RoboCup 2004: Robot Soccer World Cup VIII

RoboCup is an international initiative with the main goals of fostering research and education in Artificial Intelligence and Robotics, as well as of promoting Science and Technology to world citizens. The idea is to provide a standard problem where a wide range of technologies can be integrated and examined, as well as being used for project-oriented education, and to organize annual events open to the general public, where different solutions to the problem are compared.

In this paper we propose an approach for tracking a moving target using Rao-Blackwellised particle filters. Such filters represent posteriors over the target location by a mixture of Kalman filters, where each filter is conditioned on the discrete states of a particle filter. The discrete states represent the non-linear parts of the state estimation problem. In the context of target tracking, these are the non-linear motion of the observing platform and the different motion models for the target. Using this representation, we show how to reason about physical interactions between the observing platform and the tracked object, as well as between the tracked object and the environment. The approach is implemented on a four-legged AIBO robot and tested in the context of ball tracking in the RoboCup domain.

UCHILSIM is a robotic simulator specially developed for the RoboCup four-legged league. It reproduces with high accuracy the dynamics of AIBO motions and its interactions with the objects in the game field. Their graphic representations within the game field also possess a high level of detail. The main design goal of the simulator is to become a platform for learning complex robotic behaviors which can be directly transferred to a real robot environment. UCHILSIM is able to adapt its parameters automatically, by comparing robot controller behaviors in reality and in simulations. So far, the effectiveness of UCHILSIM has been tested in some robot learning experiments which we briefly discuss hereinafter. We believe that the use of a highly realistic simulator might speed up the progress in the four legged league by allowing more people to participate in our challenge.

RoboCup has hosted a coach competition for several years creating a challenging testbed for research in advice-giving agents. A coach agent is expected to advise an unknown coachable team. In RoboCup 2003, the coachable agents could process the coach’s advice but did not include a protocol for communication among them. In this paper we present CommLang, a standard for agent communication which will be used by the coachable agents in the simulation league at RoboCup 2004. The communication standard supports representation of multiple message types which can be flexibly combined in a single utterance. We then describe the application of CommLang in our coachable agents and present empirical results showing the communication’s effect on world model completeness and accuracy. Communication in our agents improved the fraction of time which our agents are confident of player and ball locations and simultaneously improved the overall accuracy of that information.

The Segway Human Transport (HT) is a one person dynamically self-balancing transportation vehicle. The Segway Robot Mobility Platform (RMP) is a modification of the HT capable of being commanded by a computer for autonomous operation. With these platforms, we propose a new domain for human-robot coordination through a competitive game: Segway Soccer. The players include robots (RMPs) and humans (riding HTs). The rules of the game are a combination of soccer and Ultimate Frisbee rules. In this paper, we provide three contributions. First, we describe our proposed Segway Soccer domain. Second, we examine the capabilities and limitations of the Segway and the mechanical systems necessary to create a robot Segway Soccer Player. Third, we provide a detailed analysis of several ball manipulation/kicking systems and the implementation results of the CM-RMP pneumatic ball manipulation system.

We describe a new method for detecting features on a marked RoboCup field. We implemented the framework for robots with omnidirectional vision, but the method can be easily adapted to other systems. The focus is on the recognition of the center circle and four different corners occurring in the penalty area. Our

constructive approach

differs from previous methods, in that we aim to detect a whole palette of different features, hierarchically ordered and possibly containing each other. High-level features, such as the center circle or the corners, are constructed from low-level features such as arcs and lines. The feature detection process starts with low-level features and iteratively constructs higher features. In RoboCup the method is valuable for robot self-localization; in other fields of application the method is useful for object recognition using shape information.

We propose to use a robust method for appearance-based matching that has been shown to be insensitive to illumination and occlusion for robot self-localization. The drawback of this method is that it relies on panoramic images taken in one certain orientation, restricts the heading of the robot throughout navigation or needs additional sensors for orientation, e.g. a compass. To avoid these problems we propose a combination of the appearance-based method with odometry data. We demonstrate the robustness of the proposed self-localization against changes in illumination by experimental results obtained in the RoboCup Middle-Size scenario.

In this paper, we present a new approach for omnidirectional vision-based self-localization in the RoboCup Middle-Size League. The omnidirectional vision sensor is used as a range finder (like a laser or a sonar) sensitive to colors transitions instead of nearest obstacles. This makes it possible to have a more reach information about the environment, because it is possible to discriminate between different objects painted in different colors. We implemented a Monte-Carlo localization system slightly adapted to this new type of range sensor. The system runs in real time on a low-cost pc. Experiments demonstrated the robustness of the approach. Event if the system was implemented and tested in the RoboCup Middle-Size field, the system could be used in other environments.

