Intelligent Systems

Cellular automata are discrete, rule-based systems that exhibit complex behavior emerging from simple local interactions. This paper explores the foundational principles of cellular automata and their applications in unconventional computing paradigms, including the modeling of natural phenomena, emergent computation, and robotics. We highlight how cellular automata systems provide a versatile framework for solving computational problems beyond traditional architectures, leveraging their inherent parallelism, adaptability, and scalability. Furthermore, we examine their potential in robotics for decentralized control and swarm behavior, emphasizing their role in advancing unconventional computation.

The paper proposes design of an automatic motion stabilization system for a given control object along a desired trajectory of its motion in space. The stabilization system is constructed by means of symbolic regression. In order to obtain control parameters, a reference model is built. In the reference model, the speed of motion is determined by geometrical parameters of the desired trajectory. The considered problem is formulated in this paper as an extended optimal control problem where the cost function is the integrated accuracy of the motion along the desired trajectory. In order to implement the control function for a real control object, the control synthesis problem is solved by applying the variational genetic algorithm. An example of stabilization system designed by proposed method is given for a quadcopter moving along a given trajectory.

This paper presents an innovative application of Transformer-XL for long sequence tasks in robotic learning from demonstrations (LfD). The proposed framework effectively integrates multi-modal sensor inputs, including RGB-D images, LiDAR, and tactile sensors, to construct a comprehensive feature vector. By leveraging the advanced capabilities of Transformer-XL, particularly its attention mechanism and position encoding, our approach can handle the inherent complexities and long-term dependencies of multi-modal sensory data. The results of an extensive empirical evaluation demonstrate significant improvements in task success rates, accuracy, and computational efficiency compared to conventional methods such as Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs). The findings indicate that the Transformer-XL-based framework not only enhances the robot’s perception and decision-making abilities but also provides a robust foundation for future advancements in robotic learning from demonstrations.

Taking into account that people differ from each other not only externally, but also internally, have their individual psychological and physiological features, it is impossible to build a single model of a human being that fits perfectly to any person. And so it is impossible to build a universal intellectual assistant of a person. Building a personalized human assistant requires a close connection between a person and their intellectual assistant, which is built through iterative feedback. Another significant factor that needs to be addressed is the continuous change in human behavior and characteristics throughout life. As such, approaches are needed to describe the dynamics of human and intelligent assistant change. People’s behavior at any given moment is influenced by their internal state and environment. The intellectual assistant should take this into account, adapt, and most importantly develop itself and contribute to the development of the person, both individually and humanity as a whole. This is possible only in the presence of interaction between a person and an intellectual system and reflexion functions in all participants of development. These are the necessary conditions that must be satisfied by the feedback module in the cognitive architecture of the intellectual assistant designed on the basis of the concept of co-evolutionary hybrid intelligence. In this paper, the importance of personalizing feedback to establish a better contact between the human and the AI system is shown based on a case study of a system for human state detection. The paper demonstrates the proposed approach on the example of a system for estimating a person’s emotional state based on analyzing data about his activity of using a computer keyboard and environmental data. The results show an average of 4.12% improvement in the quality of human state prediction in systems with extended feedback.

At the moment, one of the promising areas of development is the technology of connected vehicles (V2X), which allows vehicles to receive information about other vehicles, thereby increasing the safety and efficiency of traffic.

Modern models of traffic flows based on the presence of a probabilistic component are divided into deterministic and stochastic. However, when describing the behavior of vehicles using only a stochastic or only a deterministic approach, problems with the correctness of the model may arise. In 2003, a research team consisting of A.P. Buslaev, V.M. Prikhodko, A.G. Tatashev and M.V. Yashina developed a deterministic-stochastic model of traffic flow and described the rules for it. The key feature of this model is that it takes into account the individual maneuver of the vehicle. This model allows you to create different types of car behavior on the road depending on the type of driver.

Many cities around the world have already implemented intelligent transport systems (ITS) to combat the problem of traffic congestion. These systems allow traffic control in real time and adjust vehicle routes to select the most optimal solution. Thanks to ITS, it was possible to significantly reduce traffic congestion in cities.

