Proceedings of the 3rd International Symposium on Autonomous Minirobots for Research and Edutainment (AMiRE 2005)

We developed a new robotic systems that can change morphology. The system is made of ATRON modules, which are individually simple, attach through physical connections, and perform 3D motions by collective actions. We produced 100 ATRON modules, and performed both simulation and real world experiments. In this paper, we report on the ATRON hardware design and investigations related to the verification of the suitability of the ATRON module design for self-reconfigurable robotics.

This paper presents an extension turret for efficient data processing for autonomous minirobots, in particular the minirobot Khepera. The Khepera robot is a base platform of a mobile autonomous system for our experiments, but the robot’s main processor cannot accomplish complex processing tasks with real-time requirements, because of the processor’s low performance. Thus, we have designed an extension hardware module based on a field programmable gate array (FPGA) to run the real-time processing and concurrently control other peripheral modules. The module with coupling of FPGA and a microcontroller gains robot’s ability and possesses FPGA’ flexibility of reconfiguration.

We show that a map of empty space around objects near a mobile camera in an unknown environment can be obtained from a naive light field model of image data. We observe that, between the nearest objects, cells with different light in different directions correspond to empty space. We describe experiments with a robot in an environment with black objects and white objects seen against a black and white background. As there are no absolute silhouettes in this environment, ordinary silhouette methods cannot be used. We demonstrate that the robot is able, in real-time, to find and pass through a gap in a row of objects or stop if no gap is wide enough.

Climbing is a challenging task for autonomous mobile robots primarily due to requirements for agile locomotion, and high maneuverability as well as robust and efficient attachment and detachment. A novel miniature wall-climbing robot is proposed. The robot is adapted for the wall-climbing task by taking advantage of down scaling and its low design. Challenges encountered during robot miniaturization and performances of the robot are reported. The miniature robot prototype proved to be able to climb on inclined surfaces with a slope of up to 90° at a speed of 3.3mm/s. It is equipped with sensors that enable it to avoid obstacles, follow walls and detect free-falls. It can be controlled by remote control or act autonomously. Animals, such as Geckos, have developed amazing climbing ability through micro- and nano-fibers on their feet. These structures have inspired the study of dry adhesion and the design of synthetic fibrillar pads presented in the paper.

This paper describes the Teleworkbench system, a teleoperated platform for managing experiments in multi-robotics. This testbed is intended as a benchmark tool for managing and analyzing experiments involving one or many minirobots. The Teleworkbench is connected to the internet to allow easy access for remote users located in any part of the world. Three unique features of our testbed are the ability to download a user-defined program for each robot, the support for running 36 robots simultaneously, and the visualization of occurring events during the experiments.

A web service is a web-based application that allows client programs to access its functionality using open, standardized protocols. This approach ensures interconnectivity and interoperability so that the functionality of the web service can be accessed by virtually any client, regardless of location and programming language or platform. Autonomous mobile robots have the ability to perceive and physically interact with the real world. The ability of physical manipulation would fundamentally extend the service capability of web services. On the other hand, by equipping mobile autonomous robots with web service interfaces, a standardized, interoperable way for world-wide access to the robots could be provided. In this project we study the suitability of the web service standards and technologies for the teleoperation of mobile autonomous robots. As project result we provide access to a mobile autonomous robot via a web service interface.

This paper describes a communication and control infrastructure for autonomous mini-robots, which is based on Bluetooth wireless communication technology. In mobile robots, one of the most important issues is power consumption which limits the lifetime of the robot. In this respect, for communication, a technology that requires the least energy is the most appropriate choice. We have developed a Bluetooth stick for the mini-robot Khepera to enable simple point-to-point wireless connectivity. Using this stick, one can establish communication between one master and seven slaves, called a piconet. We have also extended the networking capabilities to establish Bluetooth scatternets through hardware enrichment. The described infrastructure in general can be applied to all kinds of autonomous robots.

In this paper, we develop a goal keeper for simulated RoboCup soccer. The developed agent is an improved version of UvA Base team. We rst explain RoboCup and UvA Base team that is used as a base team. We also point out several flaws in the original keeper of the UvA Base team. Then, we describe how our keeper is developed. Finally, we show through computational experiments the effectiveness of the developed keeper.