In the RoboCup four-legged league, robots mainly rely on artificial coloured landmarks for localisation. As it was done in other leagues, artificial landmarks will soon be removed as part of the RoboCup push toward playing in more natural environments. Unfortunately, the robots in this league have very unreliable odometry due to poor modeling of legged locomotion and to undetected collisions. This makes the use of robust sensor-based localization a necessity. We present an extension of our previous technique for fuzzy self-localization based on artificial landmarks, by including observations of features that occur naturally in the soccer field. In this paper, we focus on the use of corners between the field lines. We show experimental results obtained using these features together with the two nets. Eventually, our approach should allow us to migrate from landmarks-only to line-only localisation.

This paper describes a behavior-based architecture which integrates existing potential field approaches concerning motion planning as well as the evaluation and selection of actions into a single architecture. This combination allows, together with the concept of competing behaviors, the specification of more complex behaviors than the usual approach which is focusing on behavior superposition and is mostly dependent on additional external mechanisms. The architecture and all methods presented in this paper have been implemented and applied to different robots.

Navigation is one of the most fundamental tasks to be accomplished by many types of mobile and cognitive systems. Most approaches in this area are based on building or using existing allocentric, static maps in order to guide the navigation process. In this paper we propose a simple egocentric, qualitative approach to navigation based on ordering information. An advantage of our approach is that it produces qualitative spatial information which is required to describe and recognize complex and abstract, i.e., translation-invariant behavior. In contrast to other techniques for mobile robot tasks, that also rely on landmarks it is also proposed to reason about their validity despite insufficient and insecure sensory data. Here we present a formal approach that avoids this problem by use of a simple internal spatial representation based on landmarks aligned in an

extended panoramic representation

structure.

Collision detection in a quadruped robot based on the comparison of sensor readings (actual motion) to actuator commands (intended motion) is described. Ways of detecting such incidences using just the sensor readings from the servo motors of the robot’s legs are shown. Dedicated range sensors or collision detectors are not used. It was found that comparison of motor commands and actual movement (as sensed by the servo’s position sensor) allowed the robot to reliably detect collisions and obstructions. Minor modifications to make the system more robust enabled us to use it in the RoboCup domain, enabling the system to cope with arbitrary movements and accelerations apparent in this highly dynamic environment. A sample behavior is outlined that utilizes the collision information. Further emphasis was put on keeping the process of calibration for different robot gaits simple and manageable.

We show how to apply learning methods to two robotics problems, namely the optimization of the on-board controller of an omnidirectional robot, and the derivation of a model of the physical driving behavior for use in a simulator.

We show that optimal control parameters for several PID controllers can be learned adaptively by driving an omni directional robot on a field while evaluating its behavior, using an reinforcement learning algorithm. After training, the robots can follow the desired path faster and more elegantly than with manually adjusted parameters.

Secondly, we show how to learn the physical behavior of a robot. Our system learns to predict the position of the robots in the future according to their reactions to sent commands. We use the learned behavior in the simulation of the robots instead of adjusting the physical simulation model whenever the mechanics of the robot changes. The updated simulation reflects then the modified physics of the robot.

In this paper, we report on the status of the RoboCupJunior league, four years after it was founded. Since its inception in 2000, we have been surveying and/or interviewing students and mentors who participate in the international event. Here we present a high-level overview of this data. We discuss demographics of participants, characteristics of preparation and educational value. We highlight trends and identify needs for the future, in terms of event organization, educational assessment and community-building.

Real-time decision making based on visual sensory information is a demanding task for mobile robots. Learning on high-dimensional, highly redundant image data imposes a real problem for most learning algorithms, especially those being based on neural networks. In this paper we investigate the utilization of evolutionary techniques in combination with supervised learning of feedforward nets to automatically construct and improve suitable, task-dependent preprocessing layers helping to reduce the complexity of the original learning problem. Given a number of basic, parameterized low-level computer vision algorithms, the proposed evolutionary algorithm automatically selects and appropriately sets up the parameters of exactly those operators best suited for the imposed supervised learning problem.

To date, RoboCup games have all been played under constant, bright lighting conditions. However, in order to meet the overall goal of RoboCup, robots will need to be able to seamlessly handle changing, natural light. One method for doing so is to be able to identify colors regardless of illumination:

color constancy

. Color constancy is a relatively recent, but increasingly important, topic in vision research. Most approaches so far have focussed on stationary cameras. In this paper we propose a methodology for color constancy on mobile robots. We describe a technique that we have used to solve a subset of the problem, in real-time, based on color space distributions and the KL-divergence measure. We fully implement our technique and present detailed empirical results in a robot soccer scenario.

