Annamária R. Várkonyi-Kóczy

Generalization of a sigmoid generated Fixed Point Transformation from SISO to MIMO systems

Recently a systematic method was presented for the generation of whole families of “Fixed Point Transformations” that can be used in nonlinear adaptive control of “Single Input - Single Output (SISO)” systems as alternatives of Lyapunovs “direct method”. In the present paper further development of this alternative method is considered. It consists in the generalization of the method for “Multiple Input - Multiple Output (MIMO)” systems. The applicability of the novel method is demonstrated by the adaptive control of a 2 “Degree of Freedom (DoF)” system, a cart indirectly driven in the horizontal direction by a rotated pendulum. Results of numerical simulations illustrate and substantiate the usability of the suggested approach.

Adaptive control of underactuated mechanical systems using improved "Sigmoid Generated Fixed Point Transformation" and scheduling strategy

With the aim of evading the difficulties of the Lyapunov function-based techniques in the control of nonlinear systems recently the Sigmoid Generated Fixed Point Transformation (SGFPT) has been introduced. This systematic method has been presented for the generation of whole families of Fixed Point Transformations that can be used in nonlinear adaptive control of Single Input Single Output (SISO) as well as Multiple Input Multiple Output (MIMO) systems. This paper proposes a new control strategy based on the combination of the adaptive and optimal control by applying time-sharing in the SGFPT method. The scheduling strategy supports error containment by cyclic control of the different variables. Further, this paper introduces new improvements on SGFPT technique by introducing Stretched Sigmoid Functions. The efficiency of the presented control solution has been applied in the adaptive control of an underactuated mechanical system. Simulation results validate that the proposed scheme is far promising.

Anytime sport activity risk level calculation using HOSVD based hierarchical fuzzy models

In this paper a fuzzy logic-based risk calculation model is introduced, which is used to assess the risk level of sport activity in real-time. In these kinds of systems the computational complexity is a key factor, because the sufficiently accurate results should be available in time. The aim is to find the balance between the computational complexity and the accuracy. Anytime techniques are well-suited for these types of problems, because the combination of the soft computing and anytime algorithms can cope with the dynamically changing and possible insufficient amount of resources and reaction time and it is able to adaptively work with the available information which is usually imperfect or even missing. In this study the Singular Value Decomposition (SVD)-based algorithm is used to reduce the basic fuzzy model complexity.

Integration of Machine Learning and Optimization for Robot Learning

Learning ability in Robotics is acknowledged as one of the major challenges facing artificial intelligence. Although in the numerous areas within Robotics machine learning (ML) has long identified as a core technology, recently Robot learning, in particular, has been witnessing major challenges due to the theoretical advancement at the boundary between optimization and ML. In fact the integration of ML and optimization reported to be able to dramatically increase the decision-making quality and learning ability in decision systems. Here the novel integration of ML and optimization which can be applied to the complex and dynamic contexts of Robot learning is described. Furthermore with the aid of an educational Robotics kit the proposed methodology is evaluated.

A Soft Computing-Based Hierarchical Sport Activity Risk Level Calculation Model for Supporting Home Exercises

With the spread of active styles of living, regular health monitoring and risk estimation of exercises became an essential part of everyday life. In this paper, a fuzzy logic-based hierarchical, classified risk calculation model is introduced, which can be used to assess the risk level of sport activity in real-time. The model considers the current physical status and the preliminary assessed medical conditions of the person, the activity load of the exercise, as well as the environmental conditions. Based on this information, a hierarchical fuzzy decision making system evaluates the risk level and sends warning (to the person to stop the activity) or alerting (to a medical doctor/hospital) or both if necessary. By this, serious health problems/crisis situations can be avoided. The complexity of the model is optimized by the application of the singular value decomposition-based complexity reduction algorithm. In critical situations [when the available (dynamically changing) amount of time, resources, and data become insufficient], the anytime mode of operation helps to cope with the temporal conditions and to find a tradeoff between the computational complexity of the evaluations and the accuracy of the results. The system can be operated real-time at home, thus making more comfortable and safer the active (preventive) life and rehabilitation processes of conscious people.