This paper proposes an action rule discovery technique from simulated RoboCup soccer logs. Our proposed technique analyzes log files of past games and generates effective action rules for agents. From simulation results, we show the effectiveness of our action rule discovery technique.

In this paper we describe adapting recognition of shootable situations for an agent in a RoboCup soccer simulation game. An agent needs to adapt its recognition of shootable situations to their opponents in a game, because the recognition whether a shot will succeed or not depends on abilities of opponents’ interception. When an agent tries to adapt by learning in a game, the agent faces a problem of a limitation of its shooting chances. We apply LEO (Learning from Experience and Observation) to the agents to let them increase their learning opportunities indirectly by seeing teammate agents’ shots. LEO is a learning method for multi-agent environments which we proposed. LEO consists of “Learning from Observation” (LO) and “Learning from Experience” (LE). In the experiments of RoboCup Soccer Simulation games, the agents with the LEO can improve a success rate of “shooting” action to 0.12 from 0.04 (non-learning) and 0.06(LE only).

This paper presents the current efforts and ideas of members in the RoboCup Simulation and the Humanoid Leagues to take successful concepts from both environments and extend them in ways so that each league can profit from the results. We describe the ongoing development of the 3D simulator which is being extended to simulate a real humanoid robot. At the same time, we give an insight into the current behavior development framework of the Humanoid League team

Senchans

which makes heavy use of techniques which have been successfully used in the Simulation League before. Furthermore, we give some suggestions for a collaboration between the different leagues in the RoboCup from which all the participants could benefit.

Machine-learning algorithms Reinforcement Learning is one of powerful machine learning algorithms. We have known that this algorithm is the effect of robotic learning and that a lot of researchers have proposed the basic idea using this algorithm. A system in which the algorithm has been built is needs a lot of trials and errors. Moreover it requires a huge amount of calculation in order to achieve some effectiveness. This paper has proposed a method using reinforcement learning with hand-coded rules for a soccer agent in Ro-boCup Soccer Simulation League. We have analyzed a lot of scenes that agents have made scores in simulation soccer games and have extracted some rules. We have called them “hand coded rales”. Moreover we have applied our method for an offensive soccer agent in RoboCup Soccer Simulation League and have done some experiments. From the results, we have confirmed that our team has improved its capability of getting scores.

Recently, student teams in Yumekobo, the factory for dreams and ideas, in Kanazawa Institute of Technology obtain remarkable results in various domestic and world championships. As an example, the WinKIT team in the middle-size league won the vice champion at the RoboCup 2002, 2003, and 2004 world championship. The educational system in Yumekobo attracts attention very much among people who engage in the educational system reform in many universities in Japan. Yumekobo is not also the place that only learns knowledge and skill but the place of character building. Students develop their good character, which include independence, creativity, morals, teamwork, international nature and so on.

Q-learning in the Reinforcement Learning (RL) field is the powerful and attractive tool to make robots generate autonomous behavior. But it needs large amount of computational cost because of its discrete state and action. To generated smooth trajectory with less computational cost, we propose two ingredients for Q-learning. We applied Q-learning to the simulated two wheeled robot to generate trajectory for Ball-To-Goal task in robot soccer. …

In this research, we propose a multiple omnidirectional vision system (MOVIS) with three omnidirectional cameras and its calculation method for the measurement of the object position and the self-localization in an autonomous mobile robot. On the identification of the self-position, we try to improve the accuracy of the measurement by correcting the absolute position based on the measurement error of landmarks in the origin of the absolute coordinate. Furthermore, we propose the omnidirectional behavior control method for collision avoidance and object chasing motion by using fuzzy reasoning in an autonomous mobile robot with MOVIS. In this paper, we also report some experimental results to confirm the efficiency of the proposed method by using a soccer robot in dynamic environment.

Object detection on colour images is often performed by specialised microchips. A disadvantage of these implementations is often high power consumption and due to their specialised nature the chip usually can not be used for other tasks. By parallelising the image processing on FPGA hardware, the calculational intensive task can be executed very fast at reduced power consumption. In this paper we focus on the FPGA implementation of an algorithm for object detection in robot soccer application.

