While service robots are about to infiltrate everyday life still a significant effort is given to the research of robotic manipulators. As service robots become friendlier and more natural the underdevelopment of operation and programming possibilities of a manipulator becomes clear. Since small and medium enterprises (SMEs) are turning to automation the need for flexible robot cell integration is rising. Flexibility may depend on several factors:

Hardware flexibility (robot, CNC machine)

Integration flexibility (reconfiguration of robot cell)

Operation flexibility(human robot interaction)

The first two components are addressed my both the robot manufacturers and informatics technology researches. The main challenge on this field is to provide a standard protocol for the different components building up the whole production system. The service oriented architecture (SOA) paradigm [3] offers a scientifically accepted approach and it is still in the main flow of interest [4, 5, 6].

On the other hand the least flexible point of a whole system defines its overall flexibility. The problem is well defined in [7]: most flexible robot cells are sold with custom operating software and as a result the key to operational flexibility remains in the hands of the integrator. To bring more adaptively in operation [7] presents a set of robust control software. Furthermore one has to keep in mind that SMEs are often not in the position to employ highly trained operational personnel thus the easy programming and configuration contribute to flexibility (see 12. ábra). Human robot cooperation is still in the focus because of the goal for simple robot programming.

1.1.1. Robot operation in shared space

The easiest and most natural way of programming a robot is to teach the task by hand. Moving the manipulator by hand implies danger: the robot should be energized thus a robust and safe control system is needed. Any autonomous movement of the manipulator is potentially dangerous when humans must stay or work in the reach of the robot. Controlling the human robot interaction force offers a solution and implementing impedance control [8] with use of force sensor showed feasibility of collaboration without fences. Another control scheme is presented by [9] with the comparison of PD and sliding mode controller algorithms.

The other possibility to overcome this issue is to develop compliant robot systems. The lack of robustness in previous systems prompted research for new technology. [10] Presented a control system for manipulator actuated by pneumatic muscles while [11] and [12] uses direct-drive motors with wire rope mechanism.

The trend in industrial manipulator control research points clearly towards robust but compliant systems to merge accuracy and safety.

Figure 1.2.  Flexiblity factors

1.1.2. Flexible human robot interaction

In the framework of human robot cooperation the most important aspect is the bilateral communication. Physical interactions carry safety problems mentioned in the previous sub-section and environment awareness of the robot is still a significant problem. The more effective and flexible interaction with the user is intended the more sophisticated recognition technology is required.

Interesting example is presented by [13]. With traditional approach an additional channel should be used to communicate information about the object handed to the robot by the user. The use of a tactile sensor gives better potential for the robot system to understand the human intentions thus expanding the modularity of the interaction.

From the operators point of view better understanding of the system is possible to realize through user interfaces adapted to various contributors:

Complexity of the robot cell

Complexity of the task performed by the robot cell

Competence of the operator

Information needed

The key toward better flexibility is to transform the traditional and very technical graphical user interfaces between humans and industrial robots (see 1-3. figure). The focus should be shifted from the practical operation of a robot cell towards the cognitive programming and operation. Research on this includes other fields like psychology, usability, human factors.

Figure 1.3.  Traditional (upper) and flexible (lower) user interface for industrial robots

Traditionally industrial robots are machines with as minimal human interaction as possible. Including the human in the operation not only increases the flexibility and efficiency but brings improved understanding in human robot relations and takes a new step toward better trust in automation in overall.

Robots at the service sectors have to fulfill more and more sophisticated tasks. This is the same trend like at the evolution of industrial robots. The first industrial robot was created in 1937 by Griffith P. Taylor. [14] It contained almost only mechanical parts, and only one electric motor. It could stack wooden blocks in pre-programmed patterns. The program was on a punched paper, which activated solenoids. Nowadays industrial robots can weld, paint, mill, etc.

In a point of view the simplest robots at home are the washing machines, blenders, dishwashers, etc. These devices are created to help our daily routine tasks. The next step of this evolution is the help at our daily service type tasks. Inherently from the service tasks most of the robots in the service sector are mobile robots.

There are several responsive robot platforms can be found for to help us somehow, or simply just for fun. Roomba [15], or Navibot [16] helps to clean up the household in simple and convenient way like never before. There are also some doglike equipment like AIBO [17], or Genibo but these items are just trying to act as a real dog in a very primitive level. Some human inspired robots are also available, like Honda's ASIMO [18], or SONY's QRIO [19], but these complex and technically very advanced platforms are also just scratching the surface of the real social interaction between machine and human.

For the elders there is Paro [20] or Kobie, a kinds of a therapy robots, which are capable to perceive light, sound, temperature and touch, thus they are capable to sense their environment and the surrounding people. These robots could interact with the user in simple ways. They could learn the preferences of the elder in behavior, and react to their names. Even “simple” toys like these could reduce the stress level of the user, help to increase the socialization between the elders and make the communications with the caregivers more flawless.

1.2.1. Movement of the robots

Several types of mobile robots exists like tracked, wheeled, legged, wheeled-legged, leg-wheeled, segmented, climbing or hopping. The control methods of these robots are implemented individually for every construction. In most of the cases the control algorithms uses the virtual center of motion method. The path of the motion describes the path of this point [21], [22]. We can consider the whole mobile robot as a point (the centre of gravity, virtual center of motion). The velocity and the angular velocity of the centre of gravity define the motion of the robot (at constant speed). At the path of the motion we prescribe the velocity and the angular velocity of the robot in every moment.

The most often ways for mobile robot motion are the differential type drive and the steered vehicle concept. The path planning of these mechanical constructions can be extremely difficult [23]. Indoor environments increase the number of questions in this method. (see in 14. ábra)

Figure 1.4.  Steered vehicle path planning

At different driven type mobile robots path planning is simpler, because the robot can turn around without the change of the position. The orientation of the robot is always the same with the moving direction. These robots (and the driven vehicles also) have only 2 DoF-s on the ground plane.

In 1994 Stephen Killough invented omni wheels. [24] These are wheels with small discs around the circumference which are perpendicular to the rolling direction. This type of wheel does not have a geometrical constrain perpendicular to the rolling direction. With this type of configuration we can get a 3 DoF holonomic system (x, y, rot z). The most of the overland animals can move in 3DoF: turning and moving in one direction in the same time. The ordinary drive systems (steered wheels, differential drive) cannot perform this.

1.2.2. Informatics concepts

The artificial intelligence of a service robot system can be implemented on the robot, or distributed between the robot and the external environment (Intelligent Space). The Intelligent Space (iSpace) is an intelligent environment which provides both information and physical supports to humans and robots in order to enhance their performances. In order to observe the dynamic environment, many intelligent devices, which are called Distributed Intelligent Network Devices, are installed in the iSpace. [25, 26]

At the current state of science the greatest problem is caused by self-localization of robots (orientation, obstacle avoidance, path planning, etc.). There are several types of sophisticated methods for this problem. [27]

Could they be moved without requiring our presence? [28] Optical markers are generally developed for this kind of positioning. [29] (15. ábra) The most generally know optical marker is the QR code. QR code is designed only with the consideration of visual recognition and redundant information storage. (16. ábra)

Figure 1.5.  Marker localisation concept

Figure 1.6.  QR code

Ethology will play a major role in future development of social robotics because the emergence of new robotic agents does not only pose technological questions [30]. Only a broad synergic view of biological causality (ultimate and proximate), which has been practiced in ethology for at least half a century [31], can provide the necessary theoretical network which is able to handle future challenges.

May be a recent analogy could help in revealing our intentions with regard to social robotics. [32], investigated the head structures of woodpeckers in order to find out how the woodpecker’s head is protected against the vibration caused by drumming behaviour. The motion of a biological being has communication related senses.

According to human evolution we, can usually recognize the purpose, the behavior and the mood of an animal. The goal is to enhance the artificial devices with this natural interoperability, creating a way of human-machine interactions which stands for human needs. Instead of specifying any improvised implementation based on individual intuitions, the characteristics of the etho-inspired machine behavior has to be defined with a standardized methodology. These definitions are relying on organized observations of the natural behaviors, with a collection of perceptions. During the pre-organization of these tests, the observed and analyzed behavioral variables have to be defined those can make any sense for a human witness in a given scene and situation. As in many disciplines, the tests have to be performed with many subjective individuals for getting an objective result. The analyzing method consists of two simple steps: For first, the impressions and potential lateral information should be collected, which caused by the main character of the test situation. Then the impressions and information have to be conjugated to the behavioral variables in a systematic way, which allows a weighted overlap between the nodes. The weight factors must be given by using the statistics of the many performed observations.

An illustrative example, when a predator animal runs for a specified target (position) in a non-straight trajectory, because of an interfering obstacle. The most of the observers can conclude the goal position while the direction of the run is not pointing to the target. Here, the essential behavioral variable is the looking direction which contains the lateral information.

Using these analyzed measurements, a fuzzy model can be created that implements the initial concept. In many cases, the implementations require the use of new technologies. The solution of the animal like locomotion with different looking direction, body orientation and moving direction was induced the usage of holonomic robot platforms. 17. ábrashows the above example scene, where the implemented solution can be validated by repeating the observation tests with the robot, concluding that the lateral information is reflected by the behavior.

1.3.1. Informatics concepts with etho-inspiration

The best concept for the artificial intelligence is to use the abstract ethological model of the 20,000 year old human-dog relationship as the basis for the human-computing system interaction. [6] The mathematical description of the dog’s attachment behavior towards its owner is an essential aspect of setting up the ethological communication model. Such approach needed an interaction among various scientific

Figure 1.7.  Omnidirectional movement

disciplines including psychology, cognitive science, social sciences, artificial intelligence, computer science and robotics. The main goal has to find ways in which humans can interact with these systems in a “natural” way.

These systems do not have emotions, but they can act in a way that makes us believe that they are actually fond of us. While there is a need for a mathematical model of the system, the complexity of the applied solution must fit the complexity of the problem. In case of neural networks it concerns the number of the neurons and connections, in case of fuzzy logic system; there is a minimal value of the number of the fuzzy sets and fuzzy rules. Since elderly people are naturally mistrustful toward new technologies, a solution like this could makes easier to use the modern devices and fell the interactions more natural. A device like this could substitute assistant dogs in some special situations, or could cooperate with the pets to serve the elderly people in need.

As it mentioned in the previous section markers are used for robot localization. Existing markers were designed to be practical and not aesthetic, thus people did not welcome them in their homes. Who would want the walls of their home or work place to be filled with QR codes? Aesthetical appearance depends on proportional harmonization of composition, tone, pattern and rhythm. Through aesthetics marker design the goal is to create harmony between human visual recognition and image processing of robots. [29] (see 18. ábra) Without appropriate engineering methods and devices the information science would not be sufficient to express emotions. As it mentioned in section 1.2.1, the holonomic movement is one of these etho-inspired engineering methods. The eyes of the robot are especially important at the expression of emotions. (see 19. ábra) With the use of basic geometrical shapes, colors and arrangements basic emotion can be recognized.

Often machines can replace humans for more effective manufacturing because machines outperform humans in terms of strength, precision and endurance. Humans, however, perform better than machines when flexibility and intelligence is required.

Figure 1.8.  Aesthetic markers, [10]

Figure 1.9.  Robot eye concept

Figure 1.10.  Ethon robots and the aesthetic marker concept

This paper shows efficient methods and processes for teaching and optimizing complex robot tasks by introducing etho-robotics and flexible robotics. Intelligent user interfaces combining information from several sensors in the manufacturing system will provide the operator with direct knowledge on the state of the manufacturing operation. Thus, the operator will be able to determine the system state quicker. Sensory system calculates and proposes the optimal process settings. The key element is the new human-machine communication channels helping the human to comprehend the information from the sensory systems.

[1] Ministry of Economics and Trade (METI). (2010). 2035nen ni muketa robotto sangyou no shourai shijou yosoku (Market forecast of robot industry in 2035). Retrieved: http://www.meti.go.jp/press/20100423003/20100423003-2.pdf

[2] T. Thomessen, P. K. Sannæs, and T. K. Lien, “Intuitive Robot Programming,” in Proc. 35th International Symposium on Robotics, 2004.

[3] F. Jammes, H. Smit, "Service-oriented Paradigms in Industrial Automation" IEEE Trans. on Industrial Informatics, vol. 1, no. 1, pp. , Feb 2005.,

[4] G. Candido, A.W. Colombo, J. Barata, F. Jammes, "Service-Oriented Infrastructure to Support the Deployment of Evolvable Production Systems" IEEE Trans. on Industrial Informatics, vol. 7, no. 4, pp. 759 - 767 , Nov 2011.

[5] R. Kyusakov, J. Eliasson, J. Delsing, J. van Deventer, J. Gustafsson, "Integration of Wireless Sensor and Actuator Nodes With IT Infrastructure Using Service-Oriented Architecture" IEEE Trans. on Industrial Informatics, vol. 9, no. 1, pp. 43-51, Feb 2013.

[6] M. GarciaValls, I.R. Lopez, L.F. Villar, "iLAND: An Enhanced Middleware for Real-Time Reconfiguration of Service Oriented Distributed Real-Time Systems" IEEE Trans. on Industrial Informatics, vol. 9, no. 1, pp. 228-236, Feb 2013.

[7] A. Ferrolho, M. Crisostomo, "Intelligent Control and Integration Software for Flexible Manufacturing Cells ," IEEE Trans. on Industrial Informatics, vol. 3, no. 1, pp. 3-11, Feb 2007.

[8] G. Ferretti, G. Magnani, and P. Rocco, “Impedance Control for Elastic Joints Industrial Manipulators,” IEEE Trans. Robot. Autom., vol. 20, no. 3, pp. 488–498, Jun. 2004.

[9] L.M. Capisani, A. Ferrara, "Trajectory Planning and Second-Order Sliding Mode Motion/Interaction Control for Robot Manipulators in Unknown Environments" IEEE Trans. on Industrial Electronics, vol. 59, no. 8, pp. 3189-3198, Aug 2012.

[10] Tae-Yong Choi, Ju-Jang Lee, "Control of Manipulator Using Pneumatic Muscles for Enhanced Safety ," IEEE Trans. on Industrial Electronics, vol. 57, no. 8, pp. 2815 - 2825 , August 2010.

[11] C. Mitsantisuk, S. Katsura, K. Ohishi, "Force Control of Human–Robot Interaction Using Twin Direct-Drive Motor System Based on Modal Space Design" IEEE Trans. on Industrial Electronics, vol. 57, no. 4, pp. 1383 - 1392, April 2010.

[12] C. Mitsantisuk, K. Ohishi, S. Katsura, "Control of Interaction Force of Twin Direct-Drive Motor System Using Variable Wire Rope Tension With Multisensor Integration" IEEE Trans. on Industrial Electronics, vol. 59, no. 1, pp. 498-510, Jan 2012.

[13] K. Suwanratchatamanee, M. Matsumoto, S. Hashimoto, "Robotic Tactile Sensor System and Applications ," IEEE Trans. on Industrial Electronics, vol. 57, no. 3, pp. 1074 - 1087 , March 2010.

[14] Griffith P. Taylor, "An Automatic Block-Setting Crane". Meccano Magazine (Liverpool UK: Meccano) 23 (3): 172. March 1938.