This paper presents a vision system for robotic soccer which was tested on Sony’s four legged robot Aibo. The input for the vision system are images of the camera and the sensor readings of the robot’s head joints, the output are the positions of all recognized objects in relation to the robot. The object recognition is based on the colors of the objects and uses a color look-up table. The vision system creates the color look-up table on its own during a soccer game. Thus no pre-run calibration is needed and the robot can cope with inhomogeneous or changing light on the soccer field. It is shown, how different layers of color representation can be used to refine the results of color classification. However, the self-calibrated color look-up table is not as accurate as a hand-made. Together with the introduced object recognition which is very robust relating to the quality of the color table, the self-calibrating vision works very well. This robustness is achieved using the detection of edges on scan lines.

In this paper we present a proposal for setting camera parameters which we claim to give results better matched to applications in color-coded environments then the camera internal algorithms. Moreover it does not require online human intervention, i.e. is automated, and is faster than a human operator. This work applies to situations where the camera is used to extract information from a color-coded world. The experimental activity presented has been performed in the framework of Robocup mid-size rules, with the hypothesis of temporal constancy of light conditions; this work is the necessary first step toward dealing with slow changes, in the time domain, of light conditions.

In challenging environments where the risk of loss of a robot is high, robot teams are a natural choice. In many applications like for example rescue missions there are two crucial tasks for the robots. First, they have to efficiently and exhaustively explore the environment. Second, they must keep up a network connection to the base-station to transmit data to ensure timely arrival and secure storage of vital information. When using wireless media, it is necessary to use robots from the team as relay stations for this purpose. This paper deals with the problem to combine an efficient exploration of the environment with suited motions of the robots to keep data transmissions stable.

Cooperative localization of objects is an important challenge in multi-robot systems. We propose a new approach to this problem where we see each robot as an expert which shares unreliable information about object locations. The information provided by different robots is then combined using fuzzy logic techniques, in order to reach a

consensus

between the robots. This contrasts with most current probabilistic techniques, which average information from different robots in order to obtain a

tradeoff

, and can thus incur well-known problems when information is unreliable. In addition, our approach does not assume that the robots have accurate self-localization. Instead, uncertainty in the pose of the sensing robot is propagated to object position estimates. We present experimental results obtained on a team of Sony AIBO robots, where we share information about the location of the ball in the RoboCup domain.

Robot recognition is a very important point for further improvements in game-play in

RoboCup

middle size league. In this paper we present a neural recognition method we developed to find robots using different visual information. Two algorithms are introduced to detect possible robot areas in an image and a subsequent recognition method with two combined multi-layer perceptrons is used to classify this areas regarding different features. The presented results indicate a very good overall performance of this approach.

Humans process images with apparent ease, quickly filtering out useless information and identifying objects based on their shape and colour. However, the undertaking of visual processing and the implementation of object recognition systems on a robot can be a challenging task. While many algorithms exist for machine vision, fewer have been developed with the efficiency required to allow real-time operation on a processor limited platform. This paper focuses on several efficient algorithms designed to identify field landmarks and objects found in the controlled environment of the RoboCup Four-Legged League.

In competitive domains, the knowledge about the opponent can give players a clear advantage. This idea lead us in the past to propose an approach to acquire models of opponents, based only on the observation of their input-output behavior. If opponent outputs could be accessed directly, a model can be constructed by feeding a machine learning method with traces of the opponent. However, that is not the case in the Robocup domain. To overcome this problem, in this paper we present a three phases approach to model low-level behavior of individual opponent agents. First, we build a classifier to label opponent actions based on observation. Second, our agent observes an opponent and labels its actions using the previous classifier. From these observations, a model is constructed to predict the opponent actions. Finally, the agent uses the model to anticipate opponent reactions. In this paper, we have presented a proof-of-principle of our approach, termed OMBO (Opponent Modeling Based on Observation), so that a striker agent can anticipate a goalie. Results show that scores are significantly higher using the acquired opponent’s model of actions.

Despite efforts to design precise motor controllers, robot joints do not always move exactly as desired. This paper introduces a general model-based method for improving the accuracy of joint control. First, a model that predicts the effects of joint requests is built based on empirical data. Then this model is approximately inverted to determine the control requests that will most closely lead to the desired movements. We implement and validate this approach on a popular, commercially available robot, the Sony Aibo ERS-210A.