Robot soccer fosters AI and intelligent robotics research by providing a standard problem where a wide range of technologies can be integrated and examined. In order for a robot team to actually perform in a soccer game, various technologies must be incorporated including: design principles of autonomous agents, multi-agent collaboration, strategy acquisition, real-time reasoning, robotics, and sensor-fusion. In this paper, we discuss the speci c features of the KheperaSot league, describe the winner of the last two World Cups, and discuss the future directions of the KheperaSot league.

This paper proposes a coevolutionary algorithm for acquiring cooperative behavior of soccer robots. Fuzzy control is applied to the control of the soccer robots as local behavior, while the target point generation is used for global behavior. Furthermore, a classifier system is applied for generating a target point for the global behavior in the current situation. We show simulation results of cooperative behavior acquisition of soccer robots.

This study focuses on the relation between contour integration process and figure-ground process, both of them are fundamental functions of the visual system in the human brain. We executed a set of psychological experiments using stimuli with small triangles which have the attribute of figure. We obtained the results that contour saliency is elevated when triangles are aligned so that edges of them are on a smooth contour and figural side of those edges are identical against the smooth contour. Those results indicate that the processes of contour integration and figure-ground separation are not one-way cascade manner, but the reciprocal and interactive one.

In order to study neuronal basis to extract Angles and junctions embedded within contours, we conducted extracellular recordings, while two macaque monkeys performed the fixation task. Angle stimuli were the combination of two straight half lines. Each line was drawn from the center to outside the classical receptive fields in one of 12 directions. In the superficial layer of area V2, 91 out of 114 neurons showed selective responses to these angle stimuli. Of these, 41 neurons (36%) showed selective responses to wide angles between 60° and 150°, which were distinct from responses to straight lines or sharp angles (30°). When we tested these neurons’ selectivity to the 12 individual half lines, the preferred directions were more or less consistent with one or two components of the optimal angle stimuli. These results suggest that the angle selectivity in area V2 was largely dependent on the representation of individual line components. We suggest that area V2 is the first step to extract information of angles embedded within contour stimuli.

Although the integration of depth cues has been studied electrophysiologically and psychophysically, it is still not clear what macroscopic brain activity is correlated with the integration of depth cues. In the present study, magnetoen-cephalography (MEG) related to the integration of two important depth cues, motion parallax and binocular disparity, was recorded. Visual stimulus was a rotating random-dot cylinder in the right hemifield. There were three stimulus conditions LEFT, RIGHT and INCOHERENT. In the LEFT condition, dots moving leftward and those moving rightward had crossed and uncrossed disparity, respectively. In the RIGHT condition, dots moving rightward and those moving leftward had crossed and uncrossed disparity, respectively. In the INCOHERENT condition, 50% of all the dots had crossed disparity and the rest half had uncrossed disparity. A peak of the root mean square (RMS) amplitude of MEG responses around 180 ms after the onset of rotation was observed from nine subjects in all of the three conditions. The RMS amplitude of the peak was significantly smaller in the INCOHERENT condition than that in the LEFT condition. The results are partly consistent with the motion-disparity interaction revealed by a previous electrophysiological study.

We investigated responses of area CIP of macaque monkey to a set of 3D curvatures defined by random-dot stereogram (RDS). In 82 neurons recorded, 75 (91.5 %) had significant response to one of curvatures tested. Most of visually responsive neurons showed preference to curvature with convex or concave. To evaluate the strength of tuning to the shape, modulation index (MI) was computed for the responses of each neuron to the curvatures, and neuron with MI more than 0.6 was defined as ‘curvature selective neuron’. Of 75 visually responsive neurons, 20 (26.7 %) met the criterion. Five neurons had highly selective response to a particular curvature (MI >= 0.8). Then we plotted the preferred shape index of the curvature selective neurons computed from Gaussian function fit to the responses of neurons. The distribution of the preferred shape index was biased toward concave and convex ellipsoid, although it distributed over the range of shape index (−1.0 − 1.0). These results suggest that 3D curvatures might be represented in area CIP and this area plays a critical role in 3D vision.

TAM (Topographic Attentive Mapping) network is a biologically-motivated neural network. In this paper, we define the Gabor function type receptive field and incorporate it to the TAM network’s feature layer. We also discuss the orientation selectivity of the receptive field through some examples of character recognition.

