter circle) and rich visual input, on-board vision proved to be a complex and expensive task.

The VUB AI-lab team is mainly interested in the two loosely linked aspects of on-board control and heterogeneity. One major effort for fostering both aspects within RoboCup’s small robots league is our development of a so-to-say robot construction-kit, allowing to implement a wide range of players with on-board control. For the ’99 competition, the existing RoboCube controller-hardware has been further improved. In addition, some solid and precise mechanical building-blocks were developed, which can easily be mounted on differently shaped bottom-plates. On top of these engineering efforts, we report here a computational inexpensive but efficient algorithm for motion-control, including obstacle avoidance. Furthermore, we shortly address the issue of increased difficulties of coordinating so-to-say multiple teams due to the possible variations based on heterogeneity. Operational semantics based on abstract data-types and patter matching capabilities can be a way out of this problem.

This paper describes the Agilo RoboCuppers - the RoboCup team of the image understanding group (FG BV) at the Technische Universität München. With a team of five Pioneer 1 robots, equipped with CCD camera and a single board computer each and coordinated by a master PC outside the field we participate in the Middle Robot League of the Third International Workshop on RoboCup in Stockholm 1999. We use a multi-agent based approach to represent different robots and to encapsulate concurrent tasks within the robots. A fast feature extraction based on the image processing library HALCON provides the data necessary for the onboard scene interpretation. In addition, these features as well as the odometric data of the robots are sent over the net to the master PC, where they are verified with regard to consistency and plausibility and fusioned to one global view of the scene. The results are distributed to all robots supporting their local planning modules. This data is also used by the global planning module coordinating the team’s behaviour.

Azzurra Robot Team (ART) is the National Italian Team for F-2000 RoboCup league, developed within the RoboCup Italia project. ART99 is formed by six academic groups and Consorzio Padova Ricerche. ART started with RoboCup-98, and its goal is to exploit the expertise and ideas from all groups in order to build a team where players have different features (hw and sw), but retain the ability to coordinate their behaviour within the team. ART99 obtained the second place in RoboCup-99 F-2000 league, and coordination among players is, in our view, the most significant achievement of the team.

This paper presents the hardware and software design principles of the medium size RoboCup Team CoPS which are developed by the image understanding group at the Institute for Parallel and Distributed High Performance Systems (IPVR) of the University of Stuttgart. By adapting already successfully tested multiagent software concepts by our group to the domain of robotic soccer we intend to improve those concepts at the field of realtime applications with uncertain sensory data.

One of the interesting challenges in designing a successful robotic soccer team is the need to cover the entire loop from sensing over deliberation to acting. For example, successful ball passing needs good estimations of the position and velocity of the other players and the ball, projections into the future, planning ahead in order to create and exploit opportunities, and, finally, it requires to act accordingly.

The annual Robocup soccer competition is an excellent opportunity for our robotics and agent research. We view the competition as a rigorous testbed for our methods and a unique way of validating our ideas. After two years of competition, we have begun to understand what works (we won the competition in Tokyo 97) and what does not work (we failed to advance to the second round in Paris 98). This paper presents an overview of our goals in Robocup, our philosophy in building soccer playing robots and the methods we are employing in our efforts.

This article gives a brief sketch of the scientific and engineering approach taken at the GMD RoboCup Team. We sketch (i) the robot hardware, (ii) the “Dual Dyanmics” model of behavior control that we develop, and (iii) the integrated “Dual Dynamics Designer” environment that we use for programming, simulation, documentation, and code generation.

The SocRob project was born as a challenge for multidisciplinary research on broad and generic approaches for the design of a cooperative robot society, involving Control, Robotics and Artificial Intelligence researchers. In this paper the basic aspects of last year implementation as well as the improvements made meanwhile are briefly recalled and presented. Naturally, a special emphasis is given here to the novel solutions proposed for this year implementation, the results obtained and the expected future developments.

Autonomous soccer robots should recognize the environment from the captured image from a video camera and plan to proper behavior. Furthermore, when some robots play cooperatively, communication system between robots is important inputs. We choose simple vision and actuator system, then the gap between the real world and simulation environment are small. Our research target is to accomplish multiagent system using reinforcement learning.

This paper describes our research interests and technical information of our team for RoboCup-99. Our robots have been developed to have advantages for playing soccer. That is, the capability of kicking a ball and high mobility. We developed pneumatic kickers and omnidirectional bases.

Through robotic soccer issue, we focus on “perception” and “situation and behavior” problem among RoboCup physical agent challenges [1]. So far, we have implemented some behaviors for playing soccer by combining four primitve processes (motor control, camera control, vision, and behavior generation processes)[2]. Such behaviors were not sophisticated very much because they were fully implemented by the human programmer. In order to improve the performance of such behaviors, a kind of learning algorithm would be useful during off/on-line skill development phase.