This paper presents a new approach to optimize gait parameter sets using evolutionary algorithms. It separates the crossover-step of the evolutionary algorithm into an interpolating step and an extrapolating step, which allows for solving optimization problems with a small population, which is an essential for robotics applications. In contrast to other approaches, odometry is used to assess the quality of a gait. Thereby, omni-directional gaits can be evolved. Some experiments with the Sony Aibo models ERS-210 and ERS-7 prove the performance of the approach including the fastest gait found so far for the Aibo ERS-210.

Generation of a sequence of behaviors is necessary for the RoboCup Humanoid league to realize not simply an individual robot performance but also cooperative ones between robots. A typical example task is passing a ball between two humanoids, and the issues are: (1) basic skill decomposition, (2) skill learning, and (3) planning to connect the learned skills. This paper presents three methods for basic skill learning (trapping, approaching to, and kicking a ball) based on optic flow information by which a robot obtains sensorimotor mapping to realize the desired skill, assuming that skill decomposition and planning are given in advance. First, optic flow information of the ball is used to predict the trapping point. Next, the flow information caused by the self-motion is classified into the representative vectors, each of which is connected to motor modules and their parameters. Finally, optical flow for the environment caused by kicking motion is used to predict the ball trajectory after kicking. The experimental results are shown and discussion is given with future issues.

Kicking the ball with high power, short reaction time and accuracy are fundamental requirements for any soccer player. Human players acquire these fine low-level sensory motor coordination abilities trough extended training periods that might last for years. In RoboCup the problem has been addressed by engineering design and acceptable, probably sub-optimal, solutions have been found. To our knowledge the automatic development of these abilities has not been yet employed. Certainly no one is willing to damage a robot during an extended, and probably violent, evolutionary learning process in a real environment. In this work we present an approach for the automatic generation (from scratch) of ball-kick behaviors for legged robots. The approach relies on the use of UCHILSIM, a dynamically accurate simulator, and the

Back to Reality

paradigm to evolutionary robotics, a recently proposed method for narrowing the difference between simulation and reality during robot behavior execution. After eight hours of simulations successful ball-kick behaviors emerged, being directly transferable to the real robot.

The joints of a humanoid robot experience disturbances of markedly different magnitudes during the course of a walking gait. Consequently, simple feedback control techniques poorly track desired joint trajectories. This paper explores the addition of a control system inspired by the architecture of the cerebellum to improve system response. This system learns to compensate the changes in load that occur during a cycle of motion. The joint compensation scheme, called Trajectory Error Learning, augments the existing feedback control loop on a humanoid robot. The results from tests on the GuRoo platform show an improvement in system response for the system when augmented with the cerebellar compensator.

Humanoid dynamic walk and kick are two main technical challenges for the current Humanoid League. In this paper, we conduct a research aiming at generating dynamically stable walking and kicking gait for humanoid soccer robots with consideration of different constraints. Two methods are presented. One is synthesizing gait based on constraint equations, which has formulated gait synthesis as an optimization problem with consideration of some constraints, e.g. zero-moment point (ZMP) constraints for dynamically stable locomotion, internal forces constraints for smooth transition, geometric constraints for walking on an uneven floor and etc. The other is generating feasible gait based on human kicking motion capture data (HKMCD), which uses periodic joint motion corrections at selected joints to approximately match the desired ZMP trajectory. The effectiveness of the proposed dynamically stable gait planning approach for humanoid walking on a sloping surface and humanoid kicking on an even floor has been successfully tested on our newly developed Robo-Erectus humanoid soccer robots, which won second place in the RoboCup 2002 Humanoid Walk competition and got first place in the RoboCup 2003 Humanoid Free Performance competition.

This paper proposes a new algorithm for the automatic segmentation of motion data from a humanoid soccer playing robot that allows feed-forward neural networks to generalize and reproduce various kinematic patterns, including walking, turning, and sidestepping. Data from a 20 degree-of-freedom Fujitsu

hoap

-1 robot is reduced to its intrinsic dimensionality, as determined by the

isomap

procedure, by means of nonlinear principal component analysis (

nlpca

). The proposed algorithm then automatically segments motion patterns by incrementally generating periodic temporally-constrained nonlinear

pca

neural networks and assigning data points to these networks in a

conquer

-and-divide fashion, that is, each network’s ability to learn the data influences the data’s division among the networks. The learned networks abstract five out of six types of motion without any prior information about the number or type of motion patterns. The multiple decoding subnetworks that result can serve to generate abstract actions for playing soccer and other complex tasks.

