The paper proposes algorithms for adaptive control of pure-feedback objects based on the strategy gradient method. A reinforcement learning framework known as Actor-Critic is used. The Critic algorithm proposes analytical method for calculating the assessment of the value of the state. The Actor algorithm is based on a given structure of the control law and the time difference method. The novelty of this algorithm lies in the modification of the learning algorithm, which limits the growth of the adjustable coefficient. An analysis of the stability and convergence of the processes of stabilization of the controlled variable and estimates of the value of the condition is given. Conditions have been obtained under which the controlled variable does not exceed the specified constraints during the learning process and is asymptotically stable relative to the zero state. Two approaches to the consideration of constraints on the control are proposed. The first approach uses a neural network approximation of the control algorithm and suggests zeroing the delta error when the control reaches constraints. In the second approach, control is designed in the form of an analytical function with constraints. The conditions for the stability of the system relative to the zero position are obtained, taking into account the constraints on the control action. The obtained algorithms are generalized to a multidimensional object presented in canonical form. The algorithms ensure that state variables and control actions are found in a given area during the learning process, which allows them to be applied without prior training.

This first part of the article discusses the design of an output feedback controller for a multivariable control system. The controller is designed to provide specified or achievable performance in terms of: control errors, stability margins, and response time. The plant is subject to bounded external disturbances. The design problem is solved using a modified H∞-optimization procedure. The second part of the article provides an interpretation of the stability margin radii for a multivariable system, formulated in terms of Nyquist plots with breaking points at individual plant inputs. A direct relationship is established between the absolute stability of a closed-loop multivariable system with sector nonlinearities at the plant’s input and its stability margin radii. The proposed approach is illustrated by an example of controller design for a load-interconnected electric drive, demonstrating its relevance to engineering practice. This paper addresses the synthesis of output controllers for linear multivariable systems, using performance indicators that are widely adopted in engineering practice. These will be referred to as engineering quality indicators. These indicators are used in practice to assess the effectiveness of a designed closed loop control system. It is essential to note that the engineering quality indicators admit experimental verification. These indicators characterize the system’s accuracy (assessed by control errors under bounded disturbances), its response time (determined by the settling time), and its stability margins (evaluated via Nyquist plots of the open-loop system for each measured and control variable, i.e., at the plant’s physical output and input). In the classical automatic control theory for scalar systems, stability margins are typically quantified by gain and phase margins. However, in some cases, these indices can be misleading; the Nyquist plot of the open-loop system may pass very close to the critical point (–1, j0) without a corresponding significant reduction in their values. Therefore, this article employs the stability margin radius defined as the minimum distance from the critical point (–1, j0) to the Nyquist plot to assess robustness. This radius allows for the direct specification of guaranteed lower bounds on the classical gain and phase margins. This first part of the work considers multivariable systems subjected to unmeasured, bounded external disturbances. The disturbances are assumed to be continuous and piecewise differentiable, a class that covers most scenarios encountered in engineering practice. The objective is to synthesize an output controller that guarantees specified or achievable engineering quality indices: control errors for each output, settling time, and stability margins evaluated at the plant’s physical input. Moreover, a dedicated stability margin radius is guaranteed for each individual control input. The synthesis problem is solved via a modified standard H∞ optimization procedure. The order of the synthesized controller does not exceed that of the plant. The operating accuracy requirements are satisfied by selecting a diagonal weight matrix for the controlled variables in the optimization criterion. The matrix elements are determined via rigorous formulas that utilize the known amplitudes of disturbances and the specified control errors. The second part of this work analyzes the frequency-domain properties of the synthesized system using Nyquist plots of the loop transfer function, broken at each individual plant input. The absolute stability of the closed-loop system with sector nonlinearities at the plant input is proven. The size of the sector is directly determined by the achieved stability margin radius for that input. An example illustrating the method’s effectiveness is provided, based on the synthesis of a multivariable output controller for a practical electromechanical system.