V2X (Vehicle-to-Everything) and Intelligent Transport Systems (ITS) are closely related to each other, as both technologies are aimed at improving the safety, efficiency, and sustainability of transport systems.

V2X is a communication technology that allows cars to interact with other vehicles (V2V), infrastructure (V2I), pedestrians (V2P), and even the network (V2N). The main goals of V2X include improving road safety, optimizing traffic, and reducing accidents by exchanging information in real time.

ITS is a set of systems and technologies that use data and communications to improve transport management, reduce congestion, manage parking, and improve overall safety. ITS includes traffic management systems, road condition monitoring, emergency services, and other aspects related to the efficiency of transport infrastructure.

This study considered a one-way intersection with one lane. The intersection is described by Buslaev’s contour network. The software package allows choosing one of three modeling options - using S. Wolfram’s cellular automaton, a stochastic model, a deterministic-stochastic model, and a deterministic-stochastic model connected to an intelligent road. The rules for each of the models are described and a comparative analysis of the main characteristics of the traffic flow is carried out.

Designing an intelligent system for monitoring environmental metrics requires the system to be integrating different technologies and applications. One of the aspects includes the development of software and hardware solutions for visualization and analysis of data from environmental monitoring systems. In additional, monitoring schemes for biodiversity based on input data underlines the importance of developing indicators based on scientific data and perspectives of the approach to achieve effectiveness and performance of monitoring programs. Also, realization of adaptive semantic models in intelligent information systems responds to the difficulties connected with the management and application of geochemical and ecological data for effective systems of observation and tracking. By Combining these approaches, a integrated and intelligent environmental monitoring system that expands the monitoring opportunities is developed. The research paper included the metrics and parameters that affect the biogeocenosis, determined the hardware to design the monitoring system, and offered a software realization of the system.

The reliability of non-destructive testing methods at key stages of production plays a crucial role in ensuring the structural safety of the product and minimising production costs. Early detection of defects allows them to be rectified at minimal cost, helping to strengthen the business and increase economic sustainability. It also makes it possible to extend the life of components and reuse them.

During operation, minor defects are usually caused by external factors such as human factors or environmental conditions. To improve the quality of product inspection, a new method is proposed involving a Decision Support System in the context of Second Opinion using Artificial Neural Networks. A functional diagram of the DSS architecture is developed, describing and justifying the choice of technologies to be implemented. First experiments were carried out to train the ANN YOLOv10 on a small consistent dataset and to test it on real images. The model demonstrates a sufficient ability to accurately predict the bounding box and class of low contrast small defects in over 50% of cases. Conversely, the accuracy of prediction for those categories with strong visual attributes rises to over 90%. This observation not only underscores the accuracy of the model, but also highlights the well-known problem of unbalanced data and local feature extraction. Such improvements promise to significantly increase the overall reliability of the results obtained.

Further research will focus on collecting a specialised defect detection dataset and improving the model architecture.

The possibilities of increasing confidence in the conclusions and results of AI systems through the use of interpretations and explanations formed on the basis of cause-and-effect relationships are discussed. However, the task of determining the causes by the observed effects does not always have a correct solution. Some examples of semantic defects of the causes based on the need to represent the semantics of objects and phenomena analyzed by the computer AI system by syntactic means are analyzed. Using the example of identifying fraudulent loan issuance schemes, some important features of mathematical tools for identifying cause-and-effect relationships hidden in empirical data are demonstrated. A mathematical model for monitoring the correctness of performed actions (i.e. prohibition and absence of fraudulent operations) is proposed and it is shown that in such monitoring false alarms in general cannot be reduced to zero.

The paper presents an optimization approach to the problem of choosing channels and parameters in the data transmission process within a Software-Defined Wide Area Network (SD-WAN). The approach presumes designing an intelligent subsystem integrated in transport agents (such as QUIC) and aimed at satisfying the given Service Level Agreement (SLA) by choosing the appropriate congestion control algorithm and Forward Error Correction (FEC) scheme. The SLA represents the constraints on four indices related to each SD-WAN channel: bandwidth, delay, jitter, and loss rate. The optimality criterion is to minimize the total average cost of successful transmission per one payload packet subject to the SLA, taking into account the quality of service provided by each channel. A valuable part of the present paper is devoted to a survey on mathematical models of data transmission networks and transport communication channels. Fundamental results and possible shortcomings are highlighted for every specific approach based on deterministic dynamic models, network calculus, and fluid approximation. The proposed optimization problem has the form of a data-driven stochastic program. Its details are provided by evaluating parameters of FEC coding schemes and their performance indices.