[15] Jone J.L., “Robots at the tipping point: the road to iRobot Roomba”, IEEE Robotics & Automation Magazine, Vol. 13, Issue 1, pp. 76-78, Marc. 2006.

[16] Eric Guizzo, “Robot Vacuums That Empty Themselves”, IEEE Spectrum, 16. Jan. 2012.

[17] Fujita M. “On activating human communications with pet-type robot AIBO”, Proceedings of the IEEE, Vol. 92, Issue 11, Nov. 2004.

[18] Sakagami Y, Watanabe R, Aoyama C, Matsunaga S, Higaki N, Fujimura K, “The intelligent ASIMO: system overview and integration”, Intelligent Robots and Systems, Vol. 3, pp. 2478-2483, 2002,

[19] Tanaka F, Suzuki H, “Dance interaction with QRIO: a case study for non-boring interaction using an entrainment ensemble model”, Robot and Human Interactive communication, pp. 519-424, 2004.

[20] Shibata T, Kawaguchi Y, Wada K, “Investigation on living people with Paro at home”, Robot and Human Interactive Communication, pp. 1131-1136, Toyama, 2009.

[21] Martin Udengaard, Karl Iagnemma, “Kinematic Analysis and Control of an Omnidirectional Mobile Robot in Rough Terrain” in Intelligent Robots and Systems, San Diego, pp. 795-800, 2007.

[22] H. Adachi, N. Koyachi, T. Arai, A. Shimizu, Y. Nogami, “Mechanism and Control of a Leg-Wheel Hybrid Mobile Robot” in International Conference on Intelligent Robots and Systems, Kyongju, pp. 1792-1797, 1999.

[23] S.A. Arogeti, Danwei Wang, Chang Boon Low, Ming Yu, " Fault Detection Isolation and Estimation in a Vehicle Steering System ," IEEE Trans. on Industrial Electronics, vol. 59, no. 12, pp. 4810-4820, Dec 2012.

[24] Stephen Killough, "1997 Discover Awards". Discover Magazine, Retrieved 22, September 2011.

[25] Kovács B., Szayer G., Tajti F., Korondi P. Nagy I., "Robot with Dog Type Behaviour", EDPE 2011, pp. 4, 5

[26] QinZhang,L. Lapierre, XianboXiang " Distributed Control of Coordinated Path Tracking for Networked Nonholonomic Mobile Vehicles ," IEEE Trans. on Industrial Informatics, vol. 9, no. 1, pp. 472-484, Feb 2013.

[27] D. Lee, W. Chung, "Discrete-Status-Based Localization for Indoor Service Robots," IEEE Trans. on Industrial Electronics, vol. 53, no. 5, pp. 1737-1746, Oct 2006.

[28] W. Chung, S. Kim, M. Choi, J. Choi, H. Kim, C. Moon, J.-B. Song, "Safe Navigation of a Mobile Robot Considering Visibility of Environm," IEEE Trans. on Industrial Electronics, vol. 56, no. 10, pp. 3941-3950, Oct 2009.

[29] Korondi Péter, Farkas Zita, Fodor Lóránt, Illy Dániel Aesthetic Marker Design for Home Robot Localization. In: 38th Annual Conference of the IEEE Industrial Electronics Society (IECON 2012). Montreal, Kanada, 2012.10.25-2012.10.28. pp. 5510-5515.

[30] Miklósi, Á., Gácsi, M. 2012. On the utilization of social animals as a model for social robotics. Frontiers in Psychology, 3: 75

[31] Tinbergen, N (1963). On aims and methods of ethology. Zeitschrift für Tierpsychologie, 20, 410- 433.

[32] Yoon, S-H, Park S. (2011). A mechanical analysis of woodpecker drumming and its application to shock-absorbing systems. Bioinsp. Biomim. 6, 1-12.

RT-middleware (Robot Technology Middleware) is a technology that implements several key concepts needed for the development of complex robot systems, even in geographically distributed environments. Through a useful API, reusable standardized components and communication channels, and some degree of automatization, RT-middleware helps the user to build easily reconfigurable systems. The behavior of the components and the manner of interaction among them are standardized by the Robot Technology Component (RTC) specification. Until now there have been two implementations of the specification. OpenRTM.NET is written in .NET, while the other implementation, called OpenRTM-aist.

Within the robotics area, the task of robot systems can change quickly. If the job and environmental circumstances change frequently, reusable and reconfigurable components are needed, in addition to a framework which can handle these changes. The effort needed to develop such components depends on the programing language, the development environment and other frameworks used. Many programming languages can be applied for this process, however developing a brand new framework is a difficult task. Applying an existing framework as a base system, such as the OpenRTM-aist in the case of this chapter, the associated development tasks can be dramatically simplified.

Frameworks and middleware are gaining popularity through their rich set of features which support the development of complex systems. Joining together any robot framework and robot drivers can form a complex and efficient system with relatively small effort. In many cases, the task of the system designer is reduced to the configuration of an already existing framework.

On the other hand, when an existing framework requires only a set of completely new features, it can simply be extended with them. Improving an existing framework is an easier job than developing a new one because the design and implementation of such a system requires specialized knowledge and skills (design and implementation patterns). In most cases, the missing functionalities can be simply embedded into the existing framework, nonetheless there are some cases when this is hard to achieve. An existing framework can be improved simply only in case it is well designed and implemented. In spite of this, building a brand new system from is a much longer process that requires much effort.

In the area of robotics, there are many robot parts that share similar features, so the concept of robot middleware as a common framework for complex robot systems is obvious. Probably there is no framework which can fulfill all of the above requirements entirely.

In this section, I explore the requirements of robot middleware technologies in general and examine some existing robot approaches such as YARP, OpenRDK, OpenRTM-aist and ROS.

Robot middleware is a software middleware that extends communication middleware such as CORBA or ICE. It provides tools, libraries, APIs and guidelines to support the creation and operation of both robot components and robot systems. Robot middleware also acts as a glue that establishes a connection among robot parts in transparent way.

First of all, we should define the requirements and the actors of robot middleware. Generally, two actors use such technology: end-users and developers. Every robot middleware has to provide tools and mechanisms to facilitate the work of these actors.

Figure 2.1.  Main use cases of robot middleware

The main use cases of robot middleware are shown in 21. ábra. After the analysis of these use case we can define some functional requirements. The component developers design and implement components. For this development procedure, many programming languages and operating systems should ideally be supported. Tools are needed for the generation of skeletons and semi-working components for quicker development. The main activity of the developer actor should be the generation of semi code, the filling out of business logic, the compilation of binary code and the execution of test cases. Developers also create robot parts as components using their knowledge about programming languages, robot middleware and the specification of the given robot system. Moreover, the person who develops the component has to validate it as an end-user. In summary, from the developer's point of view, the main requirements for robot middleware are as follows:

As opposed to the developer actor, an end-user is someone who has no knowledge about programming languages, as is not interested in developing components. End-users simply want to use components for their own work. In their case only a minimum number of system parameters should be visible, which are sufficient for the creation and operation of the robot system. Nice graphical interfaces and visual aid features are useful to end-users in allowing for the straightforward operation of the robot system. End-users have two main tasks. First, they often create complex robot systems using existing robot parts, and second, they often want to run previously created robot systems. The creation of a robot system can include a search for available robot parts during which the user obtains information about the accessibility of available components. This accessibility information can depend on the specific robot middleware which is applied for the task, and may include IP numbers, or any internal identifiers or symbolic names. Unfortunately, most robot middleware systems only support online components. For this reason, robot parts have to be available during the robot system creation process, i.e. have to be already running and must be registered.

End-users want to register their components in order to share functionalities with other end-users. After registration, the component can be found by all other participants. Usually end-users create custom robot systems using available components, so the second task is to find out which of the components are running. Once the robot systems are running, users will want to control them, i.e. activate/deactivate any or all of the components, and run them with various different parameter sets.

In summary, from the end-user point of view, the main requirements for robot middleware are as follows:

In summary, the key consideration for a robotics middleware system to be useful is that it should provide an efficient API and as much automatization as possible. Additionally, middleware systems also have to support as many operating systems and programing languages as possible. If a framework is applied for a specific task, some efforts will always be needed to understand the relevant concepts and application details.

One distributed environment for robot cooperation is OpenRDK. The concept is detailed by Calisi in [1] . The user’s robot system can be developed using a set of Agents, through a simple process. A Module is a single thread inside an agent process. Every module has a repository in which a set of internal properties are published. Inter-agent (i.e., inter-process) communication is accomplished by two methods: through property sharing and message sending. RConsole is a graphical tool for remote inspection and management of modules. It can be used as both the main control interface of the robot and for debugging while developing software. It is just an agent that happens to have some module that displays a GUI. The framework can be downloaded from [2] .

Alternative important robot middleware platform is Yet Another Robot Platform (YARP). Communication in YARP generally follows the Observer design pattern (for more details see [3] ). Every YARP connection has a specific type of carrier associated with it (e.g., TCP, UDP, MCAST (multi-cast), shared memory, within-process). Ports can be protected by SHA256 based authentication. Each port is assigned a unique name and it is registered into a name server. The YARP name server, which is a special YARP port, maintains a set of records, the keys of which are text strings (the names of the ports). The remainder of each record contains whatever information is needed to make an initial connection to a given port.

The third important robot middleware technology is OpenRTM-aist (documentation about it can be found [4] ), which is a convenient modular system also built on the Common Object Request Broker Architecture (CORBA). This concept uses the distributed components approach, where the components have to be registered into central naming service. (The openRTM-aist uses the name server of CORBA system.). The whole concept is detailed by Ando et al in [5] , [6] and [7] . The components can contain modules what can be loaded any time. The framework is implemented using design patterns, robust software system, which provides huge feature set, more than previously mentioned ones. The components of robot system have same rights no central logic (except the naming service). Each RT-Component has an interface (a "port") for communication with other components. The RT system is constructed by connecting the ports of multiple components to each other in order to aggregate RT-Component functions. The advantage of OpenRTM-aist is that it provides a simple way to create and co-operate various robot parts. Two kinds of port can be used the service port and the data port. The data ports provide the asynchronous messages and the service port establishes the synchronous communication. The usage of service port is complicated because we have to know the CORBA technology that is an out of date system. The connection among components has to be established at runtime by the end-user using a graphical interface. A comfortable state machine concept is provided i.e. the developer has to override the appropriate methods only. More programing languages and mostly used operating systems are supported. This is the second popular robot framework in the world. (The most popular in Japan.)

The final investigated robot framework is the ROS (Robot Operating System), that seems a basic operating system, but it provides similar features as previously mentioned (available at [8] ). The concept is detailed by Quigley in [9] . The items of robot system have to be registered a central server as well, but this component is called ROS node in this case. This node is the basic structure that can not be divide more parts in official ROS, but a 3rd party component exists called nodelet which support run more algorithm in one process (node). A central process has to be run (one roscore file creating a ros master and rosout nodes) providing the central log opportunity and a distributed parameter set. The connections among node are established automatically, i.e. if a component is running that advertises the topic and the other which subscribe to the topic, then the two nodes store information about the other using the roscore API. The developers' work is assisted completely, but only few tools are available end-user. This is the most popular robot middleware all over the world.

In the case of all investigated robot framework a concept is handled as distributed entity that is an agent or a port or a component or a node. The YARP provides more communication possibility nevertheless it has poor set of tools. Unfortunately the development of YARP and OpenRDK is broken. Only two robot frameworks are used nowadays, the OpenRTM-aist and the ROS, which is in progress so in the later part these robot system will be compared and detailed.

In this section the ROS and the OpenRTM-aist robot middleware are investigated in point of view of developer. The complexity of significant procedures are compared and additionally the helper tools.

2.4.1. Creating a new project

In the case of ROS the new package can be generated by a tool named catkin_create_package that creates a folder containing a CMakeList.txt and a package.xml. These files have to be modified if another package is needed or a new executable file is necessary. If a simple message sending or receiving is should be used then the official tutorial can be modified easily and the business logic can be placed into a single function. It doesn’t generate any file or class, but it is not necessary for usage of ROS. The structure of beginning project is very simple, understandable. The basic features of ROS can be accessed only via three classes and the implementation of business logic not requires serious knowledge of ROS system. C++ and Python programming languages are supported officially. A custom creation tool is presented called catkin_make that is an extension of the platform independent CMake build system. More Integrated Development Environment (IDE) can be used that are supported the CMake, because the CMake can export it (at the moment CodeBlocks, Eclipse, Xcode, KDevelop, and Visual Studio).

For project creation, the OpenRTM-aist middleware also contains a command line tool (rtc-template) and graphical editor, called RTC Builder (the graphical interface is shown in 22. ábra.

Figure 2.2.  The graphical interface of RTC Builder

The definition of each RTC component is stored in a programming language independent xml file, and the source codes are automatically generated by the robot middleware framework using this XML file. Four source code files are created. One header and one C++ file for class definition, one C++ source file containing the main function and a makefile for building process. The developer has only to fill out the body of each skeleton member function and to insert other external classes. After building our system using these generated source we get a minimal working system, but it is hard to understand and to modify because substantial knowledge is needed about OpenRTM-aist system and the RT-specification. In the generated files the class is a subclass containing overriding methods and codes of design patterns. The knowledge of 5-10 classes are necessary for beginners. The generation process of a beginner project requires a lot of parameters at least 10, but in a complicated case can be 20-30 parameters too. C++, Java, Python programming languages Eclipse and Visual Studio are supported.

2.4.2. Component development

More than 2000 3rd party components are available for ROS that can be used freely for own robot system. Using these components the development period of the new robot system can be quick and easy. If our system is well designed i.e. the workflow of the components and the interface among them are designed precisely then the development process can be done in parallel due to available ROS tools. A message to a topic can be sent an easy way by rostopic from command line or listening to arriving messages to a specified topic. The construction of the component is not block the development of another component. Sharing a new functionality needs only a new method and variable definition and a simple function call.

Figure 2.3.  The architecture of RT component

In case of OpenRTM-aist approximately 1000 components are available. Some of them can be used in Windows operating system only and the others restrict us to Linux. Unfortunately the documentation of the most components written in japan only and we have to trust the result of any language translator. The Debug and test tools are missing that helps to developer, we have to implement dummy components for this reason. In most cases the correct usage of the system services requires deep knowledge about CORBA technology that increases the full time of development. If a behaviour of any component can be used remotely we have to do more steps in the provider and consumer side also. In this case the usage of CORBA can not be omitted. The developer has to create a class containing methods of state machine and the business logic. The architecture of the component is show in the 23. ábra.

The OpenRTM-aist robot framework is responsible for more services for the end-user than other ones. It supports the substitution of component, robot system creation, change of parameters of the robot, and the control of the component. Some command line tools are presented in rtc_shell package which are Python scripts. In addition, a graphical system editor is shipped as Eclipse plugin. The size of this system editor (and the Eclipse itself) is too large and the graphical capabilities are limited, but it is an effective supervisory interface of the system (the graphical interface is shown in the 24. ábra.)

Figure 2.4.  The graphical interface of the System Editor of the OpenRTM-aist

In the left side of the window the end-user can browse the online components in the middle part the actual robot system is shown. In the right side the main parameters of the component are shown. The end-user can start or stop the robot system or any component of the system at any time. If this GUI is not sufficient, than a new tool should be created. Unfortunately, the detailed documentation of OpenRTM-aist system is missing so the source code investigation is needed.