This paper describes an Autonomous Mobile Robot team which plays football, developed by the Group of Automation and Robotics at the Industrial Electronics department of the University of Minho, in Guimarães (Portugal). In this competition each team is free to use and/or build all the different electronics, sensory systems, playing algorithms, etc. as far as they cope with the rules imposed by the organisation. Instead of using several different sensors increasing electronics complexity, this team decided to use only one major sensor: a vision system with a small colour camera. All the image processing algorithms were developed from scratch and they consist on the heart of the whole project. This vision system uses an innovative approach: in order to see the whole field, a convex mirror was placed at the top of the robot looking downwards with the video camera looking upwards towards the mirror. This way, the robot can see all around itself with a top view, which means continuous vision of the ball, goals and other robots.

The hardware and software architectures of the Real MagiCol robots are presented. The hardware remains the same as the 1998 Robot Cup one, our research bearing on the cooperative architecture based on agent concept. The Real Magicol soccer players team for Robot Cup 99 is based on a Behavior Oriented Commands (BOC) architecture extension which combines reactive and deliberative reasoning by the distribution of the knowledge system into modules called behaviors.

The RMIT Raiders team is composed of three custom-made robots and one Pioneer robot. The most significant feature of the custom platform is a powerful kicking device, which proved itself in the first game by kicking a goal from the centre of the field. Compared with previous competitions, our custom robots were much more reliable and the batteries lasted much longer.

Arvand is a robot specially designed and constructed for playing soccer according to RoboCup rules and regulations for the medium size robots. This robot consists of three main parts: mechanics (motion mechanism and kicker), hardware (image acquisition, processing unit and control unit) and software (image processing, wireless communication, motion control and decision making). The motion mechanism is based on a drive unit, a steer unit and a castor wheel. We designed a special control board which uses two microcontrollers to carry out the software system decisions and transfers them to the robot mechanics. The software system written in C++ performs real time image processing and object recognition. Playing algorithms are based on deterministic methods. We have constructed 4 such robots and successfully tested them in a soccer field according to RoboCup regulations for middle size robots.

This is the team description of Osaka University “Trackies” for RoboCup-99. We have worked two issues for our new team. First, we have changed our robot system from a remote controlled vehicle to a self-contained robot. The other, we have proposed a new learning method based on a Q-learning method so that a real robot can aquire a bhevior by reinforcement learning.

This paper describes the 5dpo team. The paper will be divided into three main sections, corresponding to three main blocks: the Global Level, the Local Level and the Interface Level. These Levels, their subsystems and some implementation details will be described next.

Our team focused on manipulating the ball. We developed the kicking and heading motion. Two kinds of neural networks are utilized in order to recognize the ball and kick it. The recognition and manipulation did not work sufficiently, but intdicates our interest of playing soccer with legged robots. We confronted the team LRP and the team McGill in the round robin. We defeated McGill by PK, and lost the game with LRP 0-2.

Our interests are learning issues such as action selection, emergence of walking, and self localization without 3D-reconstruction. We implemented teaching, self localization without 3D-reconstruction and embodied trot walking.

The robots used in this competition were generously provided by Sony [3]. The robots are the same as the commercial AIBO robots except for slight hardware changes and programming capabilities. These autonomous robots are about 30cm long and have 18 degrees of freedom. The neck pans ±90° allowing the robot to scan the field with its on board camera. Six uniquely colored landmarks are placed around the field (at the corners and center-line) to help the robots localize. Each team consists of three robots. Like our team last year, CMTrio-98 [5], we divided our team between two identical attackers and one goalie.

The team members include students as well as members of the teaching stuff from the Department of Computer Science at the Humboldt University. They represent the groups of Artificial Intelligence, Responsive Computing, and Signal Processing, respectively. It was the aim of the project to combine the skills of these disciplines to program soccer playing legged robots.

After a period of experimentation and investigation into the dog’s possibilities we defined six core areas in which we would concentrate our development efforts. In concert, we felt, these areas of functionality would give a strong system capable of succeeding at the proposed challenges. The six major areas of functionality can be roughly divided into two groups, informally labeled input and output. Input consists of the tasks of Vision, odometry and localization/map-building. Output embodies the tasks of moving, path planning and AI (decision making). These areas are treated in greater detail in the remainder of this document, after a brief overview of our system’s infrastructure.

“Team Sweden” is the Swedish national team that entered the Sony legged robot league at the RoboCup 99 competition. We had two main requirements in mind when preparing our entry to the competition: 1.The entry should effectively address the specific challenges present in this domain; in particular, it should be able to tolerate errors and imprecision in perception and execution; and2.it should illustrate our research in autonomous robotics, by incorporating general techniques that can be reused in different robots and environments.

The system we developed uses a hierarchial approach to software design, and consists of three subsystems: vision, acting, and planning. We also developed an offline classification system that uses concepts derived from Machine Learning.

The main areas of focus for our team were the development of solid algorithms for merging vision data with other inputs and basic strategies for moving the ball towards the goal and for defending the goal. Our team came in fifth place overall in the competition. We defeated Team Sweden 2-0, though both goals were “own-goals” (scored by Team Sweden). We lost 2-0 to Osaka Univ., in a game where the first half we spent 90% of the time in the corner of the field near their goal, unable to score, while in the second half they scored one quick goal, just making it past our (slow) goalkeeper, and then a second goal later on in the game. We also had two scrimmage games-a 2-0 victory over Team Sweden, and a 0-0 tie against the team from Humboldt.