Kyoto University (京都大学) / 0048 / 新制・課程博士 / 博士(工学) / 甲第15664号 / 工博第3322号 / 新制||工||1501(附属図書館) / 28201 / 京都大学大学院工学研究科機械理工学専攻 / (主査)教授 椹木 哲夫, 教授 松原 厚, 教授 西脇 眞二 / 学位規則第4条第1項該当

Alors que les systèmes de reconnaissance gestuelle actuels privilégient souvent un usage intérieur, nous nous intéressons à la conception d'un système dont l'utilisation est possible en environnement extérieur et en mobilité. Notre objectif est le contrôle temps-réel d'un robot mobile dont l'usage est destiné aux fantassins débarqués. La contribution principale de cette thèse est le développement d'une chaîne de reconnaissance gestuelle temps réel, qui peut être entraînée en quelques minutes avec: un faible nombre d'exemples ("small data"); des gestes choisis par l'utilisateur; une résilience aux gestes mal réalisés; ainsi qu'une faible empreinte CPU. Ceci est possible grâce à deux innovations clés: d'une part, une technique pour calculer des distances entre séries temporelles en flux, basée sur DTW; d'autre part, une rétro-analyse efficace du flux d'apprentissage afin de déterminer les hyperparamètres du modèle sans intervention de l'utilisateur. D'autre part, nous avons construit notre propre gant de données et nous l'utilisons pour confirmer expérimentalement que la solution de reconnaissance gestuelle permet le contrôle temps réel d'un robot en mobilité. Enfin, nous montrons la flexibilité de notre technique en ce sens qu'elle permet de contrôler non seulement des robots, mais aussi des systèmes de natures différentes. / Although gesture recognition has been studied for several decades, much research stays in the realm of indoors laboratory experiments. In this thesis, we address the problem of designing a truly usable, real- world gesture recognition system, focusing mainly on the real-time control of an outdoors robot for use by military soldiers. The main contribution of this thesis is the development of a real-time gesture recognition pipeline, which can be taught in a few minutes with: very sparse input ("small data"); freely user-invented gestures; resilience to user mistakes during training; and low computation requirements. This is achieved thanks to two key innovations: first, a stream-enabled, DTW-inspired technique to compute distances between time series; and second, an efficient stream history analysis procedure to automatically determine model hyperparameters without user intervention. Additionally, a custom, hardened data glove was built and used to demonstrate successful gesture recognition and real-time robot control. We finally show this work's flexibility by furthermore using it beyond robot control to drive other kinds of controllable systems.