Exiguous amount of tool exist in ROS aiding the end-user. Some graphical tools are implemented in qt helping to view the log messages and draw the structure of robot system. Moreover, we should mention a promising tool that runs in a browser called Robot Web Tool. This is a Java script collection communication to the Python server application (a bridge to ROS) but additional development is necessary.

The OpenRTM-aist contributes more features to end-users via graphical interface, but more knowledge is needed for developer. In the case when the researcher would like to try more solutions i.e. varying the component the OpenRTM-aist is better than ROS.

It supports more programming languages and operating systems also, so it is easy to integrate it into the existing topology and easy to develop the new system. Unfortunately, the documentation is poor and some of them are exist in japan only which is not popular except in Japan. The support of more operating systems is mentioned as advantage, but it is drawbacks also, because it can lead to complicate and heterogeneous robot topology. If our robot system requires components which are found, however that components can be run only different operating system, it requires additional PC-s. The OpenRTM-ais provides more tools and features to developers, but more knowledge and experience are needed, nevertheless the documentation of the system is poor. More useful code can be found in the sample projects complementing the missing documentation, however the developer has to search it and has to try it with different parameters. An active community of OpenRTM-asit exists in Japan, nevertheless the questions and the replies are presented in Japanese language only in most cases.

The ROS is suggested in the case of autonomous system when the connection of robot parts is mandatory. In this case the graphical editor is not necessary. Because of huge number of existing components, probably one or more component exists which fulfil our requirements which are small, so the understanding and the modification are an easy task. It is easy to create a new project from the scratch (easier than OpenRTM-aist) because the creation of classes is optional we have to use existing classes and methods only. The development of the ROS core is continuous by a community, the seventh version is in progress. There are a lot of information about it and a lot of questions and answers on the web page.

At the previous decades the industrial robots and NC machines becomes available for smaller companies. Industrial robots has slightly different programming mode than other Industrial machines like CNC machines and Lathes. Programming of NC machines is more standardized (G-code), while different programming languages and interfaces exist for different robots even from the same manufacturer. It is important to follow standards at production, however this standardization is absent in industrial robotics. Despite the standard G code NC machines usually do not have any appropriate port (and sometimes I/O opportunities) for high level communication in a manufacturing cell. Due to these limitations of existing NC and robot controllers, many researches were started to develop a new, open-source, flexible middleware software and hardware architecture for replacing outworn controllers [1, 13].

We are the first in the field of universal control development who are combining the following two popular existing concepts to merge their benefits: the reliable LinuxCNC and the flexible RT-Middleware technology.

The organization of the paper is as follows: Section II describes the origin of LinuxCNC software system. Section III introduces the RT-Middleware technology, Section IV presents three different concepts for hardware architecture, Section V gives an example solution with experimental results, and Section VI concludes the paper.

The US government sponsored Public Domain software systems for numeric control of milling machines were among the very first projects developed with the first digital computers in the 1950`s. In fact, the need and concepts of universal motion controllers is not a novel issue in the industry. The universality was described by modularity, portability, extensibility, and scalability requirements before two decades when the development of LinuxCNC was already started [2]. The project was originally launched by the National Institute of Standards and Technology (NIST) in 1989 [2, 3] and the software was moved under public domain in 2000, allowing external contributors to make changes and reuse the code. The characteristics of universality are still forming an actual topic as the available hardware elements are improved and the control architectures are developed. Today, LinuxCNC is a very reliable and popular open source software system that can be used under General Public License for numerical control.

The Japanese Ministry of Economy, Trade and Industry (METI) in collaboration with the Japan Robot Association (JARA) and National Institute of Advanced Industrial Science and Technology (AIST) started a 3 year-national project “Consolidation of Software Infrastructure for Robot Development” in 2002 [4]. An ICE Extension was proposed in [14,15]. With the intention of implementing robot systems to meet diversified users' needs, this project has pursued R&D of technologies to make up robots and their functional parts in modular structure at the software level, and to allow system designers or integrators building versatile robots or systems with relative ease by simply combining selected modular parts. The robot technology middleware have been developed as infrastructure software for implementing the proposed robot architecture, named "OpenRTM-aist" (Open Robot Technology Middleware).

To manage the rapidly growing need for sensor communication in robotic applications several suitable architectures, named middlewares, are being developed for easy system integration. Unfortunately, most of these middleware technologies are developed independently of each other and are often dedicated for specific user applications [5]. RT-Middleware is the only middleware solution that is under standardization [6] and also this solution has proved to be industry ready and used by many industrial partners: Toshiba (different system components), Honda (ASIMO humanoid robot), AIST (OpenHRP humanoid robot, etc.) and also many research institutes. The most important middleware solutions can be found in [7- 11] and a comparison can be found in [11, 12]. The main goal of the middleware system is the development of a common language and control concept for different users and tasks. The base of RT-Middleware concept is that system is build up from low-level, real-time, platform independent components (RTCs, RT-Components). These components are developed for certain machines, sensors and actuators. From these components more complex machine configurations can be made: production cells or an entire production line. The advantage of this architecture is its well-defined hierarchy, modularity, scalability and fast development ability. More details about the RT-middleware can be found in references [9].

Following chapter describes three different experimental systems for different machines. The first is a joint controller which can be the smallest element of a manufacturing cell and controlled by an RTM server. The second is a compact 3-axis desktop CNC controller, which works as a small plug&play RT-Middleware unit. The third concept is combining the low-level flexibility (able to control different kind of systems on low level) of the LinuxCNC software system and the high-level flexibility (connectable to each other to form a production cell) of the latest RT-Middleware technology.

3.4.1. The first one is a joint controller

In case there is no hard-real-time attachment necessary between machine joints, it is possible to develop one RT-Middleware RTC hardware and software unit to drive any joint. In this concept more instances of joint component is controlled by a higher level RTC component. Main advantage of this architecture is its flexibility: new joints can be attached or joints can be reconfigured with the reuse of joint RTC hardware and software component. It has certain small delay in control which makes it not fit for example simultaneous movement of high speed CNC machine joints, but it fits to most automation applications like pick&place. An RT-Middleware based joint was developed to test the performance of this concept and get experimental results. Two pieces of this joint controller were manufactured and attached to the first two angular joints of an ADEPT 604-S type SCARA robot. Block diagram of the hardware architecture can be seen on figure IV.1. The logic was implemented in a SUZAKU-V FPGA board. Custom logic and PowerPC processor was configured to run Linux kernel and realize real-time interfaces. The board is connected to the Middleware server via Ethernet. Furthermore the developed electronics includes power amplifier for DC motors, signal conditioning and necessary power converters. It was developed to be the smallest element of a manufacturing cell.

Figure 3.1. The block diagram of the joint controller RTC

3.4.2. The second one is a simple decentralized CNC controller

At a 3 axes CNC machine the joints are usually linearly independent. In this case the joint coordinates have effects only on the X, Y and Z coordinates of the TCP. Our decentralized CNC controller can get the joint references from an external PC based G code interpreter. The mechanical design and the shape of the workspace of milling machines can be specialized for different tasks, but the control method can be the same. After the controller and the machine is connected, the position control loops of the axes run on the CNC controller (figure IV.2), and can be tuned via RS232 port from a PC. It is possible to control additional functions along output 8 relays, and 8 isolated inputs, for example coolant, tool size measurements, and other PLC functions. These I/O interfaces can be accessed along ModBus interface. With this system design and a normal PC different CNC machines can be controlled without machine specific controller development. The G code is a common machine language so the PC based RT-Middleware G code interpreter can be used universally. Our decentralized CNC controller is a universal hardware element for these machines so the cost of this concept can be reduced by the increasing number of the manufactured controllers.

Figure 3.2. Block diagram of the 3 axis CNC controller

3.4.3. Linux based modular multi-axis controller

The latest version of our controllers is a complete modular solution for most motion control applications up to 9 axes. The system is based on LinuxCNC, which reliable core was developed since more than two decades, described in the introduction. And we combined this deep experience with modern hardware elements of today's semiconductor industry. This platform is an open-source software system that implements numerical control capability using general purpose computers to control any servo driven machines. It uses Linux kernel with real time extensions (RTAI or RTLinux), and can control up to 9 axes or joints of a machine using standard G-code as input. It can handle the operation of all peripheral machine elements, e.g. tool length measurement, cooling, tool-change procedure, etc. The graphical user interface can be customized for specific kinds of usage e.g. touch screen or interactive development, see figure IV.3. Programming features include most needs of NC usage. It is possible to implement inverse kinematics, hence the software system is capable to control non-Cartesian robots as well. The hardware architecture of the generic system layout can be seen on figure IV.4. The PC installed with LinuxCNC is hosting a PCI card, which is based on a PCI bridge ASIC and an FPGA.

Figure 3.3. Default (a) graphical user interface of LinuxCNC software system

Figure 3.4. Customized (b) graphical user interface of LinuxCNC software system

Figure 3.5. (a) Typical system layout of the LinuxCNC based motion controller

Figure 3.6. (b) Typical system layout of the LinuxCNC based motion controller

(b)

As the LinuxCNC has a simple ladder diagram editor and handles standard PLC functions, the hardware architecture is extended with a RS485 bus for I/O extension modules. The bus is handled by a separate 8-bit RISC microcontroller on the PCI card, which communicates via SPI with the FPGA. Currently digital input, relay output, ADC/DAC, and teach pendant modules are exist. These modules can be connected in chain with a termination resistor at the end of the bus, see figure IV.5./a. Each module has a unique, four bit long address on the bus, hence up to 16 modules can be chained to one PCI card.

Figure 3.7. RS485 expansion (a) bus

Figure 3.8. RS485 expansion (b) module concepts

Basically, the modules have a bus powered and a field or application powered side with an optical isolation between them, see figure IV.5./b. But for example an optical isolated input module has only field power side as the inputs from the field are only simple opto-coupler inputs. Each node connected to the bus is recognized by the PCI card automatically. During startup the driver exports pins and parameters of all available modules automatically. All the I/O-s of these modules can be routed to a function in the hardware abstraction layer (HAL) of LinuxCNC, and can be controlled by a custom logic implemented with the ladder editor.

The specialty of the system is in the servo connection interface. A novel modular concept is introduced for connecting different types of servo controllers including the old analogue systems, the incremental systems, and modern CAN based servo modules. Four different small and simple axis-interface modules were developed, and combining them in many different ways makes the possibility of interfacing with a servo controller. These types are optical isolator, DAC module, differential line driver, and a breakout for simple terminal connection. Figure IV.6. shows only two different servo examples from the possible 8: a analogue, and an incremental with differential outputs and encoder feedback.

Figure 3.9. Examples of (a) analogue incremental servo interface

Figure 3.10. Examples of (b) differential incremental servo interface

This chapter presents an application to multi-axis simultaneous control. LinuxCNC was connected to a SCARA type robot. The theoretical background of the implemented control algorithm is described in point A). The modular hardware structure resulted a flexible system which can be connected to any similar machines without long time demanding hardware development (described in point B).

3.5.1. Kinematics of the experimental non-Cartesian system

Three different universal architectures were introduced from which two were connected to a SCARA type robot. This chapter describes the implemented SCARA kinematics (Fig. V.1.) which is followed by hardware and driver-level software description of an experimental systems.

Figure 3.11. Mechanical drawing of the Adept 604-S SCARA robot. m1, m2, m3 are the masses, l1,l2,d0,d3 are the length, q1,q2,q3,q4 are the angles of the corresponding joints. (These data are necessary only for the calculations of robot dynamics: lc1 and lc2 are the masses position on joint 1 and joint 2, respectively.)

The Cartesian coordinates of TCP can be expressed as equation (5.1), (5.2), (5.3).

(5.1)

(5.2)

(5.3)

Forward kinematics can be expressed deriving equations (5.1), (5.2) as (5.4),(5.5).

(5.4)

(5.5)

The connection between the Cartesian velocities and joint velocities can be represented as equation (5.6).

(5.6)

Where

Where is a Jacobean matrix as equation (5.7)

(5.7)

3.5.2. Application example

At a manufacturing line an assembler or palletizing cell is usually indispensable. The most common palletizing type robot is the SCARA. The LinuxCNC has a kinematic module, which software part is open source as well. We designed the kinematic module for a SEIKO D-TRAN TT 4000SC SCARA robot. The original controller was broken and obsolete, but the mechanics of the machine worked perfectly. Our controller has a modular built up, so the main parts of the system are separated in three shelves. The first one is universal for most kind of machines which is the PC and the PCI card. (Fig.V.2.a) The machine specific signals (end- and homing- switches), the RS485 bus, and the CAN references are also connected to this part of the controller.

The second shelf contains the modular components, which is chosen for the specific machine. In this case these are the DC power amplifiers and the RS485 modules. (Fig.V.2.b) (These can be digital and analogue I/O-s.) The power amplifiers are connected to the DC bus, to DC logic power, to CAN references and to the encoders. The motor parameters, the encoder parameters and control loops at the power amplifiers can be tuned and measured via USB.

The third shelf contains the power electronics. (Fig.V.2.c) These elements are mainly saved from the original controller. It has an E-Stop circuit, fuses, DC powers, brake resistors and thermal protection. With this concept we could save and modernize the original robot mechanics (and some of the power electronics). The machine can be integrated into a modern and flexible manufacturing line as a palletizing cell. According to the modular built up of the controller we could use the SCARA robot again without any long time demanding electronic and software development. As a part of the system we also developed a teach pendant with touch screen and joysticks for the X,Y,Z and rotZ jogging. The controller has standard Ethernet, USB, etc. interfaces. With the LinuxCNC RT-Middleware component the robot can be connected and integrated into a high level manufacturing control system.

Figure 3.12. Shows (a) control of the modular controller

Figure 3.13. Shows (b) power amplifier of the modular controller

Figure 3.14. Shows (c) power electronics shelves of the modular controller

The LinuxCNC & RT-Middleware based modular control system was implemented with success for more machines now: Two 3-axes desktop CNC, a 3-axes dental CNC, a 5-axes with tool-change and tool-length measurement, and at least two SCARA robots. All of these machines can be connected along the RTM interface of the LinuxCNC and can work together in real-time applications. The improvement of the modules and all other hardware elements was in progress with every system integration. The first controller was running since more than one year with negligible maintenance now, showing that the system is reliable too. The application was started by external international system integrators in Europe this year.

Bence Kovács, Géza Szayer, Ferenc Tajti, Solvang Bjorn, Péter Korondi, Design of a universal robot controller Robi, International Engineering Symposium at Bánki (IESB 2011). Budapest, Hungary, pp. 1-13, 2011.11.15.

Proctor, F. M., and Michaloski, J., "Enhanced Machine Controller Architecture Overview," NIST Internal Report 5331, December 1993. Available online at ftp://129.6.13.104/pub/NISTIR_5331.pdf

Albus, J.S., Lumia, R., “The Enhanced Machine Controller (EMC): An Open Architecture Controller for Machine Tools,” Journal of Manufacturing Review, Vol. 7, No. 3, pp. 278–280, September 1994.

