Information technology for building behavioral models of multiagent robotechnical systems

Размещено на http://www.allbest.ru/

Department of Applied Mathematics and Informatics, State Institution "South Ukrainian National Pedagogical University named after K. D. Ushynsky",

Information technology for building behavioral models of multiagent robotechnical systems

Mazurok Tetiana Leonidivna D.E.Sc., Professor, Korablov Viacheslav Anatoliovych Senior Teacher

Chernykh Volodymyr Volodymyrovych PhD in Pedagogy, Senior

Odesa

Abstract

This work has the goal to derive a set of tasks and systematize methodologies suitable for solving those tasks in the course of developing a united theoretical basis for modeling multi-agent robotic systems (MARS). It features the character of a reflection experiment, the subject of which is gradually becoming the need for such systems, the required technological solutions and the optimal way of implementing such systems using the example of a chosen task. Thus, we shall examine the existing problems that can be solved through the implementation of these systems, their current methods of solutions and the alternative that will appear due to MARS. We shall also describe both the existing theoretical developments within the discipline, optimal for introducing into the future production process, and the recommended areas for research with indications of specific points by either using inclinometric techniques, or introducing methods of cognitive intelligence, and with references to specific works specialists who have a good understanding of these issues. We shall present the data of our own theoretical developments taking into account international experience and the synthesis of related disciplines, which are the derivation of the doctrine for standardizing MARS, when used in real space conditions, for tasks on resolving crises related to the life and limb risks. Both the recommendations for the hardware and the mathematical, and the software parts of the approach to the implementation of such systems will be provided, with indicating the strengths and weaknesses of the chosen methods, such as emphasis on a centralized agent management system, communication protocols, required sensor equipment, etc. Based on the results of this observation, we shall develop the direction for further activities on the standardization of the MARS models.

Keywords: multi-agent system, multi-agent robotic system, cognitive intelligence, centralized approach, behavioral model.

Анотація

Мазурок Тетяна Леонідівна доктор технічних наук, професор, завідувач кафедри прикладної математики та інформатики, Державний заклад «Південноукраїнський національний педагогічний університет імені К. Д. Ушинського», м. Одеса

Корабльов В'ячеслав Анатолійович старший викладач кафедри прикладної математики та інформатики, Державний заклад «Південноукраїнський національний педагогічний університет імені К. Д. Ушинського», м. Одеса,

Черних Володимир Володимирович кандидат педагогічних наук, старший викладач кафедри прикладної математики та інформатики, Державний заклад «Південноукраїнський національний педагогічний університет імені К. Д. Ушинського», м. Одеса

ІНФОРМАЦІЙНА ТЕХНОЛОГІЯ ПОБУДОВИ ПОВЕДІНКОВИХ МОДЕЛЕЙ МУЛЬТИАГЕНТНИХ РОБОТОТЕХНІЧНИХ СИСТЕМ

Дана стаття має на меті вивести комплекс завдань і систематизувати методології, придатні для вирішення цих завдань в ході розробки єдиної теоретичної бази для моделювання мультиагентних робототехнічних систем (MARS). Вона має характер рефлексивного експерименту, предметом якого поступово стає потреба в таких системах, необхідні технологічні рішення та оптимальний шлях реалізації таких систем на прикладі обраної задачі. Таким чином, ми розглянемо існуючі проблеми, які можуть бути вирішені завдяки впровадженню цих систем, існуючі методи їх вирішення та альтернативу, яка з'явиться завдяки MARS. Ми також опишемо як існуючі теоретичні напрацювання в рамках дисципліни, оптимальні для впровадження в майбутній виробничий процес, так і рекомендовані напрямки досліджень із зазначенням конкретних точок або з використанням інклінометричних методик, або з впровадженням методів когнітивного інтелекту, а також з посиланнями на конкретні роботи фахівців, які добре розуміються на цих питаннях. Наведемо дані власних теоретичних розробок з урахуванням міжнародного досвіду та синтезу суміжних дисциплін, які є похідними від доктрини стандартизації MARS, при використанні в реальних космічних умовах, для задач з вирішення кризових ситуацій, пов'язаних з ризиком для життя та здоров'я людей. Будуть надані рекомендації щодо апаратної, математичної та програмної частин підходу до реалізації таких систем із зазначенням сильних та слабких сторін обраних методів, таких як акцент на централізованій системі управління агентами, протоколах зв'язку, необхідному сенсорному обладнанні тощо. На основі результатів цього спостереження буде розроблено напрямок подальшої діяльності зі стандартизації моделей MARS.