Denna förstudie behandlar robotisering av en process som monterar två stiftade tåtar i en plastkontakt, detta bildar ett kablage. Idag monteras tåten till plastkontakten manuellt. Förstudien undersökte om robotisering är tekniskt möjlig och ekonomiskt försvarbar. Robotiseringen sker genom att en robot fattar tag i en stiftad tåt i stiftmaskinen med ett gripdon, sedan förs den stiftade tåten in i en plastkontakt med hjälp av roboten. Stiftmaskinen är en Schleuniger 67 cc. Simuleringar av två koncept genomfördes med en mock-up av maskinen och ligger som underlag för de uppskattade cykeltiderna för den robotiserade processen. Två laborationer genomfördes för att undersöka svårigheterna vid robotiseringen av processen och om det är tekniskt möjligt. Första laborationen bestod av att utforma ett gripdon som kunde hantera produkterna utan att skada dem. Andra laborationen gav underlag för ett uppskattat kvalitetsbyte för processen.  Första laborationen bestod av hur det konstruerade gripdonet skulle kunde fatta tag i den stiftade tåten utan att skada tåten eller ändra dess position eller orientering. Genomförandet av laborationen var i enlighet med ``try and fail`` metoden. Laborationen gav riktlinjer avseende på konstruktionen av gripdonet och vilket arbetstryck som är lämpligt för operationen. För högt arbetstryck resulterade i att gripdonet skadade tåten, för lågt arbetstryck resulterade i att friktionskraften mellan tåten och gripdonet blev för låg för att stiftet skulle föras in i kontakten. Andra laborationen bestod av att utröna ett approximativt värde av den robotiserade processens kvalitetsutbyte. I denna laboration utfördes 65 stycken försök till att bilda ett kablage med roboten och antalet godkända noterades. Resultatet från laborationen med de berörda produkterna som används idag vid montering av ett kablage gav kvalitetsutbytet 51%. Företaget som skall eventuellt investera har krav på att kvalitetsutbytet är 95% för att robotisering av processen skall genomföras. En omfattande analys av resultatet från denna laboration ger indikationer på att kravet kan uppnås med vidare produktutveckling av berörda produkter och gripdon.   En ekonomisk analys i form av en pay-off kalkyl ligger som underlag för att bedöma om robotiseringen är ekonomiskt försvarbar, och gav ett approximativt värde för hur mycket förädling av produkten som den robotiserade processen skall ta över från den manuella monteringen. Den förädling som görs av den robotiserade processen kallas i denna förstudie för automatiseringsgrad. Analys av pay-off kalkylen, produkterna samt resultat från laborationerna, anser författaren att robotisering av processen är teknisk möjligt och ekonomiskt försvarbart med de approximativa värdena för automatiseringsgrad på 22 % av den manuella monteringen och med 95% kvalitetsutbyte. / This pilot study investigated the robotization of a process that connects two pined conductors to a plastic contact and thereby forming a cabling. The pilot study examined if the process is technically possible and economically feasible to robotize. Today the pined conductors are attached to the cabling manually. Introduction of robotization is done so that a robot seizes one pined conductor at a time in the pin machine (Schleuniger 67) with a gripper and inserts same into the plastic contact. To determine if the process is technical possible to robotize two experiments were conducted with a gripper that was designed for this purpose. The purpose of the first was to investigate how the gripper could engage the pined conductor without damaging same or change its position or orientation. The experiment was carried out as a “try and fail” type. The experiment resulted in guidelines regarding the design of the gripper and which working pressure was recommendable for the process. A too high pressure could damage the pined conductor and a too low would cause the friction force between the conductor and the gripper to be too low to insert the pined conductor into the plastic contact. The goal of second experiment was to find an approximate value for the quality yield of the process (that is the number of approved cablings in %). In this experiment 65 attempts to form a cabling was performed and the number of approved cablings was checked and documented. The result from the experiment – using the present conductors and plastic components – was a quality approval rate of 51%. The company requires that the quality yield is higher than 95% in order to robotize the process. A thorough analysis of the result of the experiment and discussion with company representatives indicates that the required quality yield can be reached with some further development of the involved components and the gripper making the designs more “automation friendly”. A financial analysis was made using a pay-back calculation in order to determine when and if robotization is financially justified and resulted in an approximate value regarding how much of the value –added in the operation that is to be automated. This value-added that will be performed by the robot is defined as the degree of automation. The conclusion is that robotization (automation) is technically feasible with a degree of automation of 22% and a quality yield of 95%. This is based upon: the pay-back analysis, the present design of the used components and the result from the experiments.

With the emerging development of new technology and the constantly falling prices, more companies will find interest in industrial robots. Until today, the typical robot users have been large scale car manufacturers. But there exists a big potential market within the small to medium businesses that already uses of CNC machines. Attracting smaller businesses to start using industrial robots could open up the doors to new possibilities and increase their production. Unfortunately, most people still lack the knowledge of operating and programming industrial robots. But many companies have knowledge in G-code which is normally used in CNC machines. That is why this work is focussing on the development of a software that opens up the possibility to make use of G-code to program advanced robot paths with minimal user input. By striving for easier handling of robots, the vision about a more efficient and automated society will become one step closer. The introduction straightens out the different basic principles of CNC milling machines and robots. This will introduce the reader and highlight the different similarities and differences that exist between robots and CNC mills. The goal of this work has been to create an add-in application for ABB’s off-line programming software RobotStudio that was able to import at least one type of file format commonly used for CNC milling. The program should be able to handle some basic functionality, the focus has been on the standard iso6983 type of G-code. The project started with a literature study that gave the author a better insight in both the previous research within the area but also deeper knowledge of the systems CNC mills and robots. The work continued with the development of a software able to import the specified file format. The software has been built in C# and is built as an add-in software for ABB’s offline programming software RobotStudio. The result presents a software that is able to read different types of G-code and translate them into generated paths in RobotStudio. The software also has an inbuilt function in order to parameterize the G02 and G03 commands that represent curves in G-code into straight line segments of the type MoveL in RobotStudio.