Bence Kovács, Géza Szayer, Ferenc Tajti, Péter Korondi, István Nagy: "Robot with dog type behavior”, 17th Int. Conference on Electrical Drives and Power Electronics The High Tatras, Slovakia 28–30 September, 2011

Kotoku, Tetsuo: "Robot Middleware and its Standardization in OMG" in International Conference on Intelligent Robots and Systems (IROS'06) Workshop on Robotic Standardization, Beijing, China, 11-13 Oct. 2006

J.U. Cho, Q.N. Le and J.W. Jeon, An FPGA-Based Multiple-Axis Motion Control Chip, IEEE Trans. Ind. Electron., Vol. 56, No. 3, pp. 856-870, Mar. 2009.

Ozaki Fumio, Oaki Junji, Hashimoto Hideaki, Sato Hirokazu, "Open Robot Controller Architecture (ORCA)" in Journal: Nippon Kikai Gakkai Robotikusu, Mekatoronikusu Koenkai Koen Ronbunshu, Vol. 2, 2003.

M. Mizukawa, H. Matsuka, T. Koyama, T. Inukai, A. Noda, H. Tezuka, Y. Noguchi, N. Otera, "ORiN: open robot interface for the network - the standard and unified network interface for industrial robot applications" in Proceedings of the 41st SICE Annual Conference (SICE 2002), Tokyo, Japan, Vol. 2, pp. 925- 928, Aug. 2002.

N. Ando, T. Suehiro, K. Kitagaki, T. Kotoku, Yoon Woo-Keun, “RT-middleware: distributed component middleware for RT (robot technology)” in Proceedings of 2005 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2005), pp. 3933- 3938, ISBN: 0-7803-8912-3, 2-6 Aug. 2005.

G. Veiga, J. N. Pires,K. Nilsson “On the use of SOA platforms for industrial robotic cells” in Proceedings of Intelligent Manufacturing Systems (IMS2007), Spain, 2007.

J.-C. Baillie, “URBI: towards a universal robotic low-level programming language” in Proc. of IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2005), pp. 820- 825, ISBN: 0-7803-8912-3, 2-6 Aug. 2005.

B. Solvang, G. Sziebig, and P. Korondi, “Multilevel control of flexible manufacturing systems,” in Proc. IEEE Conference on Human System Interactions (HSI’08), pp. 785– 790, 2008.

Solvang Bjørn, Refshal Lars Kristian, Sziebig Gábor “STEP-NC Based Industrial Robot CAM System.” In: 9th IFAC Symposium on Robot Control (SYROCO2009). Gifu, Japán, 2009.09.09-2009.09.12. (IFAC) Gifu: IFAC by Pergamon Press, pp. 361-366.

Zoltán Krizsán, “ICE Extension of RT-Middleware Framework”, in Proceedings of 10th International Symposium of Hungarian Researchers on Computational Intelligence and Informatics, pp 513-521, 2009.

Zoltán Krizsán, Szilveszter Kovács: Structural Improvements of the OpenRTM-aist Robot Middleware, Applied Computational Intelligence in Engineering and Information Technology Topics in Intelligent Engineering and Informatics, Eds: R.E. Precup, S. Kovács, S. Preitl and E.M. Petriu, 2012, Volume 1, pp. 287-300, ISBN: 978-3-642-28304-8

The Internet is a very fast evolving new technology, allowing people to electronically connect places that are thousands of miles apart. By combining network technologies with the capabilities of mobile robots, Internet users can discover and physically interact with places far away from them, creating opportunities in resources sharing, remote experimentation and long-distance learning. This paper provides an assessment of the state of art of Internet based control and presents a general approach to internet based telemanipulation. The concept is based on layers. This paper proposes these layers as well as their functions. One of the most important problems of a closed-loop internet-based telemanipulation application is the transmission time delay, which can destabilize the system. Furthermore, when using a computer network for sending data, time delay is not even constant. In this paper the Internet transmission time delay is discussed, various remote control modes for robotic systems are surveyed and an internet-based tele-handshaking as an applications is discussed in details. The task is to realize a handshake between two people at two different corner of the world via Internet.

Human operators perform their tasks mostly by hand and the master device must fit the operator’s hand. In the field of telemanipulation, many haptic devices have been developed. The applications of the haptic devices cover various fields, such as medical application, working in nuclear power plants and entertainment. In this section four important master device concepts are presented.

4.4.2. The point type master device

It similar to a six DOF manipulator. The operator can point at the remote site. The Phantom [12] is a point type device (see 4-10. ábra) using direct drive and is noted for its high resolution. However it can cover only fingertips. The operator can control a tip type slave device with this device. Tip type slave systems are widely used in surface manipulation applications, such as telemanipulation in the nano space [11].

Figure 4.10.  A point type master device

4.4.3. The arm type master devices

It covers the human arm from shoulder to wrist (see in in 4-11. ábra). It has 6 degrees of freedom. The arm type haptic device is useful in applications where the operator uses his arm to do some task in the remote environment, for example some mining workplace. In several applications the arm type master device is a subsystem of a complex master device.

Figure 4.11.  An arm type master device

4.4.4. The glove type master device

It allows the most complex and dexterous manipulation [13]. The human hand is the most widely used tool. The final goal is to make such a device that the operator can feel as if the function of whole his hand were expanded. Several types of manipulation cannot be performed without force feedback. A 20 DOF sensor glove type device with force feedback (see in 4-12. ábra) is discussed in the next section.

Figure 4.12.  A glove type master device

4.4.4.1. Mechanical structure

The wrist location is measured by an inductive sensor. The sensor glove is equipped with force feedback capability to each joint of the operator fingers, so the operator can feel the remote environment as he/she is there. The mechanical structure of a finger is shown in 4-13. ábra. The device is designed with consideration to the space for sensors and actuators. The device is attached to a human hand by bands. A fixed base for all movements of the glove is constructed on a metallic plate fixed on the back of human palm. Five yawing drive units are fixed in the plate. Three pitching drive components are placed on the yawing drive units.

Figure 4.13.  Mechanical structure of the sensor glove

Figure 4.14.  Finger movement in the glove

4-15. ábra shows the structure of one D.O.F. of the device. The torques of the finger joints are measured by strain gauges. Rotary encoders in the motors are used to measure the angles of the joints. Ideally, the motors move the joints directly, however, there was no motor available which was small and light enough to be attached on the hand. The pulleys and wires in tubes are used to transmit forces to the motors. Single motors can control all the 20 joints of a human hand independently. One joint of the finger corresponds to 2 links of the device. The main problem is the friction. Because of this additional force, the operator cannot feel, which force is emerged in remote environment. The friction compensation is discussed in Session IV.

Figure 4.15.  Structure of one D.O.F. of the Sensor Glove

4.4.5. Micromanipulation Systems

In this section, the tele-micromanipulation system is introduced (see 4-16. ábra). In micromanipulation, visual information of microenvironment is usually fed back by microscope. It is difficult to manipulate micro object using only 2D visual information that is why a 3D animation of slave device (see left part of 4-16. ábra) helps the operator. In this system, the master input device used by the human operator is called as haptic interface, which is a 6 DOF joystick type master device (see in 4-17. ábra). The slave manipulators used directly to perform manipulation tasks are shown in 4-18. ábra. The slave manipulator and master device systems are connected using Ethernet and they are used to perform teleoperation through the network.

Figure 4.16.  Concept of the Micro Telemanipulation

Figure 4.17.  The photo of the Master Device

Figure 4.18.  The photo of the Slave Device

4.4.5.1. The Master Device of the micromanipulation system

The structure of master device is ought to be same as the slave device. However the parallel link structure has small workspace, which is not convenient for the human operator. That is why a serial link structure is applied to get large workspace, although it decreases accuracy and stiffness. The master device is stick-type device composed of 3 linear servo motors which realize 3 axis (X, Y, Z) parallel movement and 3 AC servo motors which realize rotation around each axis (, , ) [12] (see 4-17. ábra). It covers all 6 degrees of freedom of the slave device. The operator holds the master-stick in his hand and he can move and rotate it. The workspace is 340mmx340mmx340mm for linear movement, and the possible rotation is +/- 15 degrees for each axis. All degrees of freedom have force feedback.

4.4.5.2. The Slave Device of the micromanipulation system

As a slave device, a parallel link manipulator [13, 14, 15] was developed (see in 4-18. ábra), which has 6 degrees of freedom. In general, the parallel link structure has good features of precision and stiffness, although it has small workspace. This structure is fit for precise manipulation. It has 3 degrees of freedom for XYZ linear motion and 3 degrees of freedom for rotation of each axis. Its workspace is almost 30mmX30mmX30mm for linear motion and +/- 15 degrees for rotation. The positioning accuracy is 10 m. It is different from the well-known Stewart Platform structure [15], it has novel structure: Six links, which sprouted verticality from base table can expand and contract. Each two links of six links are combined into one by sublink. Ultimately, the end effector table is held by total 3 points (details in [13]). This manipulator is controlled by a PC (dual Pentium III 500MHz). Operating System is Real Time Linux (RT-Linux) in order to perform motor control with 2.5 kHz sampling time. Input-output using motor, rotary encoder, force sensor are performed by AD, DA, counter, DIO boards connected to an extended bus. As actuators, six AC servomotors of the same type are installed, and 1mm pitch ball screws are connected to each motor. The main drawback of this structure is the complex kinematics and dynamics with singular points as discussed in [13].

Visual feedback is aimed at giving a quite real representation of the operator’s hand, and the program enables the user to wander it in full three-dimensional space.

The program was designed to run either locally in full simulation mode or over TCP/IP connection, such as Internet or Local Area Network. In simulation mode, the program can be used to represent all the motions of the human hand. To accomplish it, both anatomical and mathematical models were built up, and these models were implemented in OpenGL. The mathematical deduction uses the Denavit-Hartenberg notation, because it can be quite well applied both for hands and in OpenGL. The model uses several basic assumptions, these are:

It is important that coordinate frame 0 is the same for all fingers, and is a common reference to the whole hand. The individual i angles and l0 variables determine the exact position and orientation of the next frame of the next joint in each finger. Between the two frames di is just equal to the length of the link, l0. Since joint 0 is of the revolute type, q0 is 1 the relationship between the two joints is (4.1).

In (4.1) A01is the 44 homogeneous transformation matrix between the two adjacent joints and Xsare position vectors in the respective coordinate frames.

As it was stated, in this case is 0º, which means that its cosine is 1 and its sine is 0. For the two joints d is taken to be 0 and a is the length of the link. Substituting these values simplifies the previous equation and yields to (4.4).

Now the position and the orientation of the coordinate frame at joint 1 is received, further transformations can be carried out.

Now there is a general formula for calculating the position of the finger tip with reference to the common base coordinate frame. (4.3) shows this, and it is very similar to (4.1). Matrix T contains the position and orientation of the tip in the base coordinate system.

It is useful to state that joint 2 has two transformation matrices and frames because of the two possible axis of rotation.

where X15 is the position vector of the fingertip in its local co-ordinate frame.

The position and orientation of all five fingertips with respect to the common base frame (the wrist), is given by (4.7) as a function of joint angles.

4.5.1. Grasping

4-20. ábra illustrates the transformations between finger-tip, contact and object coordinate frames [10]. Let us define homogeneous transformation matrices Ui R4x4 and Vi R4x4, where Ui transforms from the finger-tip coordinate frame Ti R4x4 to the contact frame Ci R4x4 and Vi R4x4 from the contact frame to the object frame Bi R4x4.The x- and y-axis of the contact coordinate frame Ci are tangents and the z-axis is perpendicular to the object surface curvature. (4-21. ábra). The transformation matrices i R6x6 and i R6x6 are defined such a way that i transforms finger-tip forces Tifi into forces Cifi in the contact coordinate frame and i transforms these to forces Bfi in the object reference frame B.

Figure 4.19.  Object Grasped by 3 Fingers

Figure 4.20.  Contact Point and Contact Frame

The relationship of coordinate transformations can be written as

Ci= Ti Ui B= Ci Vi= Ti Ui Vi Cifi =i Tifi Bfi=i Cifi=i i Tifi

(4.8)

where Ui, Vi are homogeneous transformations and i , i are force transformations [10]. The sensor glove has two types of sensor outputs, the torques actuated by the human operator R5x4 and the finger joint angle values R5x4. Ti are calculated from the forward kinematical equations using .

A handshake via Internet is an interesting type of telemanipulation [29]. The task is to realize a handshake between two people at two different corner of the world. Let one be the one the operator and the other the far environment reacting to the operator (hand) movement. The master device will be a hand-shape device in the operator’s hand. The slave device is identical. When the operator grasps the master device (see in REF _Ref364272598 \h a and b), the master device transforms the movement and force to the slave device. The slave device gives this movement to the far environment (the other person’s hand). This person gives a reaction force and a movement to the slave device. This information is transmitted to the operator via master device. The handshake will be realized between two people far from each other. The main challenge is compensation of the time delay.

Figure 4.39. Tele Handshaking Device: (a) Photo (b) Structure

4.7.1. Virtual Impedance with Position Error Correction

Virtual Impedance with Position Error Correction (VIPEC) is based on the concept of the impedance control. VIPEC is composed of two major parts: a virtual impedance model (VI) and a position error correction part. The virtual impedance model generates a motion trajectory for a manipulator based on input forces and its dynamics. The position error correction part is added to decrease the position difference of the master and slave.

4.7.1.1. A. 1 Virtual Impedance

In the case of a 1 DOF linear motion manipulator, it is assumed that a virtual mass, Mv, is attached to a manipulator arm. This virtual mass is supported by a virtual damper, Dv, and a virtual spring, Kv. The structure of the VI is shown in 4-41. ábra.

Figure 4.40.  One DOF linear motion manipulator with virtual impedance

When a force is applied to the virtual mass, the position of the virtual mass is modified according to the dynamics of VI. The dynamic equation of VI is as follow.

(4.45)

where F(t) is the force detected at the virtual mass and x(t) is the displacement of the virtual mass. The transfer function of the VI acts as a second order low pass filter to convert force into displacement. Thus the displacement is always smooth even if the force contains high frequency components.

4.7.1.2. Position Error Correction

The structure of a teleoperator with time delay utilizing the VIPEC is illustrated in 4-42. ábra. The PEC part (the shaded part) is added to the original approach [30] in order to improve the response of the teleoperator by reducing the position error between the master and slave positions.

Figure 4.41.  Virtual Impedance with Position Error Correction for a teleoperator system with time delay

In 4-42. ábra, Fodenotes the force that an operator applies to the master arm, and Fe denotes the reflected force when the slave arm contacts with the environment. xm and xmrdenote the position and the reference position of the master arm. xsand xsr denote the position and the reference position of the slave arm. Td represents a delay time in data transmission between the two sites. Af is a scalar unit, which can be varied depending on the size ratio of the slave and the master.

Each VI generates the reference motion trajectory for each manipulator arm based on the sum of two forces: the force from its own site and the force received from the other site. The data transmission from one site to the other site is delayed by Td . The reference motion trajectory from VI is sent to a servo controller to control the position of the manipulator arm.

The control approach can be used for the master and the slave arms, which are different in size by appropriately adjusting Af and the parameters of each VI. For example, when the slave is a micro robot, which is smaller than the master.