A new method to recognize corridors for autonomous mobile robot is described. The method, based on the both genetic-algorithm and adaptive matching of surface-strip model length, detects the edge markers of a corridor. To recognize the branch of corridor, the surface-strip model with layered structure is proposed. The experimental results show the method is effective for real-time recognition of corridor and its branch.

To operate successfully in any environment, mobile robots must be able to localize themselves accurately. In this paper, we describe a direct method (in the sense it does not use an iterative search) based on vision for localizing a mobile robot in an environment with only two observations along a linear trajectory. We only assume that the robot can visually identify landmarks and measure their bearings. Contrary to other existing approaches to landmark based navigation, we do not require any other sensors (like range sensors or wheel encoders) or the prior knowledge of relative distances between the landmarks. Given its low cost, the range of potential applications of our localization system is very wide. In particular, this system is ideally suited for domestic robots such as autonomous lawn-mowers and vacuum cleaners.

Two key objectives of robot vision are autonomous navigation and mapping. Digital cameras have become relatively cheap in recent years and have appeared in a variety of consumer devices, such as PDAs (Personal Digital Assistants). It therefore makes sense to try to build systems that use vision as their primary input instead of the more traditional sonar and infrared sensors that have been used in the past. The camera can also be used for a range of other tasks. Service robots and toys typically operate in an indoor environment but rarely have a map of their environment when they are first turned on. This paper therefore addresses the problem of vision-based mapping where a camera has deliberately been chosen as the only sensor.

This paper presents a prediction method of future position of a fish using a Neural Network. Our previous experiments to catch fishes using a robot with visual feedback has shown that even a small fish can find some ways by itself to escape from catching net. So we started a research to make more intelligent robot than a fish. As the first step, we tried to predict its future position using a neural network. Experimental results using a swimming real fish show that the proposed method can predict the motion of fish.

Visual homing is a navigational technique allowing an agent to return to a location which it has visited previously, using visual information alone. Traditional visual homing techniques eschew goal distance information. We show that, using our novel algorithm, such information is relatively easy to acquire. We then show that the use of such goal distance information reduces the computational cost of homing.

This paper presents the

Smarty Board

; a new micro-controller board designed specifically for the robotics teaching needs of Australian schools. The primary motivation for this work was the lack of commercially available and cheap controller boards that would have all their components including interfaces on a single board. Having a single board simplifies the construction of programmable robots that can be used as platforms for teaching and learning robotics. Reducing the cost of the board as much as possible was one of the main design objectives. The target user groups for this device are the secondary and tertiary students, and hobbyists. Previous studies have shown that equipment cost is one of the major obstacles for teaching robotics in Australia.

The new controller board was demonstrated at high-school seminars. In these demonstrations the new controller board was used for controlling two robots that we built. These robots are available as kits. Given the strong demand from high-school teachers, new kits will be developed for the next robotic Olympiad to be held in Australia in 2006.

There is little software and hardware support for the structure of computations in robots, which is inherently parallel and distributed. This article describes the

Agora

distributed, parallel software/hardware computing architecture primarily intended for autonomous robotics, and how the software part of the architecture can be realised with simple extensions to object oriented programming. The Squeak open source Smalltalk implementation is used for this purpose. The advantages of the Smalltalk environment for robot software development are discussed. In addition we describe the use of Smalltalk as a vehicle for robotics education.

This paper reports education for creativity in the Department of Human and Artificial Intelligent Systems (HAIS), Faculty of Engineering, Fukui University.

Students will train their creativity through the designing and making line trace robots while they are studying the drawing, electrical circuit design, and computer programming, and so on.

This paper describes the education curriculum of “humanoid” class using a small size humanoid robot and introduces the educational tools which helps these kinds of curriculms. “MONOZUKURI” is one of the most important factors in education field, this class also study “MONOZUKURI” through a humanoid robot.

Purpose of this study is to develop self-organization algorithm of spiking neural network applicable to autonomous robots. We first formulated a spiking neural network model whose inputs and outputs were analog. We then implemented it into a miniature mobile robot Khepera. In order to see whether or not a solution(s) for the given task exists with the spiking neural network, the robot was evolved with the genetic algorithm (GA) in an environment. The robot acquired the obstacle-avoidance and navigation task successfully, exhibiting the presence of the solution. Then, a self-organization algorithm based on the use-dependent synaptic potentiation and depotentiation was formulated and implemented into the robot. In the environment, the robot gradually organized the network and the obstacle avoidance behavior was formed. The time needed for the training was much less than with genetic evolution, approximately one fifth (1/5).