Ключові слова: мультиагентна система, мультиагентна робото- технічна система, когнітивний інтелект, централізований підхід, поведінкова модель.

Introduction

The term “agent” is found almost universally when it comes to scientific or applied disciplines. Just as the methods of these kinds of disciplines are different, the meaning of the term “agent” also differs in them, except for the following points: the ability to react to environmental factors, the possibility of influence that effects on the environment, the possibility of interaction with other agents. This basis has also migrated to the agent- oriented approach to programming, where it was supplemented by the specifics determined by this discipline. Applying an agent-based approach is the next logical step in the development of robotics, explained by the continuity of this discipline in cybernetics and computer science. Which, after some manipulations, can be represented in the form of the Lacanian transformation of the virtual into the real, but in the applied sense of the term. A number of restrictions caused by environmental properties, both known and hidden, both conditional and unconditional, must accompany any similar transition. This leads to a narrowing of the range of methods for implementing this approach. This way, in our case, from the environment where the only limitation is time, we move into the real world with real problems. The latter fact is quite often, omitted to please theoretical speculation, but if the project is really needed, the following factors will have to be faced: the task itself, the qualifications of the operators, the threats to the environment, the physical laws, orientation in an incomplete prception of the environment, critical situations, breakdowns, dimensionality of devices, etc., which for the most part rests on the current level of available technologies.

Fig. 1. Block diagram of the MARS agent

Multi-agent systems is the direction of artificial intelligence for solving a complex problem, which uses systems consisting of many interacting agents. This area of artificial intelligence is actively developing and is currently still in its infancy. In multi-agent systems, the whole range of tasks, according to certain rules, is distributed among all agents, each of which is considered a member of an organization or group. Distribution of tasks means assigning each agent a role, the complexity of which is determined based on the capabilities of the agent. It is known that a multi-agent robotic system (MARS) can be considered as one of the options for implementing a multiagent system (MAS) model, so that each robot agent has the known properties of agents in aggregate [1]. The control systems of such complex structures should ensure the adaptability of robotic devices to the range of tasks to be solved, the coordination of the development of motion paths, etc. Therefore, an urgent issue is to increase the adaptive properties of the control system of complex robotechnical structures (CRS). In order to provide for the full functioning of such systems, it is necessary to improve the information support of the control system. Thus, it is proposed the development of special information technology that integrates into the robotic complex, and is designed to perform automation tasks aimed at improving the efficiency of the operation of the CRS. Such information technology allows automating the process of constructing behavioral models of a multi-agent system, based on the use of the principles of centralization of analysis and control processes, as a component of virtual simulation (Fig. 2).

Fig. 2. Block diagram of the MARS agent management system

Analysis of the latest research and publications

To sum it up, in the information technology under development, it is pro-posed to place the central control unit (CCU) on the remote server, and conduct strategic planning inside a virtual environment that simulates the real space (strategic level). This approach is appropriate when performing tasks in an environment with high volume of conducted research, for example, if there are virtualized building plans with all the indicators (escape routes, material characteristics, weak and strong structural elements). Therefore, based on the initial global formation of the suggested environment and the feedback system with agent robots, the system allows building up dynamically a virtual space identical to a real one, and form an algorithm for solving a crisis. The CCU, owing to its high capacities, can produce an algorithm for solving a problem at a faster pace. On the other hand, if a non-standard solution is required, requiring a heuristic approach, it is possible to work out with the maximum speed the necessary number of simulations to obtain a solution close to optimal, even before a direct empirical experiment in the area [2]. Primary training of a neural network may be insufficient, which means that it is necessary to introduce adaptive mechanisms and tools of dynamic field training. This subsequently means that the latest communication technologies [3] and synchronization of real space simulation with these protocols will be required.