The dynamic equation of VI at the operator site is

(4.46)

The dynamic equation of VI at the remote site is

(4.47)

Without loss of generality, Af =1 and both VIs have the same parameters. When the parameters of VI are appropriately determined, the system should remain stable. In the steady state, Fo is approximately equal to Fe when Kv is set to a low value, and the master and the slave positions should be identical. However, when time delay is present in the teleoperator, the control approach VI without PEC part causes the position error between the master and slave positions, since there is no comparison between these both positions. To reduce this position difference, the PEC part is added in order to make a closed loop system. Therefore the position error can be reduced. The slave position, delayed by Td, is transmitted to the master site and compared to the master position. The difference between the two positions is sent through the PEC gain, and then modifies the reference motion trajectory of the master arm.

Control model of the Handshaking device

4-43. ábra shows the control system diagram of the master-slave system inside the tele-handshaking system with VIPEC approach. The armature inductance, La of each motor is neglected. The two linear motion motors represented by the master block and the slave block can be modeled as simple DC-motors with the transfer function.

(4.48)

where X and Ea are the Laplace transformed of motor shaft displacement and the applied armature voltage, Km is the motor gain constant, Tm is the motor time constant, and Kp is a constant to convert the angular displacement into displacement.

Figure 4.42.  Control diagram the Handshaking device

The servo controller is a proportional controller whose gain is G. The closed-loop transfer function of the manipulator arm and the controller is

(4.49)

Gain P represents PEC gain. For simplicity, the environment is modeled by a spring with stiffness constant Ke. Therefore the dynamics of the spring interacting with the slave arm is modeled as follows.

Fe (t) = Ke xs(t),

(4.50)

where xs is the displacement of the spring. The dynamics of the operator including the dynamic interaction between the operator and the master arm is approximated by

(4.51)

where Fop is the force generated by the operator's muscles. Mop, Bop and Kop denote mass, damper, and stiffness of the operator and the master arm respectively. Similarly to REF _Ref364274698 \h \* MERGEFORMAT , the displacement of the operator in (4.51) is represented by xm since it is also assumed that the operator is firmly grasping the master arm and never releases the master arm during the operation.

4.7.2. Experiment

Using the tele-handshake system, two human operators, operator A and operator B, shake hands with each other. When operator A tries to shake the hand of operator B by moving the master arm A, the slave arm A should move the same as the master arm A does. When the slave arm A contacts the palm of operator B, operator B can feel grasping force of operator A. Operator B also can grasp the virtual hand of operator A in the same way. Two operators succeeded to shake hands with each other and achieved force sensation through the tele-handshaking Device. Experimental results of simple VI and VIPEC methods are compared in 4-4. figure and 4-5. figure. In 4-4. figure, the time delay is negligible. The difference between the two methods is not significant. In both cases, the position of master and slave device are approximately the same. 4-5. figure illustrates the experimental results with 400 ms time delay. In both cases, the system is stable and the operators can feel grasping force. However, the position difference between the master and slave device is smaller in case of VIPEC method.

Figure 4.43. Experimental results of tele handshaking device without time delay (a) Results with VI and without PEC

Figure 4.44. Experimental results of tele handshaking device without time delay (b) Results with VIPEC

Figure 4.45. Experimental results of tele handshaking device with 400 ms time delay (a) Results with VI and without PEC

Figure 4.46. Experimental results of tele handshaking device with 400 ms time delay (b) Results with VIPEC

The Internet based telemanipulation is a relatively new and attractive research field. It is fruitful ground for applying new and sophisticated control methods to eliminate the time delay and nonlinear friction force, which are the main challenges in this filed. Of course, the total elimination is impossible, but a stable operation may be achieved. This paper presented a promising method, namely Virtual Impedance method with Position Error Correction. The effectiveness of this method has been proved by measurement results of a tele-handshaking device.

[1] T.B. Sheridan, “Telerobotics” Automatica, Vol. 25, No. 4, pp 487-507, 1989

R. C. Goertz and W. C. Thompson, „Electroically Controlled Manipulator” NUCLEONICS, Vol.12, No. 11, pp. 46-47, 1954.

[2] S. Lee, B. Bekey and A.K. Bejczy, “Computer Control of Space-borne Teleoperators with Sensory Feedback,” Proc. IEEE Int. Conf. On Robotics and Automation}, pp. 204-214, 1985.

[3] Barney Dalton, Ken Taylor: A Framework for Internet Robotics. Proceedings of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, WS2, pp. 15-22, 1998.

[4] S. Mizik, P. Baranyi, P. Korondi, and M. Sugiyama, “Virtual Training of Vector Function based Guiding Styles” Transactions on AUTOMATIC CONTROL and COMPUTER SCIENCE ISSN 1224/600X vol. 46(60) No.1 pp. 81-86. 2001.

[5] Handlykken M., Turner T., “Control systems analysis and synthesis for a six degree-of-freedom universal force-reflecting hand controller”, In proceedings of the IEEE Conference on Decision and Control, ‘80, vol 2, pp 1197-1205. 1980.

[6] Sheridan, T.B., “Telerobotic, Automation, and Human Supervisory Control” Cambridge, MA: The MIT Press, 1992.

[7] Burdea, G. and Coiffet, P.,“Virtual Reality Technology” New York: John Wiley & Sons, Inc, 1994.

[8] J. Dudragne, C. Andriot, R. Fournier and J. Vuillemey „A generalized bilateral control applied to master-slave manipulators” Proc of 20th Conference of International Society for Intelligent Research, pp. 435-442, 1989.

[9] Y. Kunii and H. Hashimoto, “Dynamic Force Simulator for Multifinger Force Display” IEEE trans. on IE, Volume 43, pp.74-

[10] Metin Sitti, Hideki Hashimoto: Two-Dimensional Fine Particle Positioning Using a Piezoresistive Cantilever as a Micro/Nano-Manipulator. IEEE International Conference on Robotics and Automation, pp. 2729-2735 1999.

[11] Thomas H.Massie, J.Kenneth Salisburg, ``The PhantomHaptic Interface: A Device for Probing Virtual Objects'', Proceedings of the 1994 ASME International Mechanical EngineeringCongress and Exhibition, Chicago, Vol. 55-1, pp.295-302

[12] Péter Korondi, Péter T. Szemes and Hideki Hashimoto, „Sliding Mode Friction Compensation for a 20 DOF Sensor Glove”, Transaction of ASME, Journal of Dynamic Systems, Measurement and Control, ISSN 0022-0434, December 2000, Vol. 122. P. 611-615.

[13] J. E. Colgate: “Robust Impedance shaping telemanipulation”. IEEE Transactions on Robotics and Automation, 9(4):373-384, Aug. 1993.

[14] Ren C. Luo, Tse Min Chen, Chih-Chen Yin: “Intelligent Autonomous Mobile Robot Control through the Internet”. ISIE 2000, Cholula, Puebla, Mexico, 2000 W.

[15] Burgard, A. Cremens, D. Fox, D. Hahnel, G. Lakemeyer, D. Schulz, W. Steiner, S. THRUN: “The Interactive Museum Tour-Guide Robot”. Proceedings of National Conference on Artificial Intelligence, pp. 11-18, 1998

[16] Reid Simmons: “Xavier: An Autonomous Mobile Robot on the Web”. Proceedings of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, WS2, pp. 43-48, 1998

[17] B. R. Farzin And K. Goldberg: “A Minimalist Telerobotic Installation on the Internet”. Proceedings of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, WS2, pp. 7-13, 1998

[18] M. R. Stein: “Painting on the World Wide Web: The PumaPaint Project”. Proceedings of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, WS2, pp. 37-42, 1998

[19] K. Kosuge, J. Kikuchi, K. Takeo: “VISIT: A Teleoperation System via Computer Network”. Proceedings of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, WS2, pp. 61-66, 1998

[20] H. HIRUKAWA And I. HARA: “The Web Top Robotics.” Proceedings of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, WS2, pp. 49-54, 1998

[21] R. Ferrell: “Remote Manipulation with Transmission Delay”. IEEE Transactions on Human Factors in Electronics, pp. 24-32, 1965

[22] Sukhan LEE, Hahk Sung LEE: “Advanced Telemanipulation under Time-delay”. SPIE OE/Technology ’92: Telemanipulator Technology, 1992

[23] A. K. Bejczy, W. S. Kim, S. C. Venema: “The Phantom Robot: Predictive Displays for Teleoperations with Time-delays”. IEEE International Conference on Robotics and Automation, pp. 546-551, 1990

[24] Won S. Kim, Blake Hannaford, Antal K. Bejczy: “Force-reflection and Shared Compliant Control in Operating Telemanipulators with Time-delay”. IEEE Transactions on Robotics and Automation, pp. 176-185, 1992

[25] Paul G. Backes: Multi-sensor Based Impedance Control for Task Execution. IEEE International Conference on Robotics and Automation, pp. 1245-1250, 1992

[26] W. R. Ferrell, T. B. Sharidan: “Supervisory Control of Remote Manipulation”. IEEE Spectrum, pp. 81-88, 1967

[27] Barney Dalton, Ken Taylor: “A Framework for Internet” Robotics. Proceedings of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, WS2, pp. 15-22, 1998

[28] S. Manorotkul and H. Hashimoto, “Virtual Impedance with Position Error Correction for Teleoperation with Time Delay,” Proc. 4th Int. Workshop on Advanced Motion Control}, Vol. 2, pp. 524-528, 1996.

[29] M. Otsuka, N. Matsumoto, T. Idogaki, K. Kosuge and T. Itoh, “Bilateral Telemanipulator System with Communication Time Delay Based on Forcesum-Driven Virtual Internal Model” Proc. IEEE Int. Conf. On Robotics and Automation, pp. 344-350, 1995.

[30] Smith O. J. M.: “A Controller to Overcome Dead Time”. Chem. Eng. Prog. Vol. 53, 217-219, 1959

[31] R. J. Anderson and M. W. Spong, “Bilateral Control of Teleoperators with Time Delay,” IEEE Trans. Automatic Control, Vol 34, pp. 494-501, 1989.

[32] Niemeyer, G., J-J.E. Slotine, “Stable Adaptive Teleoperation” IEEE trans. On Automatic Control, Vol. 16. No. 1. pp. 152-162. 1991.

[33] Young, K. D., “Manifold Based Feedback Design, Advances in Control and its Applications”, Lecture Notes in Control and Information Sciences, Springer-Verlag London, Tokyo, New York. 1995.

[34] Utkin, V. I., “Sliding Modes in Control and Optimization”, Springer-Verlag London, Tokyo, New York. 1992

[35] D. Lawrence, “Stability and Transpareny in Bilateral Teleoperation”, IEEE Transactions on Robotics and Automation, vol. 9, pp. 624-637, 1993.

There are many researches focused human robot interaction (HMI) in the last decade because service robots work at the same environment as people. This research takes in investigation of HMI from the ethological point of view, based on the 20,000 year old human-dog relationship [3]. The ethologists observing the dog’s behavior from the attachment point of view for the last fifteen years. Based on these observations a new type of artificial intelligent can be created that is able to attach to a specified owner.

Figure 5.1.  MogiRobi: the holonomic drive based ethological robot

The goal of this paper was to suggest mechanical and control methods for a social robot in the iSpace conception. The control methods could be implemented, and tested in different situations. MogiRobi controlled with simple orders from the fuzzy logic system and worked at the iSpace conception with promising results. It is an easily built and well designed adaptable structure for many application areas at social robotics. The robot can cooperate with dogs and humans also. At the scenarios the emotions of the robot always recognized. It’s proved that the mechanical and control methods of MogiRobi were appropriate for Human-Robot, Dog-Robot interaction. The suggested mechanical structure was worked successful during emotion recognition, and cognitive communication.

In recent years there has been an increased interest in the development of Human-Robot Interaction (HRI). Researchers have assumed that HRI could be enhanced if these intelligent systems were able to express some pattern of sociocognitive and socioemotional behaviour (e.g. REF _Ref361667391 \r \h \* MERGEFORMAT ). Such approach needed an interaction among various scientific disciplines including psychology, cognitive science, social sciences, artificial intelligence, computer science and robotics. The main goal has been to find ways in which humans can interact with these systems in a ‘natural’ way. Recently HRI has become very user oriented, that is, the performance of the robot is evaluated from the user’s perspective. This view also reinforces arguments that robots do not only need to display certain emotional and cognitive skills but also showing features of individuality. Generally however, most socially interactive robots are not able to support long-term interaction with humans, and the interest shown toward them wears out rapidly.

The design of socially interactive robots has faced many challenges. Despite major advances there are still many obstacles to be solved in order to achieve a natural-like interaction between robots and humans. The ‘uncanny valley’ effect: Mori REF _Ref361656179 \r \h \* MERGEFORMAT assumed that the increasing similarity of robots to humans will actually increase the chances that humans refuse interaction (will be frightened from) very human-like agents. Although many take this effect for granted only little actual research was devoted to this issue. Many argue that once an agent passes certain level of similarity, as it is the case in the most recent visual characters in computer graphics, people will treat them just as people REF _Ref361667499 \r \h \* MERGEFORMAT . However, in the case of 3D robots, the answer is presently less clear, as up do date technology is very crude in reproducing natural-like behaviour, emotions and verbal interaction. Thus for robotics the uncanny valley effect will present a continuing challenge in the near future.

In spite of the huge advances in robotics current socially interactive systems fail both with regard to motor and cognitive capacities, and in most cases can interact only in a very limited way with the human partner. This is a major discrepancy that is not easy to solve because there is a big gap between presently available technologies (hardware and software) and the desire for achieving human-like cognitive and motor capacities. As a consequence recent socially interactive robots have only a restricted appeal to humans, and after losing the effect of novelty the interactions break down fast.

The planning and construction of biologically or psychologically inspired robots depends crucially of the current understanding of human motor and mental processes. However, these are one of the most complex phenomena of life. Therfore it is certainly possible that human mental models of abilities like ‘intention’, ‘human memory’ etc., which serve at present as the underlying concepts for control socially interactive robots, will be proved to be faulty.

Because of the goal of mimicking a human, socially interactive robots do not utilize more general human abilities that have evolved as general skills for social interaction. Further, the lack of evolutionary approach in conceptualizing the design of such robots hinders further development, and reinforces that the only goal in robotics should be the produce ‘as human-like as possible’ agents.

In order to overcome some of the challenges presented above ethologically inspired HRI models can be applied. The concept of ethologically inspired HRI models allows the study of individual interactions between animals and animals and humans. This way of handling Human-Robot Interaction is based on the concept that the robot acts like an animal companion to human. According to this paradigm the robot should not be molded to mimic the human being, and form human-to-human like communication, but to follow the existing biological examples and form inter-species interaction. The 20.000 year old human-dog relationship REF _Ref361667925 \r \h \* MERGEFORMAT is a good example for this paradigm of the HRI, as interaction of different species. One good reason of this approach in HRI is the lack of the ‘uncanny valley’ effect REF _Ref361656179 \r \h \* MERGEFORMAT , i.e. there is no need to buit robots which have any similarities to humans.

Such robots do not have to rely on the exact copy of human social behaviour (including language, etc.) but should be able to produce social behaviours that provide a minimal set of actions on which human-robot cooperation can be achieved. Such basic models of robots could be improved with time making the HRI interaction more complex.