This paper presents a method for planning three-dimensional walking patterns for biped robots in order to obtain stable smooth dynamic motion and also maximum velocity during walking. To determine the rotational trajectory for each actuator, there are some particular key points gained from natural human walking whose value is defined at the beginning, end and some intermediate or specific points of a motion cycle. The constraint equation of the motion between the key points will be then formulated in such a way to be compatible with geometrical constraints. This is first done in sagittal and then developed to lateral plane of motion. In order to reduce frequent switching due to discrete equations which is inevitable using coulomb dry friction law and also to have better similarity with the natural contact, a new contact model for dynamic simulation of foot ground interaction has been developed which makes the cyclic discrete equations continuous and can be better solved with ODE solvers. Finally, the advantages of the trajectory described are illustrated by simulation results.

We have developed an automatic calibration method for a global camera system. Firstly, we show how to define automatically the color maps we use for tracking the robots’ markers. The color maps store the parameters of each important color in a grid superimposed virtually on the field. Secondly, we show that the geometric distortion of the camera can be computed automatically by finding white lines on the field. The necessary geometric correction is adapted iteratively until the white lines in the image fit the white lines in the model. Our method simplifies and speeds up significantly the whole setup process at RoboCup competitions. We will use these techniques in RoboCup 2004.

In order to detect objects using colour information, the mapping from points in colour space to the most likely object must be known. This work proposes an adaptive colour calibration based on the Bayes Theorem and chrominance histograms. Furthermore the object’s shape is considered resulting in a more robust classification. A randomised hough transform is employed for the ball. The lines of the goals and flagposts are extracted by an orthogonal regression. Shape detection corrects over- and undersegmentations of the colour segmentation, thus enabling an update of the chrominance histograms. The entire algorithm, including a segmentation and a recalibration step, is robust enough to be used during a RoboCup game and runs in real-time.

We investigate the application of a Monte Carlo localization filter to the problem of combining local and global observations of a small, off-the-shelf quadruped domestic robot, in a simulated

Smart House

environment, for the purpose of robust tracking and localization. A Sony Aibo ERS-210A robot forms part of this project, with the ultimate aim of providing additional monitoring, human-system interaction and companionship to the occupants.

We present a complete system for obstacle avoidance for a mobile robot. It was used in the RoboCup 2003 obstacle avoidance challenge in the Sony Four Legged League. The system enables the robot to detect unknown obstacles and reliably avoid them while advancing toward a target. It uses monocular vision data with a limited field of view. Obstacles are detected on a level surface of known color(s). A radial model is constructed from the detected obstacles giving the robot a representation of its surroundings that integrates both current and recent vision information. Sectors of the model currently outside the current field of view of the robot are updated using odometry. Ways of using this model to achieve accurate and fast obstacle avoidance in a dynamic environment are presented and evaluated. The system proved highly successful by winning the obstacle avoidance challenge and was also used in the RoboCup championship games.

In this paper we show how a combination of multiple neuromorphic vision sensors can achieve the same higher level visual processing tasks as carried out by a conventional vision system. We process the multiple neuromorphic sensory signals with a standard auto-regression method in order to fuse the sensory signals and to achieve higher level vision processing tasks at a very high update rate. We also argue why this result is of great relevance for the application domain of reactive and lightweight mobile robotics, at the hands of a soccer robot, where the fastest sensory-motor feedback loop is imperative for a successful participation in a RoboCup soccer competition.

In 2004, the playing field size of the small sized league was significantly increased, posing new challenges for all teams. This paper describes extensions to our current video server software (Doraemon) to deal with these new challenges. It shows that a camera with a side view is a workable alternative to the more expensive approach of using multiple cameras. The paper discusses the camera calibration method used in Doraemon as well as an investigation into some common two–dimensional interpolation methods, as well a novel average gradient method. It also proves that (ignoring occluded parts of the playing field) it is possible to construct a realistic top down view of the playing field with a camera that only has a side view of the field.

Multiple cameras have been used to get a view of a large area. In some cases, the cameras are placed so that their views are overlapped to get a more complete view. 3D information of the overlapping areas that are covered with two or three cameras can be obtained by stereo vision methods. By shifting the shutter timings of cameras and using our pseudo stereo vision method, we can output 3D information faster than 30 fps. In this paper, we propose a pseudo stereo vision method using three cameras with different shutter timings. Using three cameras, two types of shutter timings are discussed. In three different shutter timings, 90 points of 3D position for a sec are obtained because the proposed method can output 3D positions at every shutter timing of three cameras. In two different shutter timings, it is possible to calculate the 3D position at 60 fps with better accuracy.

This paper describes a new technique for determining the distance to a planar surface and, at the same time, obtaining a characterization of the surface’s material through the use of conventional, low-cost infrared sensors. The proposed technique is advantageous over previous schemes in that it does not require additional range sensors, such as ultrasound devices, nor a priori knowledge about the materials that can be encountered. Experiments with an all-terrain mobile robot equipped with a ring of infrared sensors are presented.