In this paper, the performance of the utility function method for behavioral organization is investigated in the framework of a simple guard robot. In order to achieve the best possible results, it was found that high-order polynomials should be used for the utility functions, even though the use of such polynomials, involving many terms, increases the running time needed for the evolutionary algorithm to find good solutions.

Evolutionary robotics is a useful technique for the automatic creation of autonomous robots. However, to use them in actual environment, we have to solve a problem that they might occur the unexpected behavior in phase period during evolution. In this paper, we describe a new method of artificial evolution with which autonomous robot acquires safe and flexible actions. Analysis of sensory information of robots under various conditions revealed that the certain pattern between robot’s behavior and the flow of sensory information, and this pattern is unique under each condition. By memorizing such patterns during move in the maze field and learning a pattern that lead to crash to the wall, the robot obtains an ideal action evolutionally without clashing to the wall even in phase period during evolution. Thus, this method may have an advantage for evolution of the robot without occurring the unexpected dangerous behavior.

We propose a design methodology for self-replicating robots and describe two robots we have designed using our methodology. We specify our target system as a mechanically autonomous team of remote control Lego robots capable of constructing all members of the team from individual pieces of Lego. Each member is responsible for a specific type of work and the design process proceeds by validating sub-teams. To validate a candidate sub-team we demonstrate that each member of the sub-team is capable of doing all the work of its type required for the construction of each member of the sub-team. Other types of work, which would be done by other members of a complete team, are done manually. This methodology allows us to face the essential difficulties of the self-referential design task without building a complete system. We describe the design and validation of a two-member sub-team including a tapper robot, responsible for piece-connection work, and a vehicle robot, responsible for robot-positioning work.

In this paper, an environment for prototyping algorithms for the autonomous intersection management of vehicles is presented. It is based on a colony of Khepera mini robots that cooperate via radio communication to enable a collision-free passing of an intersection. The design of the environment, a simple distributed algorithm, and results from first experiments are described. Compared to other work, multiple robots can pass the intersection simultaneously.

This paper deals with a cooperation form of highly directed multi-agent systems (DMAS), where a coordinated play takes more important role than conventional MAS. We propose a concept of subjective distance, which is a sense of distance subjectively perceived by a directed agent. This paper formulates the subjective distance and empirically verifies proposed concept with actual canoe polo players.

This paper deals with a crack detection method for preventive maintenance of power plant. In order to detects a crack automatically, we develop a mobile robot with ECT (Eddy Current Testing) sensor composed of a transmitter-receiver coil probe. Here, ECT is a major technique for detection of cracks found out in structures of the power plant. The robot detects a crack by ECT sensor signals and localizes the crack by the position of the robot. Then, the shape of crack is reconstructed by tracing crack. For showing the effectiveness of this method, we shows some results for 2 types of crack.

A formation of path from a nest to a feeding area by ants was investigated by using small robots. An ant communicates the route from the nest to feeding area to the other ants by a pheromone. When an ant discovers food, it goes back marking a pheromone on its way to the nest. Its fellows follow the trace of the pheromone and then they also mark the pheromone on their way to the nest after arriving at the feeding area. At the early stage, the route is a complicated line. A lot of ants’ coming and going optimize the route from the nest to the feeding area. We observed the behavior of ants. Based on the observation we designed an ant type robot. Then we investigated the formation process of the route from the nest to the feeding area by artificial ants.

Multi-robot systems have been studied in tasks that require the robots to be physically linked. In such a configuration, a group of robots may navigate a terrain that proves too difficult for a single robot. On the contrary, many collective tasks can be accomplished more efficiently by a group of independent robots. This paper is about

swarm-bot

, a robotic system that can operate in both configurations and autonomously switch from one to the other.

We examine the performance of a single robot and of groups of robots self-assembling with an object or another robot. We assess the robustness of the system with respect to different types of rough terrain. Finally, we evaluate the performance of swarms of 16 physical robots.

At present, for self-assembly in autonomous, mobile robotics,

swarm-bots

is the state of the art for what concerns reliability, robustness and speed.