In ethological modeling, mass of expert knowledge exists in the form of expert’s rules. Most of them are descriptive verbal ethological models. The knowledge representation of an expert’s verbal rules can be very simply translated to the structure of fuzzy rules, transforming the initially verbal ethological models to a fuzzy model. On the other hand, in case of the descriptive verbal ethological models, the ‘completeness’ of the rule-base is not required (thanks to the descriptive manner of the model), which makes implementation difficulties in classical fuzzy rule based systems. For that reason for fuzzy modell we use low computational demand Fuzzy Rule Interpolation (FRI) methods like FIVE REF _Ref385680017 \r \h . The application of FRI methods fits well the conceptually sparse rule-based (‘incomplete’) structure of the existing descriptive verbal ethological models.

The main benefit of the FRI method adaptation in ethological model implementation is the fact, that it has a simple rule-based knowledge representation format. Because of this, even after numerical optimization of the model, the rules remain ‘human readable’, and helps the formal validation of the model by the ethological experts. On the other side due to the FRI base, the model has still low computational demand and fits directly the requirements of the embedded implementations.

For implementing ethologically inspired HRI models the classical behaviour-based control structure is suggested, described in the followings.

The main building blocks of Behaviour-based Control (BBC, a comprehensive overview can be found in REF _Ref358152640 \r \h ) are the behaviour components themselves. The behaviour components can be copies of typical human or animal behaviors, or can be artificially created behaviours. The actual behaviour response of the system can be formed as one of the existing behaviour component, which gives the best match to the actual situation, or a fusion of the behaviour components based on their suitability for the actual situation. Encoding the behaviour components can be realized with e.g. simple reflexive agents, which assign an output response to each input situation.

In case when more than one behaviour components are simultaneously competing for the same ‘actuator’ the aggregation or selection of the behaviour components is necessary. Handling multiple behaviour components in a BBC system can be done in two ways. The first is the competitive way, when the behaviour components are assigned with priorities, and the behaviour component with the highest priority takes precedence, while the behaviours with lower priorities are simply ignored. The second is the cooperative way when the outputs are fusioned based on various criteria.

Figure 6.1. Diagram of the fuzzy automaton

This structure has two main tasks. The first is a decision of which behaviour is needed in an actual situation, and the levels of their necessities in case of behaviour fusion. The second is the way of the behaviour fusion. The first task can be viewed as an actual system state approximation, where the actual system state is the set of the necessities of the known behaviours needed for handling the actual situation. The second is the fusion of the known behaviours based on these necessities. In case of the suggested fuzzy behaviour based control structures both tasks are solved by FRI systems. If the behaviours are also implemented on FRI models, the behaviours together with the behaviour fusion modules form a hierarchical FRI system.

The application of FRI methods in direct fuzzy logic control systems gives a simplified way for constructing the fuzzy rule base. The rule base of a fuzzy interpolation-based model, is not necessarily complete, it could contain the most significant fuzzy rules only without risking the chance of having no conclusion for some of the observations. The model always gives a usable conclusion even if there are no rules defined for the actual observations. In other words, during the construction of the fuzzy model, it is enough to concentrate on the main actions (rules could be deduced from the others could be intentionally left out from the model, radically simplifying the rule base creation, saving time consuming work.).

For demonstrating the main benefits of the FRI model in behaviour-based control, here the focus is only on the many cases most heuristic part of the structure, on the behaviour coordination. The task of behaviour coordination is to determine the necessities of the known behaviours needed for handling the actual situation.

In the suggested behaviour-based control structure for achieving the decision related to the relevance of the behaviour components this work adapts the concept of the finite state fuzzy automaton REF _Ref361677815 \r \h \* MERGEFORMAT (see Fig. 6-1 and the next chapter for more details), where the state of the finite state fuzzy automaton is the set of the suitabilities of the component behaviours. This solution is based on the heuristic, that the necessities of the known behaviours for handling a given situation can be approximated by their suitability. And the suitability of a given behaviour in an actual situation can be approximated by the similarity of the situation and the prerequisites of the behaviour. (Where the prerequisites of the behaviour is the description of the situations where the behaviour is applicable). This case instead of determining the necessities of the known behaviours, the similarities of the actual situation to the prerequisites of all the known behaviours can be approximated.

The system consists (see Fig. 6-1) of not only the fuzzy automaton but the behaviour fusion component and various component behaviours implemented on fuzzy logic controllers (FLC). The state variables characterize the relevance of the component behaviours. The state-transition rule base of the automaton applies fuzzy rule interpolation for state-transition evaluation. The previous states are fed back to the automaton and the conclusion given by the automaton is used as weights in the behaviour fusion component for determining the final conclusion of the BBC. The conclusion of the fuzzy automaton will be the new system state for the next step of the behaviour fusion. The behaviour fusion component can be also implemented by fuzzy reasoning (e.g. using fuzzy rule interpolation), or simply as a weighted sum. The symptom evaluation (from the terminology of fault classification) components provide a kind of preprocessing for the automaton based on the gathered observations. These components also can employ FRI techniques.

Therefore the first step of the system state approximation is determining the similarities of the actual situation to the prerequisites of all the known behaviours. The task of symptom evaluation is basically a series of similarity checking between an actual symptom (observations of the actual situation) and a series of known symptoms (the prerequisites – symptom patterns – of the behaviour components). These symptom patterns are characterizing the systems states where the corresponding behaviours are valid. Based on these patterns, the evaluation of the actual symptom is done by calculating the similarity values of the actual symptom (representing the actual situation) to all the known symptoms patterns (the prerequisites of the known behaviours). There are many existing methods for fuzzy logic symptom evaluation. For example fuzzy classification methods e.g. the Fuzzy c-Means fuzzy clustering algorithm REF _Ref361758798 \r \h can be adopted, where the known symptoms patterns are the cluster centers, and the similarities of the actual symptom to them can be fetched from the fuzzy partition matrix. On the other hand, having a simple situation, the fuzzy logic symptom evaluation could be an FRI model too.

One of the main difficulties of the system state approximation is the fact, that most cases the symptoms of the prerequisites of the known behaviours are strongly dependent on the actual behaviour of the system. Each of the behaviours has its own symptom structure. In other words, for the proper system state approximation the approximated system state is needed itself. A very simple way of solving this difficulty is the adaptation of fuzzy automaton. This case the state vector of the automaton is the approximated system state, and the state-transitions are driven by fuzzy reasoning (‘Fuzzy Reasoning (state-transition rule base)’ on Fig. 6-1), as a decision based on the previous actual state (the previous iteration step of the approximation) and the results of the symptom evaluation.

As mentioned previously, the structure of the proposed model follows the behavior-based control concept REF _Ref358152640 \r \h , i.e. the actual behavior of the system is formed as a fusion of the known component behaviors appeared to be the most appropriate in the actual situation. For behavior coordination in the applied model the concept of the ‘Fuzzy Automaton’ is adapted. Numerous versions and understanding of the fuzzy automaton can be found in the literature (a good overview can be found in REF _Ref286655753 \r \h ). The most common definition of Fuzzy Finite-state Automaton (FFA, summarized in REF _Ref286655753 \r \h ) is defined by a tuple (according to REF _Ref363657503 \r \h , REF _Ref286655753 \r \h and REF _Ref288780404 \r \h ):

where Q is a finite set of states, Q={q1,q2,...,qn}, Σ is a finite set of input symbols, Σ={a1,a2,...,am}, is the (possibly fuzzy) start state of , Z is a finite set of output symbols, Z={b1,b2,...,bn}, is the fuzzy transition function which is used to map a state (current state) into another state (next state) upon an input symbol, attributing a value in the fuzzy interval [0,1] to the next state, and is the output function which is used to map a (fuzzy) state to the output.

Extending the concept of FFA from finite set of input symbols to finite dimensional input values turns to the following:

where S is a finite set of fuzzy states, , X is a finite dimensional input vector, , is the fuzzy start state of , Y is a finite dimensional output vector, , is the fuzzy state-transition function which is used to map the current fuzzy state into the next fuzzy state upon an input value, and is the output function which is used to map the fuzzy state and input to the output value. See e.g. on Fig. 6-2.

In case of fuzzy rule based representation of the state-transition function , the rules have dimensional antecedent space, and n dimensional consequent space. Applying classical fuzzy reasoning methods, the complete state-transition rule base size can be approximated by the following formula:

where n is the length of the fuzzy state vector S, m is the input dimension, i is the number of the term sets in each dimensions of the state vector, and j is the number of the term sets in each dimensions of the input vector.

According to (6.3) the state-transition rule-base size is exponential with the length of the fuzzy state vector and the number of the input dimensions. Applying FRI methods for the state-transition function fuzzy model can dramatically reduce the rule base size. See e.g. REF _Ref358152171 \r \h , where the originally exponential sized state-transition rule base of a simple heuristical model turned to be polynomial thanks to the FRI.

In case of direct application of the suggested FRI based Fuzzy Automaton for behavior-based control structures, the output function can be decomposed to parallel component behaviors and an independent behavior fusion. In this case the structure of the above introduced fuzzy automaton can turn to a very similar form as it is expected in behavior-based control (see Fig. 6-3 and Fig. 6-1). Some more details of the model implementation can be found in REF _Ref366247252 \r \h , REF _Ref366247254 \r \h and REF _Ref366247256 \r \h .

Figure 6.2. FRI based Fuzzy Automaton.

Figure 6.3. FRI behaviour-based structure

The ethological test procedure presented here has been developed for studying the affiliative relationship between a dog and its owner. The procedure is made up from seven episodes, each lasting 2 minutes, where the dog is in different situations: first with the owner, then with a stranger, or alone (according to a pre-defined protocol). Through the test, the dog’s behavior is evaluated mainly focusing on dogs’ responses related to their proximity seeking with the owner REF _Ref307389263 \r \h .

In the first episode of the test, the dog and the owner are in the test room. First the owner is passive, and then he/she stimulates playing. The second episode is where the stranger comes into the room and starts to stimulate play with the dog. Then the owner leaves the room, so in the third episode the dog is separated from the owner; the stranger tries to play with the dog then sits for a while and offers petting. The owner comes back in the beginning of the fourth episode, this is the first reunion. Meanwhile the stranger leaves the room and the dog is with the owner for two minutes. Then the owner also leaves, so in the fifth episode the dog is alone in the room. In the next episode the stranger returns and tries to stimulate playing or comfort the dog by petting it. The returning of the owner marks the beginning of the last episode. The stranger leaves the room, the owner interacts with the dog for two minutes and the test ends.

During evaluation, ethologists record pre-defined behavioral variables for describing the dogs’ responses related to both the owner and the stranger and they analyze data to reveal significant differences in behaviors showed towards the two persons.

This complex ethological model has been implemented based on the presented structure. The agent controlled by the model is the representation of the dog, other participants in the test are behaving according to a programmed scenario or are controlled by a human operator. This agent can execute a set of behaviours (moving towards specified objects, pick up or drop the toy object, wag its tail, etc.), some of them can be active and executed in parallel in the same time, but most of them are blocking each other, this latter situation is handled by the behaviour fusion component.

The model is basically designed for simulating the dog’s behaviour in the test described in REF _Ref307389263 \r \h \* MERGEFORMAT , but the model is flexible enough to react in an ethologically correct way for arbitrary situations. A simplified block diagram of the structure of a possible simulation framework is shown in Fig. 6-4.

Figure 6.4. Structure of the simulation

A screenshot of a possible simulation framework is shown in Fig. 6-5, where a room can be seen viewed from above, where the interactions take place. The humans are symbolized with squares, the blue square represents the owner of the dog, the magenta square represents a human unfamiliar (stranger) to the dog. The dog is represented by two circles, the large circle is the body of the dog, the smaller circle is the head of the dog, which is useful for visualizing the direction of the dog A thin green line on the dog’s body represents the tail of dog, which can change its length and also its angle (tail wagging). The color of the dog is dynamically changing showing the actual anxiety level of the dog. The room has a door where the human participants can leave or enter the room, the dog cannot go outside. Also a toy object is present (small blue circle, with a hole inside), which can be picked up and dropped by all the participants.

In the followings some example behaviours from the system and their definitions are presented.

The first example behaviour set is built upon two separate component behaviours, namely ‘DogExploresTheRoom’ and ‘DogGoesToDoor’. The ‘DogExploresTheRoom’ is an exploration dog activity, in which the dog ‘looks around’ in an unknown environment (see the track marked with white in Fig. 6-6). The ‘DogGoesToDoor’ is a simple dog activity, in which the dog goes to the door, and than stands (sits) in front of it (waiting for its owner to show up again). The definition of the related state-transition FRI rules of the fuzzy automaton acts as behaviour coordination in this example.

Figure 6.5. Screenshot of the simulation application

Figure 6.6. A sample track induced by the exploration behaviour component

The states concerned are the following:

As a possible rule base structure for the state-transitions of the fuzzy automaton, the following is defined (considering only the states and behaviours for this example):

If OwnerInTheRoom=False Then DogMissTheOwner=Increasing

If OwnerInTheRoom=True Then DogMissTheOwner=Decreasing

If OwnerToDogDistance=Small And Human2ToDogDistance=High Then DogAnxietyLevel=Decreasing

If OwnerToDogDistance=High And Human2ToDogDistance=LowThen DogAnxietyLevel=Increasing

State-transition rules related to the going to the door mood (state) of the Dog:

If OwnerInTheRoom=False And DogMissTheOwner=High Then DogGoesToDoor=High

If OwnerInTheRoom=True Then DogGoesToDoor=Low

State-transition rules related to the room exploration mood (state) of the Dog:

If DogAnxietyLevel=Low And OwnerStartsGame=False And ThePlaceIsUnknown=High Then DogExploresTheRoom=High

If ThePlaceIsUnknown=Low Then DogExploresTheRoom=Low

If DogAnxietyLevel=High Then DogExploresTheRoom=Low

The text in italic represent the linguistic terms (fuzzy sets) of the FRI rule base.

Also it should be noted that the rule base is sparse, containing the main state-transition FRI rules only.

A sample run of the example is introduced in Fig. 6-6 and in Fig. 6-7. At the beginning of the scene, the owner is in the room and the stranger is outside. The place is unknown for the dog (‘ThePlaceIsUnknown=High’ in the rule base). According to the above rule base, the dog starts to explore the room (see Fig. 6-6). After a little while the owner of the dog leaves the room, and then the stranger enters and stays inside (dashed white tracks in Fig. 6-7). As an effect of the changes (according to the above state-transition rule base), the anxiety level of the dog and the ‘Missing the owner’ state is increasing and as a result, the dog goes to and stays at the door (white track in Fig. 6-7), where its owner has left the room. See the state changes in Fig. 6-8.

Figure 6.7. A sample track induced by the ‘DogGoesToDoor’ behaviour component

Figure 6.8. Some of the state changes during the sample run introduced in

Fig. 68 shows the values of four state variables on a time scale (iteration number). Two immediate changes can be seen which are both caused by events related to the movement of the owner. First when the dog realizes that the owner is heading for the door (around the 200th iteration) the exploration rate drops rapidly and the rate of the ‘DogGoesToOwner’ behaviour component is high, till the owner finally leaves the room (around the 250th iteration). When the owner had left the room, the‘DogGoesToDoor’ behaviour component will be active, causing the value of the ‘Missing the owner’ state to increase. At the same time the anxiety level of the dog also increases with the leaving of the owner.