Most existing e-learning systems have tools for creating an individual learning trajectory. These tools select the next task or test using student performance and task difficulty levels as initial data. This article attempts to complement such technologies. Specifically, the goal is to learn how to generate sets of computational questions for exam papers. Unlike individual tests, less formalized concepts should be used to characterize such sets. The exam paper is expected to be balanced in complexity and diversified in content. No models exist for such concepts, but appropriate data sets enable their creation. We used a set of exam papers from the university course on complex analysis prepared for us by experts. They composed exam papers utilizing a set of computational questions labeled with a topic, difficulty level, and set of competencies. The new methodology should leverage expert knowledge embedded in the exam papers. We propose and compare two generation methods. The first one is based on the classical probabilistic model and uses only frequency characteristics. The second one is based on generative-adversarial neural networks. The experimental results show that generative models are applicable for exam papers composing even in the presence of issues of poor convergence and partial mode collapse.

New wavelet neural networks (WNN) method for linear stochastic system (StS) synthesis based on canonical expansions (CE) and mean square error (MSE) criterion is developed. Architecture of three layer WNN with one latent layer is presented. Activation functions of latent layer are based on chosen wavelet orthonormal basis with general compact carrier. Training WNN algorithm for inverse error prevalence by method of steepest descent is used. MSE optimal operator is constructed. Special formula for MSE optimal estimation of StS output in the form of linear combination of basis wavelet functions is obtained. Numerical example illustrates CE WNN preference with wavelet CE.

Data compression is a process of transforming data in order to reduce its volume. Two important parameters of any compression algorithm are the compression ratio and speed. The compression ratio is crucial when it comes to storing or transmitting large amounts of original data, while speed becomes significant when small amounts of data are constantly exchanged (such as in message transmission). The goal of this project is to increase the speed of data compression in order to optimize memory usage in electronic devices. The work includes an analysis of existing archiving methods and the possibilities of using binary trees to ensure that the code meets the Fano condition. A coding method based on reducing the depth of leaves in the codeword tree was developed, as well as another method that optimizes the leaf depth during coding by solving the 0-1 Knapsack Problem. Theoretical complexity estimates were obtained for all algorithms and comparisons were made with the classical Huffman algorithm. Practical implementations were also provided. The tests showed that the developed algorithms are superior in terms of time and allow for effective use on small data volumes, despite the average length of the codeword increasing slightly.

Primary factors influencing IT project outcomes are personnel and their professional development. Personnel assessment is a key tool for determining employee potential and possible career paths. In most cases the assessment is based on the subjective opinions of managers, which do not possess a high degree of reliability and completeness. The solution to this problem is seen in using the results of digital footprint analysis. In this study, a model for assessing software developer competence based on fuzzy logic was constructed, utilizing the outcomes of an employee’s digital footprint data mining analysis for calculations. The model was developed using a proposed competency matrix, which incorporates significant factors identified from the structural model as competencies and additional factors derived through text mining.

At present, various models of neurons oriented at implementation in the digital element basis are known, but these models have significant differences from real biological neurons, which operate not with code combinations in binary representation, but with impulse streams. The author offers a variant of realization of the neural cell models; the “biological nature” of these models is manifested in the use of bit streams to represent information. Computational processing of streams occurs in a continuous tracking mode without conversion into traditional binary codes. In the article the digital element base for design of bit-stream modules represented by logic primitives and digital modules is shown; the generalized variant of one-layer bit-stream perceptron is considered; the possibility of implementation of the proposed technical solutions on the basis of FPGA is illustrated. The developed modules are written in the SystemVerilog language, it allows to implement them in FPGAs of different manufacturers and use them as a basis for further design of ASIC implementations.