The aim of the research is to derive a set of tasks and systematize suitable methodologies for solving them during the development of a united theoretical basis for modeling multi-agent robotic systems (MARS).

Presentation of the main material of the scientific research

To begin with, let us determine the spectrum of problems, which require us to focus our attention. At each conditional moment, a certain set of tasks is formed whose solutions have not yet been formed, and solutions that do not lose their relevance for already known problems are regular constructs. The convenience of the latter is explained only by habit. That is to say, it is necessary to deal with uncertainty, and the same is true for tradition, although the latter is an obviously losing option, which generates the need for a compromise. As an example, let us consider several options for the possible implementation of a robotic system, but it is necessary to remember that the specifics of the multiagent approach implies the use of group robotics technology, i.e. it endows machines with a certain degree of autonomy: cognitive intelligence mathematical

A system replacing a person or a group of people in a workplace with a low degree of responsibility;

A system replacing a person/group of people in a workplace with a high degree of risk and threat to the life and limb of an individual;

A system designed to work in extreme conditions, that replaces a remote operator;

A system that mimics/interacts with a person ensuring that workplace safety with a high degree of risk/error.

The first case is insignificant due to the availability of ready-made (effective and safe) solutions and the reaction factor of the society. The second case is decided by the transition to the third, otherwise it is faced with public distrust. The third here is complicated even from an ethical point of view, since it implies the abolition of a workplace that requires the highest level of qualification and narrow preparedness. In addition, the third case implies the presence of crises, the solutions to which can only be found empirically, i.e. it becomes necessary to give the machine time to learn with actual damage, or to back it up with the same specialist. These options are extremely generalized and abstracted from many specific branches of human activity, but they are by far the most common. The fourth, however, implies a kind of synthesis between the skills that are certainly qualified by workers in industries such as: army, police, rescue services, medicine, heavy raw materials production, etc.; and capacities that a robotic device owns by definition. It is the one worth considering in the first place. For convenience, hereinafter, a similar system will be referred to as autonomous robotic toolset (ARTS). The next step is to determine the level of centralization, dictated by the specifics of this task. In practice, it is preferable to consider centralized and decentralized approaches in robotics not as closed doctrines, but as directions of a bi-directional scale, where our position depends on the weight of the criteria that we are considering. Thus, the tendency toward miniaturization, which rests on the plateau of microprocessor development, requires a more centralized approach, since it leads to a decrease in the computing power of an individual agent. The mechanical simplicity of the task contributes to the unification and methods of basic group interaction, which allows a large degree of decentralization. In our case, we also have a staff of qualified employees who can interact with complex computer systems, which allows us to take the risk and transfer a significant part of the high-level data processing to a certain central control unit (CCU). Such a solution imposes great requirements on communication systems between agents, but also enables virtual modeling of the environment and the progress of tasks, which minimizes the number of “in the field” errors. The released useful volume inside the design of the physical agent should be used to install sensors that are more effective and means of quick (reactive) response to threats, both for the robot and for people in the immediate vicinity. Therefore, in this case, the model is more prone to centralization, which makes it possible to reduce the cost of production and facilitate the repair of robots, which is useful, given the specifics of the crises in the sectors described above. We will adhere to this direction further. Thus, we propose the development of a special information technology that is integrated into a robotic complex to perform automation tasks and increase the efficiency of its functioning, by constructing behavioral MAS using the principles of centralization of analysis and control processes as components of virtual simulation. Centralized management strategies involve concentrating the entire multitude of command and control functions in a single unit, which ensures the planning and coordination of the appropriate actions of the elements of the agent group in solving common applied problems. The appropriate structure of the centralized control system should provide for two-way communication channels between the command and control body and each of the robotechnical devices. Among the main advantages of centralized management systems, it is worth to include a significant reduction in the functional load on ordinary members of the group, when many important and inherently complex issues, such as:

Analysis of the set applied task;

Collection, organising and interpretation of data on the peculiarities of the current situation, working environment and the state of the external environment;

Planning appropriate actions and monitoring their implementation by specific performers are purposeful transfer to the command and control unit.