In the second example the ‘DogGoesToOwner’ behaviour will be presented in details from another point of view than the previous example. As its name suggests, this behaviour component is responsible for determining the need to go to the owner for the dog. The description of this behaviour component is more complex than the rule base in the previous example. The complete rule base for the ‘DogGoesToOwner’ behaviour is shown in REF _Ref364164389 \h , where the last column means the consequent part, all the other columns are antecedent dimensions (the ‘x’ value in the rule description means that the value of that antecedent is omitted while reasoning – does not have effect in the rule). The explanation of the terms used in the rule base is the following (‘observation’ in this case means that the value is gathered (measured) directly from the environment, ‘previous state’ means that the value was calculated in the previous iteration as a conclusion of a rule base):

dgtd – previous state – dog goes to door rate (zero-large) – see Fig. 6-12 for the corresponding fuzzy partition

dgto – previous state – dog goes to owner rate (zero-large) – see Fig. 6-12 for the corresponding fuzzy partition

oir – observation – owner is inside (true-false) – see Fig. 6-9 for the corresponding fuzzy partition

ddo – observation – distance between dog and owner (zero-large) – see Fig. 6-10 for the corresponding fuzzy partition

dgtt – previous state – dog goes to toy rate (zero-large) – see Fig. 6-9 for the corresponding fuzzy partition

dgro – previous state – dog greets owner rate (zero-large) – see Fig. 6-9 for the corresponding fuzzy partition

dpmo – previous state – dog’s playing mood rate with the owner (zero-large) – see Fig. 6-9 for the corresponding fuzzy partition

dpms – previous state – dog’s playing mood rate with the stranger (zero-large) – see Fig. 6-9 for the corresponding fuzzy partition

danl – previous state – dog’s anxiety level (zero-large) – see Fig. 6-11 for the corresponding fuzzy partition

ogo – observation – owner is going outside (true-false) – see Fig. 6-9 for the corresponding fuzzy partition

dgto (consequent) – next state – dog goes to owner rate (zero-large) (output)

The rules can be interpreted as the following textual description, in the corresponding order:

dog goes to door do not go to owner

dog should greet the owner go to owner

dog is nervous go to owner

dog's playing mood with owner is high go to owner

owner is going out of the room go to owner (follow the owner)

dog is calm and dog is not playing and not greeting do not go to owner

owner outside do not go to owner (can’t go to the owner)

dog is going to the toy do not go to owner

dog's playing mood with stranger is high do not go to owner

already going to owner and owner is near do not go to owner (arrived at the owner)

already going to owner and owner is far going to owner (keep on going)

The presented rule base is suitable to define a complex behaviour component with only eleven rules in spite of the relatively high antecedent dimension count (ten variables) thanks to the usage of FRI. In Fig. 6-9 through Fig. 6-12 the fuzzy partitions used for the various antecedents are presented. From top to bottom the curves in the mentioned figures show the membership functions, the scaling functions REF _Ref385687747 \r \h and the reconstructed membership functions (based on the scaling function). As it can be seen most of the fuzzy partitions are simple (six of the nine antecedents use the same simple fuzzy partition), consisting of only two triangular fuzzy sets with their cores at 0 and 1, and steepness of 1 on the right and the left side respectively. This is true also for the other rule bases in the model, most of the antecedent variables are defined by fuzzy partitions exactly like the one shown in Fig. 6-9.

Most of the behaviour components defined in the system are responsible for coordinating the movement (heading direction) of the dog, which must be aggregated by some kind of behaviour fusion. The simulation uses a simple weighted sum of the currently required behaviour components for determining the movement vector of the agent. Other behaviour components can be executed in parallel with movement related components. Also there are rule bases which are responsible for calculating the rate of behaviour components which are not directly executed, but instead their values are used by other rule bases as input values (in a hierarchical manner). E.g. the ‘DogGreetsOwner’ component is not executed directly, as shown in REF _Ref364164389 \h \* MERGEFORMAT . the consequent of ‘DogGreetsOwner’ is used as an input value in the rule base of the ‘DogGoesToOwner’ behaviour component, hence the actual output behaviour will be triggered by the activation of the ‘DogGoesToOwner’ behaviour component.

The model can be customized for different types of dogs. The system provides three factors for customization (‘Parameters’ in Fig. 6-4), which can be set in the interval [0,1]: ‘Anxiety’, ‘Acceptance’, and ‘Attachment’. The ‘Anxiety’ parameter defines the level of anxiety of the dog on a scale between calm and nervous, ‘Acceptance’ defines the level of the dog’s acceptance towards the stranger (between friendly and unfriendly), ‘Attachment’ defines the level of attachment of the dog to its owner (between weak and strong attachment). In fact these parameters are not directly used in the behaviour description rule bases. Secondary variables are calculated first based on the three parameters. Then these secondary parameters are used in the corresponding fuzzy rule bases. The three external parameters (‘Anxiety’, ‘Attachment’, ‘Acceptance’) mentioned earlier are used as antecedents in various rule bases, which have the task of determining the values of secondary parameters. These secondary parameters are then directly used as antecedents in the rule bases of the fuzzy automaton driving the simulation. These calculated secondary parameters are the following:

Play without the owner (when the owner is outside)

Do not play with the stranger (even when it could be possible)

Run in circles (caused by anxiety) when stranger is inside

Run in circles (caused by anxiety) when the dog is alone

Base rate of greeting the stranger

Base rate of greeting the owner

Base rate of following the stranger, when he/she is leaving the room

Base rate of greeting the owner, when he/she is leaving the room

Base rate of stress (anxiety)

Base rate of standing at the door

Base rate of standing at the door, even when the owner is inside

Base rate of passivity

This hierarchical strategy is useful for minimizing the input dimensions of the rule bases (one input variable is used instead of three input values in this case), hence fewer rules are required. The values of the secondary parameters are also calculated using fuzzy rule interpolation, each secondary parameter having its own rule base for determining the exact value based on the three input parameters. These are calculated at the time of initialization before starting the simulation, but can be also adjusted while the simulation is running, allowing the operator to switch between different types of dogs in real-time.

A distinct, but important part of the framework is the so called observer module (see Fig. 6-4), which has the task of evaluating the model (the correctness of the constructed rule bases) itself. The observer module only gets information from the environment (just what an outsider would see), no data about the inner states of the fuzzy automaton or behaviour weights are supplied to the observer. Based on these observations the observer module is able to determine what kind of activity is going on in the simulation. It can distinguish among ‘playing’, ‘exploring’, ‘standing at door’ and ‘passive’ states. If the dog is in contact (close the human and looking at the human) with the owner or the stranger, that is also measured and logged. Furthermore a special scoring system developed by ethologists used for evaluating the separation and reunion episodes (human leaves dog / human comes back) is also implemented in the observer module. Also the overall time spent in the various states along the different episodes is measured. A summary of the measurements and the scores is calculated at the end of the simulation, which can be directly used by ethologist experts to evaluate the simulated dog in the same manner as they would with real dogs.

Figure 6.9. Fuzzy partition of the following terms: dgro - dog greets owner, dpmo - dog’s playing mood with the owner, dpms - dog’s playing mood with the stranger, dgtt - dog goes to toy, dgtd - dog goes to door, oir - owner is inside, ogo - owner is going outside

Figure 6.10. Fuzzy partition of the term ddo (distance between dog and owner)

Figure 6.11. Fuzzy partition of the term danl (dog’s anxiety level)

Figure 6.12. Fuzzy partition of the term dgto (dog is going to owner) and dgtd (dog is going to the door)

When one would like to examine the current achievements and theories, it worth to look back to the past and evaluate, how our predecessors interpreted the same problem. Below, a short summary is given concentrated only on the development of friction models, from the middle ages till the mid 20th century.

8.1.1. The first friction model

It is interesting, but probably not that surprising, that the first friction model came from Leonardo Da Vinci (1452-1519). Leonardo measured the force of friction between objects on both horizontal and inclined surfaces and he soon recognized the difference between rolling and sliding friction and the beneficial effect of lubricants.

As Figure 8-1 shows, Leonardo was curious of the influence of the apparent area. He wrote to his notebooks: “The friction made by the same weight will be of equal resistance at the beginning of its movement although the contact may be of different breadths and lengths.” The above observations are entirely in accord with the first two laws of friction. (see chapter 1.4.9)

Leonardo introduced first the coefficient of friction as the ratio of the force of friction and the normal load, however he assumed, that for smooth surfaces “every frictional body has a resistance of friction equal to one quarter of its weight.”

Figure 8.1.  Leonardo da Vinci’s studies about the influence of apparent area upon the force of friction.

8.1.2. A more scientific approach

Robert Hook (1635-1703), inspired by the famous Stevin’s chariot [141] – a ship-like vehicle on wheels with masts, sails and other convenient riggings – discussed rolling friction and recognized two components: “…The first and chiefest, is the yielding, or opening of the floor, by the weight of the wheel so rolling and pressing; and the second, is the sticking and adhering of the parts of it to the wheel;…”

He further distinguished between the role of recoverable and non-recoverable deformation, and also discussed the question of sticking and adhesion.

Much of the scientific work of friction was initiated by Guillaume Amontons (1663-1705) of France. His experiments and interpretation of results were published in a classical paper presented to the Académie Royale in 1699.

The apparatus used by Amontons in his experiment is shown in Figure 8-2. The test specimens of AA and BB were loaded together by various springs denoted by CCC, while the force require to overcome friction and initiate sliding was measured on the spring balance D.

Figure 8.2.  Amonton’s sketch of his apparatus used for friction measurement in 1699.

Friction is the tangential reaction force between two surfaces in contact. Physically these reaction forces are the results of many different mechanisms, which depend on contact geometry and topology, properties of the surface materials of the bodies, displacement and relative velocity of the bodies and presence of lubrication.

Contact surfaces are very irregular at microscopic level (see in Figure 8-3).

Figure 8.3.  Contact surfaces at microscopic level

It can be visualized as two rigid bodies that make contact through elastic bristles (see in Figure 8-4).

Figure 8.4.  Visualization of rigid bodies in contact

When tangential force is applied, the bristles will deflect like springs and dampers, giving raise to microscopic displacement (stick). If the displacement becomes too large, the bristles break (see in Figure 8-5).

Figure 8.5.  The breaking of bristles

At this break-away displacement macroscopic sliding (slip) starts. The friction force where this occurs is called break-away force. The average deflection of the bristles corresponds to the internal state of the dynamic friction models. [10].

In dry sliding contacts between flat surfaces friction can be modeled as elastic and plastic deformation forces of microscopical asperities in contact. The asperities each carry a part fi of the normal load FN. If we assume plastic deformation of the asperities until the contact area of each junction has grown large enough to carry its part of the normal load, the contact area of each asperity junction is:

where H is the hardness of the weakest material of the bodies in contact. The total contact area thus can be written

This relation holds even with elastic junction area growth, provided that H is adjusted properly. For each asperity contact the tangential deformation is elastic until the applied shear pressure exceeds the shear strength τy of the surface materials, when it becomes plastic. In sliding the friction force thus is:

and the friction coefficient:

The friction coefficient is independent on the normal load or the velocity in this case. Consequently it is possible to manipulate friction characteristics by deploying surface films of suitable materials on the bodies in contact. [5].

As with many problems in the physical sciences, friction can be approached by using forces or by using energy. The best known model for friction that utilizes the force approach has been postulated by Bowden and Tabor. They suggested that surfaces adhere to form junctions and the friction force is directly related to the force needed to shear these junctions. The normal load FNis supported by the total contact area A of the deformed surfaces. If τjct is the shear strength of the adhesive junctions, then the frictional force:

where PY is the yield pressure. Thus the coefficient of friction:

Estimates of the friction coefficients from this expression have been reported as μ=0.2-0.3.

This model is based on the assumption of homogeneous and isotropic materials and uses a limited number of variables. Basic material characteristics tend to be neglected in most available model of friction. Microstructural information which significantly influences material behaviour is generally omitted.

Some plastic deformations always occur near the surfaces of sliding materials. Therefore plastic deformation and associated parameters such as microstructure as mentioned before are important parameters in setting up a more realistic friction model. P. Heilman and D. A. Rigney have presented an energy based model of friction. It is assumed that the frictional work is approximately equal to the plastic work done, or:

where τ(γ) is the shear stress, γ is the shear strain and V is the deformed volume near to the surface. This approach is applicable for metals and for some polymers and ceramics, which can deform plastically under compressive loading.

In dry rolling contact, friction is the result of a non-symmetric pressure distribution in the contact. The pressure distribution is caused by elastic hysteresis in either of the bodies, or local sliding in the contact. For rolling friction the friction coefficient is proportional to the normal load as

The elastoplastic characteristics of dry friction can be described by hysteresis theory.

Other physical mechanisms appear when lubrication is added to the contact. For low velocities, the lubricant acts as a surface film, where the shear strength determines the friction. At higher velocities at low pressures a fluid layer of lubricant is built up in the surface due to hydrodynamic effects. Friction is then determined by shear forces in the fluid layer. These shear forces depend on the viscous character of the lubricant, as well as the shear velocity distribution in the fluid film. Approximate expressions for the friction coefficient exist for a number of contact geometries and fluids. At high velocities and pressures the lubricant layer is built up by elastohydrodynamic effects. In these contacts the lubricant is transformed into an amorphous solid phase due to the high pressure. The shear forces of this solid phase turn out to be practically independent of the shear velocity. The shear strength of a solid lubricant film at low velocities is generally higher than the shear forces of the corresponding fluid film built up at higher velocities. As a result the friction coefficient in lubricated systems normally decreases when the velocity increases from zero. When the thickness of the film is large enough to completely separate the bodies in contact, the friction coefficient may increase with velocity as hydrodynamic effects become significant. This is called the Stribeck effect.

Film thickness is a vital parameter in lubricated friction. The mechanisms underlying the construction of the fluid film includes dynamics, thus suggesting a dynamic friction model. Contamination is another factor that adds complication. The presence of small particles of different materials between the surfaces give rise to additional forces that strongly depend on the size and material properties of the contaminants.

This short expose of some friction mechanisms illustrate the difficulty in modeling friction. There are many different mechanisms. To construct a general friction model from physical first principles is simply not possible. Approximate models exist for certain configurations. What we look for instead is a general friction model for control applications, including friction phenomena observed in those systems.

Static models are mostly simple mathematical representations of certain frictional behaviors. They do not deal with friction as a complex process, only with some of its sub processes, such as frictional force at rest i.e. static friction, friction as a function of steady-state velocity, etc.

8.3.1. Coulomb friction

The easiest and probably the most well known model is the so-called Coulomb friction model. Though it greatly over simplifies the frictional phenomena it is widely used in the engineering world, when dynamic effects are not concerned. Also, the Coulomb model is a common piece of all more developed models. The Coulomb friction force Fc is a force of constant magnitude, acting in the direction opposite to motion:

Figure 8.6.  Coulomb friction characteristic

When:

(8.11)

(8.12)

Where FN is the force pressing surfaces together and μ is the frictional factor. μ is determined by measurements under certain conditions. One of the biggest problems of the Coulomb model is, that it cannot handle the environment of zero velocity, hence the properties of motion at starting or zero velocity crossing, i.e. static and rising static friction. To apply the model for those cases a μ0 factor has been introduced. In case of starting the motion it replaces μ in Equation (8.12) until the process arrives to steady state. The values of μ and μ0 can be found in any major physics or engineering tabulations for different material pairs in both dry and lubricated conditions. The first tabulations of those kinds date back to the beginning of 18th century. Note, that though this is the simplest and fastest way to calculate the frictional force, this method can suffer from the error up to 20%.