In this work, a novel search strategy for autonomous search and rescue robots, that is highly suitable for the environments when the aid of human rescuers or search dogs is completely impossible, is proposed. The work area for a robot running this planning strategy can be small voids or possibly dangerous environments. The main goal of the proposed planning strategy is to find victims under very tight time constraints. The exploration strategy is designed to improve the success of the main goal of the robot using specialized sensors when available. The secondary goals of the strategy are avoiding obstacles for preventing further collapses, avoiding cycles in the search, and handling errors. The conducted experiments show that the proposed strategies are complete and promising for the main goal of a SR robot. The number of steps to find the reachable victims is considerably smaller than that of the greedy mapping method.

This paper proposes a novel approach for a Constructive Self-Organizing Map (SOM) based world modeling for search and rescue operations in disaster environments. In our approach, nodes of the self organizing network consist of victim and waypoint classes where victim denotes a human being waiting to be rescued and waypoint denotes a free space that can be reached from the entrance of debris. The proposed approach performed better than traditional self-organizing maps in terms of both the accuracy of the output and the learning speed. In this paper the detailed explanation of the approach and some experimental results are given.

The RoboCupRescue Simulation project aims at simulating large-scale disasters in order to explore coordination strategies for real-life rescue missions. This can only be achieved if the simulation itself is as close to reality as possible. In this paper, we present a new fire simulator based on a realistic physical model of heat development and heat transport in urban fires. It allows to simulate three different ways of heat transport (radiation, convection, direct transport) and the influence of wind. The protective effects of spraying water on non-burning buildings is also simulated, thus allowing for more strategic and precautionary behavior of rescue agents. Our experiments showed the simulator to create realistic fire propagations both with and without influence of fire brigade agents.

We address the problem of merging multiple noisy maps in the rescue environment. The problem is tackled by performing a stochastic search in the space of possible map transformations, i.e. rotations and translations. The proposed technique, which performs a time variant Gaussian random walk, turns out to be a generalization of other search techniques like hill-climbing or simulated annealing. Numerical examples of its performance while merging partial maps built by our rescue robots are provided.

Orpheus mobile robot is a teleoperated device primarily designed for remote exploration of hazardous places and rescue missions. The robot is able to operate both indoors and outdoors, is made to be durable and reliable. The robot is remotely operated with help of visual telepresence. The device is controlled through advanced user interface with joystick and head mounted display with inertial head movement sensor. The functionality and reliability of the system was tested on Robocup Rescue League 2003 world championship in Italy where our team placed on 1st place.

In this paper, the problem of determining if a population of mobile robots is able to travel from an initial configuration to a target configuration is addressed. This problem is related with the controllability of the automaton describing the system. To solve the problem, the concept of navigation automaton is introduced, allowing a simplification in the analysis of controllability. A set of illustrative examples is presented.

This paper describes the joint approach of three research groups to enable a heterogeneous team of robots to exchange belief. The communication framework presented imposes little restrictions on the design and implementation of the individual autonomous mobile systems. The three groups have individually taken part in the RoboCup F2000 league since 1998. Although recent rule changes allow for more robots per team, the cost of acquiring and maintaining autonomous mobile robots keeps teams from making use of this opportunity. A solution is to build mixed teams with robots from different labs. As almost all robots in this league are custom built research platforms with unique sensors, actuators, and software architectures, forming a heterogeneous team presents an exciting challenge.

This paper introduces a general formulation of relational behaviours for cooperative real robots and an example of its implementation using the pass between soccer robots of the Middle-Sized League of RoboCup. The formulation is based on the Joint Commitment Theory and the pass implementation is supported by past work on soccer robots navigation. Results of experiments with real robots under controlled situations (i.e., not during a game) are presented to illustrate the described concepts.

In this paper we focus on plan execution in highly dynamic environments. Our plan execution procedure is part of a high-level planning system which controls the actions of our RoboCup team ”Mostly Harmless”. The used knowledge representation scheme is based on traditional STRIPS planning and qualitative reasoning principles. In contrast to other plan execution algorithms we introduce the concept of plan invariants which have to be fulfilled during the whole plan execution cycle. Plan invariants aid robots in detecting problems as early as possible. Moreover, we demonstrate how the approach can be used to achieve cooperative behavior.