This thesis investigates cooperative and intelligent control of autonomous multi-robot systems in a dynamic, unstructured and unknown environment and makes significant original contributions with regard to self-deterministic learning for robot cooperation, evolutionary optimization of robotic actions, improvement of system robustness, vision-based object tracking, and real-time performance. A distributed multi-robot architecture is developed which will facilitate operation of a cooperative multi-robot system in a dynamic and unknown environment in a self-improving, robust, and real-time manner. It is a fully distributed and hierarchical architecture with three levels. By combining several popular AI, soft computing, and control techniques such as learning, planning, reactive paradigm, optimization, and hybrid control, the developed architecture is expected to facilitate effective autonomous operation of cooperative multi-robot systems in a dynamically changing, unknown, and unstructured environment. A machine learning technique is incorporated into the developed multi-robot system for self-deterministic and self-improving cooperation and coping with uncertainties in the environment. A modified Q-learning algorithm termed Sequential Q-learning with Kalman Filtering (SQKF) is developed in the thesis, which can provide fast multi-robot learning. By arranging the robots to learn according to a predefined sequence, modeling the effect of the actions of other robots in the work environment as Gaussian white noise and estimating this noise online with a Kalman filter, the SQKF algorithm seeks to solve several key problems in multi-robot learning. As a part of low-level sensing and control in the proposed multi-robot architecture, a fast computer vision algorithm for color-blob tracking is developed to track multiple moving objects in the environment. By removing the brightness and saturation information in an image and filtering unrelated information based on statistical features and domain knowledge, the algorithm solves the problems of uneven illumination in the environment and improves real-time performance. / Applied Science, Faculty of / Mechanical Engineering, Department of / Graduate

Automation of welding systems is often restricted by the requirements of spatial information of the seams to be welded. When this cannot be obtained from the design of the welded parts and maintained using accurate xturing, the use of a seam teaching or tracking system becomes necessary. Optical seam teaching and tracking systems have many advantages compared to systems implemented with other sensor families. Direct vision promises to be a viable strategy for implementing optical seam tracking, which has been mainly done with laser vision. The current work investigated direct vision as a strategy for optical seam teaching and tracking. A robotic vision system has been implemented, consisting of an articulated robot, a hand mounted camera and a control computer. A description of the calibration methods and the seam and feature detection and three-dimensional scene reconstruction is given. The results showed that direct vision is a suitable strategy for seam detection and learning. A discussion of generalizing the method used as an architecture for simultanious system calibration and measurement estimation is provided.

Randomly scattered sensors may cause sensing holes and redundant sensors. In carrier-based sensor relocation, mobile robots (with limited capacity to carry sensors) pick up additional or redundant sensors and relocate them at sensing holes. In the only known localized algorithm, robots randomly traverse field and act based on identified pair of spare sensor and coverage hole. We propose a Market-based Sensor Relocation (MSR) algorithm, which optimizes sensor deployment location, and introduces bidding and coordinating among neighboring robots. Sensors along the boundary of each hole elect one of them as the representative, which bids to neighboring robots for hole filling service. Robot randomly explores by applying Least Recently Visited policy. It chooses the best bid according to Cost over Progress ratio and fetches a spare sensor nearby to cover the corresponding sensing hole. Robots within communication range share their tasks to search for better possible solutions. Simulation shows that MSR outperforms the existing competing algorithm G-R3S2 significantly on total robot traversed path and energy, and time to cover holes, slightly on number of sensors needed to cover the hole, and the cost of additional messages for bidding and deployment location sharing.

This thesis explores a game-theoretical approach for underwater environmental monitoring applications. We first apply game-theoretical algorithm to multi-agent resource coverage problem in drifting environments. Furthermore, existing utility design and learning process of the algorithm are modified to fit specific constraints of underwater exploration/monitoring tasks. The revised approach can take the real scenario of underwater monitoring applications such as the effect of sea current, previous knowledge of the resource and occasional communications between agents into account, and adapt to them to reach better performance. As the motivation of this thesis is from real applications, in this work we emphasize highly on implementation phase. A ROS-Gazebo simulation environment was created for preparation of actual tests. The algorithms are implemented in simulating both the dynamics of vehicles and the environment. After that, a multi-agent underwater autonomous robotic system was developed for hardware test in real settings with local controllers to make their own decisions. These systems are used for testing above mentioned algorithms and future development of other underwater projects. After that, other works related to robotics during this thesis will be briefly mentioned, including contributions in MBZIRC robotics competition and distributed control of UAVs in an adversarial environment.

The diploma thesis concerns CNC machining of rotary part. The first part briefly introduces options of multipurpose CNC machines and robotized workplaces. The main body of the thesis analyses two production departments; one of them operated by a robotic manipulator. The conclusion summarises the accomplished results of the thesis.

The master thesis deals with the issue of a single-wheeled robot, especially its construction and movement in more difficult terrain with the possibility of collecting samples. A variant of the robot balanced by a pair of gyroscopes was chosen as the most suitable construction. The robot is able to move in a space with an inclination of up to 24° and is equipped with its own manipulator design.