8.3.2. Viscous friction

The viscous friction element models the friction force as a force proportional to the sliding velocity:

When:

(8.13)

Figure 8.7.  Viscous friction combined with Coulomb friction

8.3.3. Static friction

Static friction is a force of constraint:

When

(8.14)

where Fs is the static friction force, an additional upload on Fc, the Coulomb friction force.

Figure 8.8.  Viscous friction combined with Coulomb friction and static friction

8.3.4. Stribeck effect

The Stribeck curve is a more advanced model of friction as a function of velocity. Although it is still valid only in steady state, it includes the model of Coulomb and viscous friction as built-in elements.

Figure 8.9.  Stribeck friction characteristic

Complex models are mostly built upon the simple model elements with additional features, which can handle dynamic behavior. Yet the science of tribology is still far from having a complete picture on how friction works in all extent, therefore these models are mostly based on empirical experiences, rather than deep scientific background.

8.4.2. Dynamic models

8.4.2.1. Seven-parameters friction model

A popular approach is simply sewing together the already existing model elements. This is how the seven-parameters model works. Each of the seven parameters of the model represents a different friction phenomenon as presented in Table 8.2. They value principally depends upon the mechanism and the lubrication conditions, but typical values may be offered based on the works of [6], [27], [57], [75], [85], [105].

The friction is given by:

Not sliding (presliding displacement)

(8.23)

Sliding (Coulomb + viscous + Stribeck curve friction with frictional memory)

(8.24)

Rising static friction (frictional level at breakaway):

(8.25)

Where:

Table 8.1.  Parameters

Ff()	is the instantaneous friction force
FC	(*) is the Coulomb friction force
Fv	(*) is the viscous friction force
Fs	is the magnitude of the Stribeck friction (frictional force at breakaway is FC+Fs)
Fsa	is the magnitude of the Stribeck friction at the end of the previous sliding period
Fs	(*) is the magnitude of Stribeck friction after a long time at rest (with a slow application of force)
kt	(*) is the tangential stiffness of the static contact
vs	(*) is the characteristic velocity of Stribeck friction
L	(*) is the time constant of frictional memory
	(*) is the temporal parameter of the rising static friction
t2	is the dwell time, time at zero velocity
(*)	marks friction model parameters, other variables are state variables

Table 8.2.  Different friction phenomenon

	Parameter range	Parameter depends upon
FC	0.001 0.1FN	Lubricant viscosity, contact geometry and loading
Fv	0 very large	Lubricant viscosity, contact geometry and loading
Fs	0 0.1FN	Boundary lubrication, FC
kt	1/x(FC+Fs) x1 50 [m]	Material properties and surface finish
vs	0.00001 0.1 [m/s]	Contact geometry and loading
L	1 50 [ms]	Lubricant viscosity, contact geometry and loading
	0 200 [s]	Boundary lubrication

Table 8.3.  Behavior of the model

Friction model	Predicted/observed behavior
Viscous	Stability at all velocities and at velocity reversals
Coulomb	No stick-slip for PD control No hunting for PID control
Static + Coulomb + Viscous	Predicts stick-slip for certain initial conditions under PD control Predicts hunting under PID control
Stribeck	Needed to correctly predict initial conditions leading to stick-slip
Rising static friction	Needed to correctly predict interaction of velocity and stick-slip amplitude
Frictional memory	Needed to correctly predict interaction of stiffness and stick-slip amplitude
Presliding displacement	Needed to correctly predict small displacements while sticking (including velocity reversals)

Parameter identification. The parameters, which represent a non-linear relationship with the friction force, cannot be identified by using linear methods. Kim et al proposes a possible non-linear identification method for the seven parameter model, the Accelerated Evolutionary Programming (AEP) [88] and gives the description as outlined below. AEP is a stochastic optimization technique for finding the extrema of an arbitrary function. A simple optimization problem can be stated as: determine the values of the ordered set of n parameters

(8.26)

which minimize/maximize the cost function f(z). AEP is a genetic algorithm type method, working with parent/child interactions to converge to the best solution. It uses two variation operators. One of them, the direction operator determines the direction of the search according to the fitness score. The other one is a zero-mean Gaussian operator, used a perturbation added to a parent in order to generate an offspring. AEP also uses a further “age” variable to enhance the diversity of the search and to prevent individuals to remain in the local minima. The Gaussian operator is incorporated with both the direction operator and the variable age. Unlike another genetic methods, AEP generates only one child from its parent. New parents then are selected in one-to-one competition between parent and its child. An offspring is selected if the child wins in the competition with its parent. AEP incorporates the direction of evolution and the age concept into the vector to be optimized in order to increase the convergence speed without loosing the diversity. Thus it brings Equation (8.26) into the form of

(8.27)

where for allis the evolving direction of the jth parameter in the ith vector, and agei denotes the duration of the life in an integer type in the ith vector. The following two rules are applied for perturbing the parents to generate their offspring.

Rule 1

When ,

then ,

or else ,

(8.28)

Rule 2

When ,

then ,

or else ,

,

(8.29)

where denotes the jth parameter in the ith vector among Np vectors at the kth generation, denotes the evolving direction of , is a realization of a Gaussian-distributed random variable, which is polarized in the direction of , and i, i = 1, 2 are positive constants. Rule 1 indicates that if the performance of a newly generated child is better, then the search will continue towards the promising direction, by setting the direction value and generating a random number in the respective positive or negative region. Otherwise the parent will be kept, by increasing the age value. As Rule 2 “else” part shows, in the latter case the standard deviation of the Gaussian-distributed random variable is getting larger in each step, since the age is included in the product. It also makes more probable that the algorithm doesn’t stuck in the local minima but tends to find the global one.

The procedure of applying the AEP is as follows:

1.) Initialization. Generate an initial population of Np trial solutions with a uniform distribution within the given domain for the ith individual vector.

(8.30)

where és , are randomly initialized and agei is initially set to 1.

2.) Evaluation. Evaluate the fitness of each parent solution by the given cost function f(zi).

3.) Mutation. For each of the Np parents zi, generate a child by using Rules 1 and 2 successively.

4.) Evaluation. Evaluate the fitness of each new offspring.

5.) Selection by one-to-one comparison. If the newly generated child is better than its parent, then select the child and discard the parent, else vice versa. New parents are composed of Np vectors selected among 2Np vectors by one-to-one comparison. The new Np parents are sorted by fitness score.

6.) Termination check. Proceed to step 3 unless available execution time is exhausted or acceptable solution has been discovered.

The most obvious way for parameter identification is to run parallel the experiment and AEP using the same input value. Parameters are identified, when the difference between the real and calculated output values diminish or fall into the previously defined error range. Input value can be position or velocity depending on the used sensor.

For the seven parameter model, first we need to know the input v or (dx/dt) and the state variables t2 and Fs,a. For this let us define a new state variable s, which indicates, whether the system is in stick or slip.

(8.31)

where vcritical is specified by the resolution of the velocity/position sensor built into the real system. Depending on the change of the value of s, t2 and Fs,a are obtained as followed.

When s changes from

0 to 0:	t2  t2 + Ts
0 to 1:	compute Fs,b
1 to 0:	Fs,a and t2  0
1 to 1:	no operation

where Ts is the sampling time.

Let a vector z denote the set of unknown parameters of Equation (8.24).

(8.32)

where all the parameters are positive. The mismatched displacement error between the plant output and the identification output is

, i = 1, 2, …, Ns

(8.33)

where denotes the estimate of z, x(z,ti) is a displacement of the plant, x(,ti) is a displacement of the identification model at the sample time ti, and Ns is the total number of sample data. Consider the cost function of the squared errors is given by:

(8.34)

Then the parameter identification problem can be posed as an optimization problem as follows:

(8.35)

8.4.2.2. State variable friction model

Research in dynamic friction modeling of rocks in boundary lubrication carried out by geophysicist led to the born of the state variable friction model. [6], [16]. For constant normal stress, the general form including n state variables iis given by

	(8.36)
	(8.37)

Coming from this representation, a sudden change in velocity does not cause a change in  but it does affect its time derivative. For a single state variable Ruina proposes the following frictional law [16]

(8.38)

(8.39)

Where  is the scalar state variable, L is the characteristic sliding length controlling the evolution of . The pair (v0/f0) corresponds to any convenient point on the steady-state friction-velocity curve. The steady state curve is given by:

(8.40)

and the steady state variable can be related to the mean lifetime of an asperity junction.

Limitations. Unfortunately the model structure itself imposes a limitation on the use of the state variable model. At standstill, the value of v/v0 in Equation (8.40) becomes zero, which gives infinitive friction force due to the result in the logarithmic function (ln 0 = ). In simulations normally it causes a problem in the integrator. A good approach to avoid it is to include a switch, which for zero velocity gives a finite friction value. However since , it means that in the environment of very small velocities the value of the logarithmic function and thus the friction force changes very fast, resulting a strong dependence of error on the selected environment size. Hence for low velocities this model is not adequate.

Parameter identification. Dupont et al gives a detailed description on the measuring apparatus used for their experiment [49]. In principle any other methods can be used, thus we are not going into details of the original set up. The first step to identify the parameters of Equation (8.38) and Equation (8.39), one must draw the steady state velocity – friction curve (or Stribeck-curve) of the respective system. The change of velocity results a change in the system as well. This can be used to compute the values of A and B on the following way. The system’s response given to a velocity step imposed at time t0, where the pair (v1, f1) corresponds to the initial steady state, the pair (v2, f2) describes to the steady state reached after the step response, while fmax denotes the maximum friction force reached during the transient. The parameters A and B then calculated as:

	(8.41)
	(8.42)

The exponential decay of the state variable  following a velocity step is given as

(8.43)

The characteristic sliding length L can be computed from a least squares fit of the exponential to the data immediately following the velocity step.

Table 8.4 shows some state variable model parameter values calculated by Dupont et al. The used lubricants are as follows. The used lubricants are paraffin oil with maximum Saybolt viscosity of 158 and a commercial boundary lubricant paste. Test samples were made of heat-treated and oil quenched tool steel, with Rockwell hardness of 59 C.

Table 8.4.  State variable model parameter values

Parameter	Dry	Paraffin Oil	Boundary Lubricant
A 103	2.8 0.5	0.0*	1.1 0.1
B 103	9.4 1.1	11.1 2.5	4.0 0.4
L [m]	64.7 14.3	22.4 4.3	19.2 1.4

The magnitude was less then noise level of the measurement.

8.4.2.3. Karnopp friction model

In some model representations friction may be a discontinuous function of the steady-state velocity. In case of integrating discontinuous ordinary differential equations it may cause a problem, though with embedded switching functions this problem is still manageable. In addition, the integrator also has to bear with the ability to locate the point of discontinuity within the given interval. Variable-step-size variable-order methods are appropriate for this purpose and they are built in elements in all commercially available simulation software. However in the physical representation of a controller in the majority of cases we have to use fixed-step-size methods as the sampling time is fixed.

The Karnopp [84] model solves this problem by introducing a pseudo velocity and a small neighborhood of zero velocity ±DV in which the speed is set to zero. The pseudo velocity, p(t) which in the original material is referred as momentum, is integrated on the standard way.

(8.44)

where Fa is the force acting on the system, while Fm is the modeled friction force. The velocity of the system is set according to Equation (8.45), where M is the system mass.

(8.45)

The friction law is given as:

(8.46)

This includes the Coulomb, static and viscous terms (FC, Fs, Fv respectively) but can be upgraded with any other convenient curves (e.g. Stribeck curve).

8.4.2.4. LuGre model

Unlike the seven parameter model, the LuGre model is an integrated dynamic friction model. In some sources it is also referred as “integrated dynamic model”. The name comes from the abbreviation of the Lund Institute of Technology and INPG Grenoble, the two universities hosting the cooperating scientists. The LuGre model is based on the elasticity in the contact [33]. The variable z(t) represents the friction state and can be interpreted as the mean deflection of the junctions between the two sliding surfaces. The instantaneous friction Ff is given by

(8.47)

where σ0 describes the stiffness for a spring-like behavior for small displacements, and σ1 gives the negative viscous friction term. The state variable is updated according to:

(8.48)

where

(8.49)

Fc, Fs and Fv represents the Coulomb, Stribeck and static friction respectively. For some special cases, an additional element Fg can be added to the sum, representing the effect of gravity (e.g. in case of dealing with a vertical machining axis) or amplifier offsets. The steady-state friction is as follows:

(8.50)

The advantage of this model, is that besides it is valid in both rotational and linear coordinates, it describes the phenomena for normally flat surfaces as well as for rolling bearings element

Parameter identification. An identification method is presented in [2] based on the general methods used for linear dynamic systems. Note that the method includes Fg contrary to Equation (8.47)-(8.49). The parameters Fs and σ0 are easily deduced from the Dahl’s curve.

The assumption implies Fs=F. Furthermore, if , then and . Combining (8.50) with these assumptions leads to

(8.51)

Using Euler’s approximation for time derivatives, a linear regression results

(8.52)

The authors of [1] concluded that these algorithms turned to be sensitive to position measurement noise, thus they propose to use interferometers instead of inductive sensors to eliminate the bias on .

8.4.2.5. Modified Dahl model

The original Dahl model [40], [20] predicts a frictional lag between the velocity reversals and leads to hysteresis loops. It does not contain however the Stribeck effect. To overcome on this deficiency Canudas et al proposed to transform the Dahl model into a second order linear space invariant model, which emulates the Stribeck friction by producing an overshoot in the response of friction forces. However, doing so, they also eliminated the dissipative properties of the first order model. The modified Dahl model is a first order model redesigned by Canudas et al and this time combining the Stribeck phenomenon without increasing the system state dimension.

The friction force represented by Ff is described by

	(8.53)
	(8.54)
	(8.55)

where ε is a time constant proportional to 1/vsimilarly to the original Dahl model, z is a state variable should be interpreted as an “internal” friction state, vs is the characteristic velocity of the Stribeck curve, while α1, α2 and α3 are positive constants representing the Coulomb, Stribeck and viscous friction terms respectively.

The steady state friction is given by:

(8.56)

8.4.2.6. M2 model

The M2 model was developed by the same researcher group, than that of the modified Dahl model. Yet it has not got any catchy name, even in the original source it simply referred as M2. [33]. The model is based on the separate representation of stiction Fs and dynamic friction Fd. The two states are connected through a dynamic weighting factor s, which depends on the relative velocity. Thus the total friction Ff is given as

(8.57)

The weighting factor s is defined by solving the equation

(8.58)

where τs is a time constant representing the frictional lag. s = 1 means sliding whereas s = 0 implies sticking. Hence, s can be interpreted as the “measure” of sliding, having a uniform centered deterministic distribution with respect to relative displacement [33].

The so-called Dahl effect models the behavior of sticking by employing a stiff spring ks accompanied by a damper ds. Stiction however is limited by the break-away force, thus a saturation function is included in the form of

(8.59)

The saturation is described by

(8.60)

where

(8.61)

The variable  is an internal variable representing the relative displacement from the position where stiction occurred. The second term serves to reset  to zero when sliding started

(8.62)

The Stribeck effect is a result of the combination of the saturation and the weighting function, but can be traced in the steady state characteristics as well:

	(8.63)
		(8.64)

where:

(8.65)