A long-term experiment of robot assisted activity for elderly people has been conducted at a health service facility for the aged since Aug. 2003. Three therapeutic seal robots, Paro, were introduced there. This paper describes the results of the experiment for one year. Face scales that consist of illustrations of person’s faces were used to evaluate a person’s moods. In addition, the interaction between elderly people and seal robots was analyzed. As a result, their moods were improved by interaction with the seal robots, and frequency of the utterances of elderly people was increased.

This paper presents two strategies for selecting volitional facial expressions and gaze behaviors of animated agents, in the case of Online Negotiation and Soft Game Theory. It is difficult to make general models select volitional facial expressions for their use in many kinds of applications. In this paper, in order to develop these strategies, I investigated the effects of facial expressions and gaze behaviors through experiments based on Online Negotiation and Game Theory. I describe strategies by using estimations of reserve prices based on facial expressions in the case of Online Negotiation. Additionally, I deal with Soft Game Theory, which considers players’ emotions, and describe strategies for using facial expressions. We can apply these strategies to animated agents and robots as a mechanism to select facial expressions and gaze behaviors.

In this paper, we propose a Q-Learning method with the plural Q-values concerning the maximization and the minimization of rewards and punishments. We aim at the realization of a system that learns complicated emotion behaviors with the behavior selection which has positiveness and negativeness adaptively according to the situation. Furthermore, we also report the result of an experiment by computer simulation to confirm the efficiency of the proposed method.

In this paper, we present a behavior generator of a companion robot “SELF” that does not bore a human. In the simulation, a red ball and a blue ball are displayed. A human moves the red ball with an analog joystick and a PC moves the blue ball. In using this model, the impressions that the robot behaviors give to the human are examined. At first, we classify the action patterns under natural interhuman interaction. Secondly, based on these action patterns, we design a behavior generator for an autonomously learning robot, SELF, by using a statistical model: dynamic Bayesian networks. Finally, the impressions that various robot behaviors give to the human are analyzed through the Semantic Differential (SD) method. The hardware of SELF is also presented.

Up to now, the method to convert non-segmented “kana” strings to “kanji-kana” strings by looking up all possible candidates of words in the dictionary and by using 2nd-order Markov chain model of “kanji-kana” characters, has been proposed. However, this method takes a lot of time to look up all possible candidates in the dictionary of large vocabulary (e.g. four hundred and thirty thousand words). This paper proposes a new method of “kana-to-kanji” conversion. This method is composed of two processing. One is the processing to find the provisional boundaries of “kana” words. The other is the processing to convert the “kana” word strings to “kanji-kana” words strings. The proposed method is evaluated by the experiments using statistical data for 1,677 issue of daily newspaper. From the results of the experiments, it is concluded that the proposed method is superior to the conventional method in accuracy rate of “kana-to-kanji” conversion and processing time.

Recently, the new method of “kana-to-kanji” conversion using Markov chain model of words has been proposed and known to be useful. However, erroneous strings of words are still included in the sentences converted by this method. This paper evaluates these erroneous strings of words and presents the method to detect these erroneous strings of words using “yomi” information and semantic information. This method is evaluated by the experiments using statistical data for 1,677 issues of a daily newspaper.

Study on traffic flow is an attractive topic from physical viewpoint as well as from engineering viewpoint, and many researchers have studied it. Some of them have proposed a cellular-automata model of traffic flow on a trail, and showed some interesting results. Chowdhury et al. have proposed a cellular automata model motivated by the motion of ants with using pheromone, and reported that there is a critical density of elements where the traffic flow increases drastically under a certain condition. In this paper, we discuss the robustness of this phenomenon using real robot system with light sensors and device to simulate chemical signals by computer graphics.

We introduce mathematically rigorous metrics on agent experiences having various temporal horizons. Sensorimotor variables accessible to the agent are treated as information-theoretic sources modelled as random variables. The time series from the sensorimotor variables over a given temporal region for different behavioural contexts ground an agent-based view of the agent’s own experiences, and the information-theoretic differences between sensorimotor experiences induce a metric space structure on the set of the agent’s possible experiences. This could allow an autonomous mobile robot to locate and navigate between its sensorimotor experiences on a geometric landscape (an

experiential metric space

) whose points are its possible experiences of a given temporal scope and in which nearby points are similar experiences.