We are exploring the use of auction mechanisms to assign roles within a team of agents operating in a dynamic environment. Depending on the degree of collaboration between the agents and the specific auction policies employed, we can obtain varying combinations of role assignments that can affect both the speed and the quality of task execution. In order to examine this extremely large set of combinations, we have developed a theoretical framework and an environment in which to experiment and evaluate the various options in policies and levels of collaboration. This paper describes our framework and experimental environment. We present results from examining a set of representative policies within our test domain — a high-level simulation of the RoboCup four-legged league soccer environment.

Object recognition systems contain a large amount of highly specific knowledge tailored to the objects in the domain of interest. Not only does the system require information for each object in the recognition process, it may require entirely different vision processing techniques. Generic programming for vision processing tasks is hard since systems on-board a mobile robots have strong performance requirements. Such issues as keeping up with incoming frames from a camera limit the layers of abstraction that can be applied. This results in software that is customized to the domain at hand, that is difficult to port to other applications and that is not particularly robust to changes in the visual environment.

In this paper we describe a high level object definition language that removes the domain specific knowledge from the implementation of the object recognition system. The language has features of object-orientation and logic, being more declarative and less imperative. We present an implementation of the language efficient enough to be used on a Sony AIBO in the Robocup Four-Legged league competition and several illustrations of its use to rapidly adjust to new environments through quickly crafted object definitions.

The existing reinforcement learning approaches have been suffering from the policy alternation of others in multiagent dynamic environments such as RoboCup competitions since other agent behaviors may cause sudden changes of state transition probabilities of which constancy is necessary for the learning to converge. A modular learning approach would be able to solve this problem if a learning agent can assign each module to one situation in which the module can regard the state transition probabilities as constant. This paper presents a method of modular learning in a multiagent environment, by which the learning agent can adapt its behaviors to the situations as results of the other agent’s behaviors. Scheduling for learning is introduced to avoid the complexity in autonomous situation assignment.

An object recognition process in general is designed as a domain specific, highly specialized task. As the complexity of such a process tends to be rather inestimable, machine learning is used to achieve better results in recognition. The goal of the process presented in this paper is the computation of the pose of a visible robot, i. e. the distance, angle, and orientation. The recognition process itself, the division into subtasks, as well as the results of the process are presented. The algorithms involved have been implemented and tested on a Sony Aibo.

We show that previously published visual data processing methods for the simulated robotic soccer so far have not been utilizing all available information, because they were mainly based on heuristic considerations. Researchers have approached to estimating the agent location and orientation as two separate tasks, which caused systematic errors in the angular measurements. Further attempts to get rid of them (e.g. by completely neglecting the angular data) only aggravated the problem and resulted in the losses in the accuracy. We utilize all the potential of the visual sensor by jointly estimating the agent view direction angle and Cartesian coordinates using the extended Kalman filtering technique. Our experiments showed that the achievable average error limit for this particular application is about 25-33 per cent lower than that of the best algorithms published by far.

Path planning technique is proposed in the paper. It was developed for robots with differential drive, but with minor modification could be used for all types of nonholonomic robots. The path was planned in the way to minimize the time of reaching end point in desired direction and with desired velocity, starting from the initial state described by the start point, initial direction and initial velocity. The constraint was acceleration limit in tangential and radial direction caused by the limited grip of the tires. The path is presented as the spline curve and was optimised by placing the control points trough which the curve should take place.

Fast and accurate self-localization is one of the most important problems in autonomous mobile robots. In this paper, an analysis by synthesis method is presented for optimizing the self-localization procedure. In the synthesis phase of this method, the robot’s observation of the field is predicted using the results of odometry. It is done by calculating the position of the landmarks on the captured image. In the analysis phase, the local search algorithms find the exact position of the landmarks on the image from which the best matching coordinates of the robot are determined using a likelihood function. The final coordinates of the robot are then obtained from the odometry sensor, using an integrated delay compensation and correction technique. Experimental results show that precise and delay-free results are achieved with a very low computational cost.

In this study, a new method called Reverse Monte Carlo Localization (R-MCL) for global localization of autonomous mobile agents in the robotic soccer domain is proposed to overcome the uncertainty in the sensors, environment and the motion model. This is a hybrid method based on both Markov Localization(ML) and Monte Carlo Localization(MCL) where the ML module finds the region where the robot should be and MCL predicts the geometrical location with high precision by selecting samples in this region. The method is very robust and fast and requires less computational power and memory compared to similar approaches and is accurate enough for high level decision making which is vital for robot soccer.