To organize the task distribution process in multi-agent systems, either a distributed problem solving system or decentralized artificial intelligence is created. In our version, the process of decomposition of the global problem and the in-verse process of composition of the solutions found are controlled by a single center". At the same time, a multi-agent system is designed strictly from top to bottom, based on the roles defined for agents and the results of dividing the global task into sub-tasks. MARS (Figure 3) can be considered as one of the options for implementing multi-agent systems (MAS), and, therefore, each robot agent must feature the following properties:

activity, ability to organize and implement actions; reactivity, ability to perceive the state of the environment;

autonomy and relative independence from the environment; sociability de-rived from the need to solve its own problems together with other agents and is provided by developed communication protocols;

purposefulness, which provides for the availability of own sources of motivation.

Fig. 3. MARS block diagram

In the framework of this research, a variant of a multi-agent robotic system (MARS) is described, consisting of units, that is, many identical agents. Each agent is a combination of hardware and software components. The hardware component measures the parameters of the external environment with sensors, which provides the possibility of in fencing the environment, control-ling executive devices and the possibility of "communication" between agents. The software performs the tasks of: analyzing input signals from sensors, as well as building an external world model, making decisions, generating control signals and data transfer packets. Turning to the specifics, we need to choose a separate case for the implementation of such MARS. Here we will consider a system that helps the Rescue Service employees to resolve crises in high-rise buildings, where the number of floors makes re hoses and ladders ineffective, given their maximum length, as well as other difficulties arising from non-standard solutions. In such a situation, the solution to which we consider a swarm of robots, the capabilities of the Rescue Service employees are significantly expanded. United robotic units are capable of carrying equipment, and, if necessary, can be combined into rigid structures (preventing collapses, bridges, stairs, ramps and even makeshift tunnels that protect against re and debris).

All this can significantly reduce mortality, both among victims and employees of the res-cue service. When the task is formed, it is divided into subtasks (tactical level) for each group of robots. The subtasks are dynamic in nature, and they largely depend on the local space model that has developed in real time based on the sensor data of each agent, and is designed to operate on circumstances. At this stage, the distribution of tasks of navigation and manipulation of agent effectors occurs. At this level, it is necessary to increase the coefficient of autonomy. The need for reaction capabilities, in terms of constructing/restructuring the routes, or the order of interactions between agents and objects of the environment, implies the necessity for the use of modern solutions of artificial intelligence [4]. This necessity grows with increasing complexity of the terrain and the number of tasks that the system operator sets.

The critical situations for which it is being prepared carry enormous demands on the multitasking load of agent groups by default, which can cause con icts in the protocols of their own communication channels separate from the direct control signal. This will require field-testing and development of the current doctrines of multi-agent communications [5]. Moreover, the first versions of such systems must be created with extensive consulting support from professionals in the industries in which they will be involved. This is an absolute necessity, but it implements the human factor (or rather, a narrow area of personal experience of a particular specialist) in the preparation of machines that are expected to manifest universality in the approach to completing tasks. In addition, to implement such a module, it is necessary to have integrated software, both at the level of a command center and an individual agent, which would perform soft calculations, and a set of nonstandard sensors for such systems containing inclinometric devices with high accuracy and data collection speed. The question of organizing the reaction behavior of an individual agent (reactive level) remains open. It is assumed that the agent's optimal correct execution of the task is always under threat, both external (mechanical obstacles, immediate threats, etc.) and internal (error in navigation, damage to the unit itself, etc.).