Data Classification Based on Fuzzy-RBF Networks

Classification has been among the most typical computational problems in the last decades. In this paper, a new filtering network is proposed for data classification that is derived from radial base function networks (RBFNs), based on a simple observation about the nature of the classic RBFNs. According to that observation, the hidden layer of the network can be viewed as a fuzzy system, which compares the input data to the data stored in each neuron, computing the similarity between them. The output layer of the RBFN is modified in order to make it more effective in certain fuzzy decision-making systems. The training of the neurons is solved by a clustering step, for which a novel clustering method is proposed. Experimental results are also presented to show the efficiency of the approach.

Sigmoid generated fixed point transformation control scheme for stabilization of Kapitza's pendulum system

In adaptive nonlinear control Lyapunovs 2nd or Direct method became a fundamental tool in control design. Recently the application of the “Sigmoid Generated Fixed Point Transformation (SGFPT)” has been introduced for replacing the Lyapunov technique. This systematic method has been presented for the generation of whole families of Fixed Point Transformations and has been extended from Single Input Single Output (SISO) to Multiple Input Multiple Output (MIMO) systems. Furthermore, the Stretched Sigmoid Functions have been introduced. In this paper a new function of this family have been investigated in order to obtain a more precise positioning of the function in the vicinity of the solution of the control task. The applicability and effectiveness of the proposed control method have been confirmed by the adaptive control of the inverted pendulum with vertical vibration of the pivot, i.e. the so-called Kapitzas pendulum. Results of numerical simulations have revealed that the proposed control design ensures performance enhancement.

A fuzzy hypermatrix-based skin color filtering method

In this paper, a classification method is proposed. The idea behind the classifier is the pre-calculation of fuzzy membership function values that are stored in fuzzy multidimensional arrays (so-called fuzzy hypermatrices) so their data can be accessed quickly run-time using the parameter values of the input data. This paper focuses on the 3 dimensional 2-class case in order to achieve rapid skin area identification based on pixel color using the images of a camera. A training algorithm is presented and the performance is illustrated by an image processing problem.

A fast fuzzy decision tree for color filtering

Fuzzy decision trees have been gaining popularity in the past two decades. They are the fuzzy extensions of crisp decision trees, introducing fuzzy logic into the nodes of the tree, thus making their generalization capabilities more robust. This paper presents a fuzzy decision tree architecture that is optimized for quick inference, in order to make the classification process as fast as possible. Furthermore, two training algorithms are presented to incrementally train fuzzy decision trees for realtime classification applications.

Combination of RFPT-based adaptive control and classical model identification

The traditional approach in the design of adaptive controllers for nonlinear dynamic systems normally applies Lyapunovs “direct” method that has the main characteristic features as follows: a) it yields satisfactory conditions for the stability, b) instead focusing on the primary design intent (e.g. the precise prescription of the trajectory tracking error relaxation) it concentrates on proving “global stability” that often is “too much” for common practical applications, c) in the identification of the model parameters of the controlled system it provides a tuning algorithm that contains certain components of the Lyapunov functions therefore it works with a large number of arbitrary adaptive control parameters; d) the parameter identification process in certain cases is vulnerable if unknown external perturbations can disturb the system under control. In order to replace this technique by a simpler approach concentrating on the primary design intent the “Robust Fixed Point Transformation (RFPT)”-based technique was suggested that - at the cost of sacrificing the need for global stability - applied iteratively deformed control signal sequences that on the basis of Banach Fixed Point Theorem converged to the appropriate control signal only within a bounded basin of attraction. This method was found to be applicable for a wide class of systems to be controlled, it was robust against the unknown external disturbances, used only three adaptive control parameters and later was completed by fine tuning of only one of these control parameters to keep the system in the region of convergence. In the present paper theoretical and simulations based considerations are presented revealing that the two methods can be combined in the control of certain physical systems.