Adaptive control methods department
Adaptive control methods department News of adaptive control methods department Scientific activity of adaptive control methods department Research areas Active projects Completed projects Scientific seminar Education Products Demo materials Publications Offers Staff Collaboration Contacts
ISPRAS Description Logo
Home   R&D Groups   Projects   Grants   Publications   Downloads

Research areas


According to department's research goal we set the following objectives:

I. Experimental study of principles of control, that are exhibited by natural nervous systems of living organisms in specific conditions and under specific circumstances not by analysis but by means of synthesis of our own models. This approach, according to our hypothesis have to result in "re-discovering" the same "inventions" that was made by Nature while synthesis of natural systems. Of course studying of biological and cybernetic sources, that have relationships to our models is also part of this objective.

II. To develop conceptual model of nervous system, that we call Autonomous Adaptive Control (AAC). This model have to describe both macro-level, that corresponds to structure and functions of the whole "organism-environment" system and micro-level, that corresponds to functions of neurons, synapses and neural impulses in the form of binary signals. Developed model shouldn?t to be solely descriptively - hypothetical but it has to be algorithmically modellable by means of a computer device. This model has also to be simulation-friendly (bionic), i. e. from all possible variants of realization of its subsystems we will select these ones which are closer to their biologic original but not these that are useful from merely pragmatic or heuristic point of view.

III. To create software models that realize properties of developed conceptual model of nervous system (AAC system) and demonstrate announced behavior.

IV. To cooperate with biology specialists in order to increase quality of developed conceptual models of nervous system.

V. To develop applied control systems based on AAC. On success we will have program for computer that will demonstrate the following important properties of nervous system:

  • ability for automatic adaptation ? step by step adapting to properties of surrounding environment, accumulating knowledge about it and building model of the world ?environment-controlled object? in its memory,
  • ability to find ways to operate with various objects of the environment,
  • ability to use self-achieved knowledge to make non-random decisions,
  • ability to elaborate reasonable stereotypes of behavior,
  • ability to improve its behavior after obtaining new knowledge about the environment and controlled object,
  • and other properties.
To enumerate these properties we need to define them in a more precise and formal way as we?ve done in our publications about AAC method.

Such model will not be presented as "Artificial intelligence" in the meaning of "artificial brain" that will be able to solve different mathematical or chess problems (such pragmatic programs were invented a long time ago). We are going to design model that has to demonstrate properties of the brain ofa new-born living mouse, for example. However, this might be huge and unachievable success, because brain of the mouse is inconceivable for the modern science.

In compliance with our goals and objectives we provide the following researches.

to top

I. Developing and research of bionic Autonomous Adaptive Control method (AAC).

The goal of this is to develop both new conceptual model of nervous system and new type of adaptive control systems based on it. Such control systems must have ability that is typical for real biological systems ? ?adaptability?, the ability to learn in time of control, multicriterial and multiparametric style of control.

to top

II. Neural networks-like realization of AAC systems based on specially developed models of neurons and networks.

The most organic realization of AAC system as a model of nervous system is network structure, that consists of neurons, synapses, receptor elements, neural impulses. But classic "Artificial Neural Networks" (ANN) approach and corresponding ?formal neurons? do not work well as models of nervous system and neuron by various reasons. Therefore we develop our own models of neurons and networks. To underline their difference form ANN we called it neuron-like elements and networks. We developed special neuron-like networks that provide functions of the main subsystems of the AAC system: Pattern Formation and Recognition subsystem (PFR), Knowledge Base (KB). Such neuron-like realization of AAC systems has been developed and are proved to work as adaptive control system for computer model of mobile robot "Gnom#8". This means, that such control system is possible to assemble from hardware neurons, in case these neurons become available. All other realizations of AAC functioning to the date are supported by more pragmatic software decisions.

to top

III. Adaptive control systems for mobile robots.

Mobile robots are illustrative example of controlled objects, that need autonomous adaptive control. Furthermore, mobile robot are always associated with living being, that stimulates interest to this goal. Achieved results in this area are controlling obvious even to unprepared audience. Besides, nowadays we can see significant growth of theory and industry of mobile robots. To see our results in area of adaptive control systems for mobile robots visit download page.

to top

IV. Application of fuzzy logic to AAC system design.

Today fuzzy logic is widely used in are of control of mobile robots. The basic idea of this approach is to apply control rules that were specified by human-expert to the control systems. For example, driver to make decision says himself: "this car is close to me rather than far away, so I have to brake". Fuzzy logic allows to formalize such control rules to include it into the control system. For more information please visit publications page.

to top

V. Application of determined chaos to AAC system design

Technology of determined chaos is the powerful tool to organize and store big repositories of text and other information as special structures, that have attractor properties and search engines that allows finding target fragments of text by small and damaged fragments. We provide researches to include this technology into AAC systems. Positive solutions were found in researches of A. Ustyuzhanin.

to top

VI. Application of genetic algorithms to AAC system design.

Applied AAC systems as well as other control systems have to be developed optimally for given conditions. For example, it?s important to select optimal parameters of sensors and actuators. Other subsystems such as pattern formation and recognition subsystem, knowledge base have to be developed optimally too. For neuron-like subsystems this means selecting appropriate structure of neuron network and setting appropriate values to neuron parameters. Unfortunately, we can?t offer methods for analytical calculation optimal parameters of AAC systems. Obviously, there are fundamental reasons why it can?t be offered for AAC systems ? lack of enough a priori information. To illustrate this problem, let?s consider the close biological problem. Is it possible to offer analytical methods to calculate optimal shape of Australian rabbit ears? It?s very difficult to offer, because optimal shape depends on vast variety of factors and conditions such as climatic conditions of Australia, character of flora and fauna, properties of behavior of other animals, length of rabbit?s hear and other. All these conditions are non-constant, they changes it?s values all the time. So, shape of ears that was optimal earlier may not be optimal in new conditions. The single well-founded method of optimization is natural selection, that acts both with genetic mechanism of transferring inherited characteristics. "Genetic algorithms" is similar method to natural selection. With our results for using genetic algorithms you can see at publications L.V.Zemskih.

to top

VII. Parallel computing for applied AAC system development.

One of the main differences between nervous system and modern computer is the fact that nervous system consists of huge number of neurons, receptor and effectors cells, that can work simultaneously, not "step-by-step" . So, there is our natural wish to have ability to create AAC systems supported by parallel computations. In particular, we develop neuron-like realization of AAC system having in mind ability to develop it on neuron chips with high degree of parallelism. About our result you can read here.

to top

Neural networks-like realization of AAC systems based on traditional artificial neural networks (ANN).

Recently we?ve started research for abilities to develop AAC systems based on traditional artificial neural networks (ANN). But positive solution were not found yet, i. e. we can?t offer method of develop AAC systems on the based on traditional neural networks. The main problems of ANN that put obstacles in the way of creating AAC system are such properties of ANN as "catastrophical forgetting" and necessity to preliminary learning of neural networks. The second item is the fundamental problem, because one of the main principles of AAC system is learning in the control process. Our efforts in this way are caused by wish to join our neuron-like AAC technology with ANN technology that has many useful decisions, deep theory and mathematical techniques. Also we suppose that AAC method may help ANN by including new models of neurons and showing working decision announced problem of passage from "recognition paradigm" to "control paradigm".

to top


Enumerated lines of investigation are concerned with:

to top

Copyright © 2002 — 2007 ISP RAS