These factors necessitate a certain level of agent autonomy, which requires the implementation of a lesser cognitive system (LCS). Here, the LCS will be responsible for developing all possible options for actions that should be formed in case of danger to an agent, operator or an outsider, if there is no corresponding instruction from a higher level of the command hierarchy. In addition, during non-crisis times, this system will be responsible for reactive maneuvering of the agent, which is necessary to adjust its position in space in accordance with the trajectory of motion corresponding to the task from a higher level.

Conclusions

Summarizing the above, it can be stated that, given the current technological level (since the element of "progress" in science should be neglected because it is necessary to consider the facts that be, not those that should be), the MARS models aimed at solving specialized tasks for group robotics are still striving for centralization. This is stipulated by the current level of neural network technology (which was deconstructed for the sake of simplifying the mathematical model), which, for operating in a real environment, should be supported by the tight control systems based on a huge array of logic gates and preferably of control by an operator.

It is stipulated by the availability in the mass market of equipment that is necessary under the stress tolerance criteria. It is also stipulated by the unconditionally of the presence of the human factor in the multitude of decisions formed. Such a model suggests a multi-level structure (strategic, tactical and reactive), where the complexity and global decision-making are reduced, but the response of the system is increased, which ultimately leads to system balancing and emergency response to threats. The latter requires the inclusion of LCS technology at the reactive level, which will provide an imitation of instinctive (determined by the properties of an individual agent) behavior. The main drawback of this block is its potential vulnerability, since failure of the CCU inevitably leads to disruption of the system as a whole.

In addition, problems can arise due to the shielding of work areas and through third-party obstacles to the passage of the signal. In such scenario, there is a need to use complex communication protocols with duplicate data transfer methods and constant verification of their up-to-date status, and to combat the collisions. The result of this study was an attempt to indoctrinate the developed model for the further development of MARS within its paradigm, which should direct the capacities of both specialists and patrons to the real path.

References

1. Wooldridge., M.: An introduction to multiagent systems. JOHN WILEY AND SONS, LTD (2002)

2. T.C., S.: Dynamic models of segregation. The Journal of Mathematical Sociology 1(2), 143{186(1971)

3. Pitt, J., Mamdani, A.: Communication protocols in multiagent systems: A development method and reference architecture

4. Russell, S., Norvig, P.: Artificial Intelligence: A Modern Approach. Upper Saddle River, NJ, USA : Prentice Hall Press (2009)

5. Payne, T.R.: Communicating agents in open multi agent systems. Proceedings of 1st GSFC/JPL Workshop on Radical Agent Concepts (WRAC) pp. 365{371 (2002)

6. Xiao, L., Greer, D.: Modeling, auto-generation and adaptation of multiagent systems. Proceedings of the Tenth CAiSE/IFIP8.1 International Workshop on Exploring Modeling Methods in Systems Analysis and Design (EMMSAD'05) pp. 605{616 (2005)

Література

1. Wooldridge., M.: An introduction to multiagent systems. JOHN WILEY AND SONS, LTD (2002)

2. T.C., S.: Dynamic models of segregation. The Journal of Mathematical Sociology 1(2), 143{186 (1971)

3. Pitt, J., Mamdani, A.: Communication protocols in multiagent systems: A develop-ment method and reference architecture

4. Russell, S., Norvig, P.: Artificial Intelligence: A Modern Approach. Upper Saddle River, NJ, USA : Prentice Hall Press (2009)

5. Payne, T.R.: Communicating agents in open multi agent systems. Proceedings of 1st GSFC/JPL Workshop on Radical Agent Concepts (WRAC) pp. 365{371 (2002)

6. Xiao, L., Greer, D.: Modeling, auto-generation and adaptation of multiagent systems. Proceedings of the Tenth CAiSE/IFIP8.1 International Workshop on Exploring Modeling Methods in Systems Analysis and Design (EMMSAD'05) pp. 605{616 (2005)

Размещено на Allbest.ru