Soccer simulation commentary system is a suitable test bed for exploring

real time systems

. The

rapidly changing

simulation environment requires that the system generates real time comments based on the information received from the

Soccer Server

. In this article, a three-layer architecture of Caspian Soccer Commentary system is presented, and each component of the system is briefly described. The emphasis of this paper is on design and implementation of the

Analyzer

and the

Content Selector

subsystems. The Analyzer takes advantage of the

State Machine

to keep track of the game situations. The

Scheduling

and

Interruption

mechanism is proposed to improve the efficiency of the Content Selector subsystem. The presented Commentary System together with the other Caspian presentation and analysis tools won the first place in RoboCup 2003 Game Presentation and Match Analysis competitions.

The paper discusses a top-down approach to model soccer knowledge, as it can be found in soccer theory books. The goal is to model soccer strategies and tactics in a way that they are usable for multiple RoboCup soccer leagues, i.e. for different hardware platforms. We investigate if and how soccer theory can be formalized such that specification and execution is possible. The advantage is clear: theory abstracts from hardware and from specific situations in leagues. We introduce basic primitives compliant with the terminology known in soccer theory, discuss an example on an abstract level and formalize it. We then consider aspects of different RoboCup leagues in a case study and examine how examples can be instantiated in three different leagues.

Motion detection and tracking while moving is a desired ability for any soccer player. For instance, this ability allows the determination of the ball trajectory when the player is moving himself or when he is moving his head, for making or planning a soccer-play. If a robot soccer player should have a similar functionality, then it requires an algorithm for real-time movement analysis and tracking that performs well when the camera is moving. The aim of this paper is to propose such an algorithm for an AIBO robot. The proposed algorithm uses motion compensation for having a stabilized background, where the movement is detected, and Kalman Filtering for a robust tracking of the moving objects. The algorithm can be adapted for almost any kind of mobile robot. Results of the motion detection and tracking algorithm, working in real-world video sequences, are shown.

This paper describes methods used in stabilizing the walking gait of

Tao-Pie-Pie

, a small humanoid robot given rate feedback from two RC gyroscopes.

Tao-Pie-Pie

is a fully autonomous small humanoid robot (30cm tall). Although

Tao-Pie-Pie

uses a minimal set of actuators and sensors, it has proven itself in international competitions, winning honors at the RoboCup and HuroSot competitions in 2002 and 2003. The feedback control law is based solely on the rate information from two RC gyroscopes. This alleviates drift problems introduced by integrating the RC gyroscope feedback in the more common position control approaches.

The UT Austin Villa 2003 simulated online soccer coach was a first time entry in the RoboCup Coach Competition. In developing the coach, the main research focus was placed on treating advice-giving as a machine learning problem. Competing against a field of mostly hand-coded coaches, the UT Austin Villa coach earned first place in the competition. In this paper, we present the multi-faceted learning strategy that our coach used and examine which aspects contributed most to the coach’s success.

ITAS is a tool that allows a human to play soccer in the RoboCup Soccer Simulator environment. This is essentially the reverse challenge to that of RoboCup. Instead of bringing the machine to the real world, ITAS strives to seamlessly interface man to the machine world. This presents a fundamental human-computer interaction design problem. This paper shows how the reverse RoboCup challenge can benefit the RoboCup community and what value it brings to robotics and AI research in large. An overview of the features of ITAS and its development using the Usability Engineering Lifecycle are then given, followed by a comparison with a related system, OZ-RP. ITAS is an open source project. The most recent releases are available at http://itas.sourceforge.net.

This article describes a software development toolkit for programming mobile robots, that has been used on different platforms and for different robotic application. In this paper we address design choices, implementation issues and results in the realization of our robot programming environment, that has been devised and built from many people since 1998. We believe that the proposed framework is extremely useful not only for experienced robotic software developers, but also for students approaching robotic research projects.

In the effort to meet the steadily changing demands of teaching computer science and computer engineering, new methods of learning and teaching are used by which multifarious knowledge and learning techniques can be imparted, practical skills and abilities can be developed and teamwork and creativity are encouraged. A promising attempt is the use of robotic construction kits.

This paper portrays the educational environment that was used at the courses “Hamburg RoboCup: Mobile autonomous robots play soccer” at the University of Hamburg, Germany and “Advanced robotics – Soccer playing mobile autonomous robots” at the California State University of Chico, USA and compares the experiences made during both courses due to intercultural differences.

Component-based software engineering is useful for embedded applications such as robotics. However, heavyweight component systems such as CORBA overstrain the ressources available in many embedded systems. Here, a lightweight component-based approach is used to implement the system software of the so-called CubeSystem, CubeOS. Since 1998, CubeOS and its component system have been successfully used in various areas from industry projects over RoboCup-related research to edutainment applications. Many of the components used in RoboCup soccer have been carried over in the implementation of the IUB Rescue robots, demonstrating the potential for software reuse.