The paper is devoted to solving the problem of developing a synthesis method for combined position-force control systems (CS) for electric drives (ED) of multi-link underwater manipulators (MUM) mounted on autonomous underwater vehicles (AUV) operating in the mode of landing on the ground or on work sites, followed by rigid fixation of these AUV using special devices. To solve this problem, a comprehensive method is proposed in the first part of the paper. First, according to this method, a self-adjusting regulators are synthesized for the ED of each degree of freedom of the MUM, which makes it possible to stabilize its variable dynamic parameters at a given nominal level. This stabilization compensates for the influence of Coulomb and viscous friction, as well as part of the external torque caused by the mutual influences between the MUM links and their interaction with the viscous medium, on the quality of control. Then, to ensure the operation of the self-adjusting regulators and the ED CS, based on the initial nonlinear models of the ED, observers with a variable structure are synthesized to obtain information about the current values of external moments, as well as the rotation speeds and accelerations of all the ED output shafts. Using this information and measurements of only angle sensors of the specified output shafts, it was possible to synthesize position-force regulators for ED, minimizing the selected quadratic cost function, ensuring the specified movements of the output shafts of the item instance while creating the required moments. In the second part of the paper a synthesis of a combined position-force control system with all MUM elements was performed using these position-force regulators. As the results of computer modeling showed, this control system is capable of providing accurate movements of the MUM work tool along specified spatial trajectories and simultaneous creation of the required force effects on the work objects in the presence of the interaction of all its links with a viscous medium with variable parameters (viscous friction and added masses) while performing various complex technological operations.

A critical validation of the theoretical predictions for barometric parameters was carried out by direct empirical comparison with realworld data obtained from Air Data Units. Through analysis of the coincidence between theoretical models and actual In-Situ aircraft measurements, this validation revealed considerably diminished standard deviations and strong correlations, highlighting the significant convergence between theoretically predicted data and empirical data. The study conducted a thorough verification and validation of critical flight parameters, including Mach number, impact pressure, static pressure, dynamic pressure, and flight altitude, using barometric measurements and established theoretical formulations. The analysis revealed exceptional consistency between the measured values and theoretical calculations, with all parameters exhibiting remarkably low standard deviations and minimal relative errors, often represented with multiple decimal places. This high degree of precision underscores the reliability of the methodologies employed, affirming that the measurements align closely with theoretical expectations. The importance of this work lies in its contribution both to bridging the gap between theoretical atmospheric and barometric models and practical aviation applications, and to creating a reliable statistical base for validating barometric measurements, which ensures the reliability of the use of theoretical formulations in flight operations, thereby contributing to improving flight safety and optimizing operational characteristics in the aviation sector.

This research article addresses the critical challenge of subpixel object detection in long-range vision applications, where traditional convolutional neural network (CNN)-based methods demonstrate severely limited effectiveness. When objects are located at significant distances from the sensor, their visual representation degrades to mere pixels or subpixels, losing all semantic and geometric features that conventional computer vision systems rely upon. The authors propose an innovative dual-channel DRI-YOLO (Dynamic Radiometric Identification-You Only Look Once) algorithm that combines low-level radiometric characteristic analysis with neural network classification capabilities using an adapted YOLO architecture. The methodology encompasses three sequential processing stages: preliminary region-of-interest segmentation through statistical anomaly detection in pixel time series, detailed radiometric profile construction using Fast Fourier Transform (FFT) for periodic component identification, and final classification of dynamic radiometric signatures using a specially modified YOLO version optimized for frequency domain data. Experimental validation conducted on an NVIDIA Jetson Orin NX embedded platform with an IMX219 camera at 1280Ѕ720 resolution demonstrated the hybrid approach’s superiority, achieving mAP = 0.82 for subpixel objects compared to standard YOLOv8m’s performance below 0.15. Practical implementation confirmed reliable detection capabilities for birds at distances up to 500 meters and humans at 800 meters while maintaining a realtime processing rate of 26 frames per second. The proposed algorithm offers groundbreaking potential for video surveillance systems, unmanned technologies, and astronomical research applications where long-range subpixel object detection is paramount.