The aircraft manufacturing industry is distinguished by its large-scale operations, high task volume, repetitive processes, and strict precision standards. To meet these demands, the integration of robotic systems for automated production is crucial. Given the customized requirements and small batch production typical in aircraft manufacturing, human-robot collaboration has emerged as a viable solution for drilling and riveting operations on aircraft fuselages. However, a critical challenge in these collaborative systems is the efficient allocation of tasks between humans and robots, ensuring that the overall system operates in a coordinated and optimized manner. This paper models the scenario of drilling tasks on aircraft fuselages using three methods. The first method is the Mixed Integer Linear Programming (MILP) approach, which has been thoroughly explored in the literature, with the goal of minimizing the overall completion time. The second method involves an extended version of the Hungarian algorithm, where each row’s coefficients are duplicated, and virtual tasks are added to convert the cost matrix into a square form. The third method is based on the Consensus-Based Bundle Algorithm (CBBA), which designs a bidding score function that better suits the characteristics and capabilities of heterogeneous agents, such as humans and robots. Simulations were carried out in Python, focusing on the drilling areas of the stringers and main frame. The results showed that the CBBA-based algorithm exhibited the lowest time complexity, while the MILP approach demonstrated the shortest computation time. Given the assigned task set, a greedy algorithm can be chosen for solving task execution planning for humans and robots.

The paper presents the results of a research of the possibility of assessing a person’s well-being using gait parameters. Gait parameters were assessed using smartphone accelerometer data collected during the research from 2023. The work also describes the features and methodology for collecting and preprocessing data. Acsa Active and Data Analyzer proprietary software was used to collect and process data. In addition to recording the gait, the person entered data about his well-being. This paper describes the results of predicting well-being based on gait parameters using the example of headache and lower back pain. The simulation was performed in the TensorFlow package. The comparative analysis included: feedforward neural networks (with the number of layers from 2 to 5), vector clustering, K-nearest neighbors, random forest, gradient boosting, ada boosting, K-nearest neighbor-based bagging classifier and gradient-based competitive network boosting, random forest and linear regression. According to the results of comparison of the described forecasting algorithms, the best results were shown by the competitive network (F-score is 0.75), bagging (F-score is 0.74), multilayer perceptron (F-score is 0.71). The results of the research indicate the possibility of forming predictive estimates of changes in a person’s well-being based on data on his gait collected using a smartphone. The work also notes that to understand all the capabilities of the technology for assessing health by gait using a smartphone, it is necessary to expand the sample and attract additional data.

The work is devoted to countering cyber attacks on electrical power systems. For this purpose, a method is proposed for predicting system states based on the previous dynamics of its functioning. The method is two-stage and consists of the following steps: 1) the stage of building a functioning model - collecting information about the states of the system, forming states, clustering states (to combine close states into one), building a recurrent artificial neural network model, training the model; 2) the stage of modeling the behavior of the system – obtaining the current state, predicting future states, checking states for attacks, assessing current security. The method is implemented in the form of a software prototype (using the architecture of an artificial neural network – LSTM), with the help of which a number of experiments were carried out on the HAI (HIL-based Augmented ICS) dataset. As a result of the experiments, the accuracy of the trained model was assessed and the probability of correctly predicting several future states of the electric power system was determined depending on various parameters (history length, number of artificial neural network nodes, etc.). The main controversial research issues are highlighted and their explanations are given.

in this paper, methods for evaluating the quality of a new dataset using statistical analysis and classical machine learning are presented. The proposed machine learning approach is based on comparing a new dataset with a known reference sample of high quality. This process is carried out using a supervised machine learning algorithm and solving a classification problem. As a result fitted algorithm, it is necessary to estimate its quality of the ROC-AUC\(^{1}\)(Receiver Operating Characteristic - Area Under Curve) metric, on the basis of which a conclusion is made about the quality of the new sample. As an additional measure of comparison, the PSI\(^{2}\)(Population Stability Index) is taken into consideration as a statistical indicator that provides an approximate qualitative evaluation of the difference for 2 distribution densities.

The article considers one of the key problems of microservice systems associated with the rapid growth of threats to container applications, containerization and orchestration tools. The emergence of new attack vectors, as well as continuous improvement of the existing ones, creates serious difficulties in the implementation of such systems and technologies. This article presents an approach to the implementation of a software component for detecting anomalous packets in the network traffic of container systems. The approach is based on the frequency analysis of the payload of network traffic packets, constructing histograms of fixed-size network packets and using them both for training the Autoencoder (AE) – Long short-term memory (LSTM) hybrid neural network model and for subsequent detection. The experimental results of the proposed approach show a fast training process and high accuracy in detecting anomalous packets in network traffic. Additionally, the low rate of false positives makes the proposed solution suitable as an extra layer for intrusion detection in container systems.

This paper describes a control algorithm that uses an adaptive neural-fuzzy network to control the depth of an elongated cylindrical body class of autonomous underwater vehicles (AUVs) operating in a marine environment with many uncertainties. After successfully synthesizing the control algorithm, a Particle Swam Optimization (PSO) algorithm is added to determine the controller parameters. The controller parameters found based on the PSO algorithm have made the system's stability more sustainable. Finally, the simulation results of the depth control of the AUV in the vertical plane are provided to verify the effectiveness of the adaptive neural fuzzy network controller (ANFNC) algorithm combined with PSO.

This paper investigates the impact of algorithms for trajectory interpolation in collaborative robot painting processes on energy efficiency and operational execution time. The study aims to develop an algorithm for trajectory planning that works with complex geometric details and identifies a suitable trajectory interpolation method for synthesizing the painting control program.

MATLAB is used for software implementation. Analytical geometry methods and affine transformations are employed to generate the spray gun trajectory based on child geometry and technological requirements. Various interpolation techniques (spline, makima, and pchip) are examined for their effectiveness in creating smooth trajectories in joint space, minimizing deviations from the intended path.

The results demonstrate significant variations in energy consumption and trajectory deviations between the different interpolation methods. CubicSmooth/spline demonstrated the best performance in minimizing energy consumption and deviation from the planned trajectory.

The trajectory calculated by the algorithm proved its operability and adherence to technological requirements. Our findings emphasize the significance of selecting appropriate interpolation algorithms based on the specific requirements of the technological process in order to optimize robot performance and energy efficiency. Future research will focus on extending the range of robotic applications and trajectory shapes while considering safety measures for collision avoidance.

This research focuses on aluminum alloys, including wrought ones. The authors identify the factors affecting their properties like alloying components and external pressure applied to the crystallizing metal. Pressure is selected depending on the level and nature of changes in this parameter over time. This helps significantly affect the state of the alloy and the formation conditions of its properties. The authors used hardness as the output parameter.

Deductor software was used to train the neural network model. Learning also relied on the Resilient Propagation (Rprop) algorithm. The authors developed a neural network model comprising an input layer of eleven neurons, an output layer of one neuron, and a hidden layer of seven neurons.

The developed ANN meets the conditions and can be used for the forecasting of aluminum alloy hardness depending on the percentage content of chemical elements and applied pressure.

The provided solution can be used as a decision support system (DSS) at companies that use pressure molding and liquid forging.

Swarm robotics is a complex domain within multi-robot systems that encompasses formation control, movement control, and inter-UAV communication. Coordinates task execution requires effective swarm control, which relies on robust motion control algorithms and reliable data exchange mechanisms. In this work, we propose an operator-leader-followers approach for managing UAV swarms. It comprises the following components: centralized swarm control utilizing the Fixed Global Difference (FGD) algorithm for maintaining inter-UAV distances and facilitating communication; a follow-mode mechanism enabling followers to accurately track leader’s motion; and a point-by-point swarm flight path control system enhancing overall control accuracy. Implemented in the Robot Operating System (ROS) and validated in the Gazebo simulator, the algorithm’s performance is evaluated via the following simulation scenarios: \(\varGamma \)-shape, parabolic, and circular flight paths. The paper discusses experimental results in virtual environments of the Gazebo simulator and demonstrates achieved precision in terms of absolute error and root mean square error (RMSE).