In recent decades, the concept of viability, proposed by S. Beer in the 1960s, has been actively developed. Viability differs from resilience in that it is a system’s ability to adapt to change and maintain dynamic equilibrium. This concept is widely used in various fields: social, economic, environmental, technological, organizational, and informational. Therefore, viability research is relevant for scientific and applied research, especially for complex multi-mode systems under conditions of significant uncertainty. Multi-mode systems designed and operated across numerous industries require a comprehensive (systems) approach to analysis and ensuring their reliability and effectiveness. Uncertainty, classified by degree, nature, and use of information, complicates traditional analysis methods. However, the lack of a unified methodological approach to viability assessment hinders a comprehensive study of the viability of these systems. In this publication, two key situations characterized by significant uncertainty are considered through an analysis of the functional and technological structures of multi-mode systems. In the first case, which can be described as situational-parametric uncertainty, information about operating modes, components, and their interactions is available, but data on the frequency of these modes’ application is lacking. In the second case, known as structural uncertainty, information about the modes, their intensity, and the frequency of element participation is available, but how they interact with each other is unclear. The structural-parametric analysis and synthesis performed in the context of the studied scenarios convincingly demonstrated the significant potential of applying a variety of mathematical tools for a comprehensive assessment of the survivability and viability of multi-mode systems.

It is proposed to use the estimates of these noise characteristics as a carrier of diagnostic information, utilizing artificial intelligence technology to monitor the start of the latent period of accidents. It is also shown that it is possible to employ artificial intelligence, using noise as a diagnostic carrier during low-magnitude earthquakes, to monitor the initial changes in the seismic stability of buildings.

Experimental studies have shown that during the operation of the objects under consideration, the occurrence of the latent period of accidents is due to the fact that under actual operating conditions equipment experiences destruction, fatigue cracking, residual stress, fatigue damage, fatigue, wear, and abrasion. As a result, during this period, the estimates of the cross-correlation function between the useful signal and the noise, as well as the noise variance of the vibration signals, are significant values. Furthermore, violations of classical conditions, such as the normality of the distribution law and the stationarity of the signals, have a negligible effect on the resulting estimates. Thanks to these technologies, the estimates of the cross-correlation function between the useful signal and the noise serve as reliable indicators for the onset of accidents using artificial intelligence technology. To ensure the adequacy of the control results, the sampling interval must be determined adaptively in real-time, which is easily achieved by utilizing the frequency properties of the least significant bit of the vibration signal samples.

In drum boiler control, model predictive control can provide accurate tracking under constraints, but residual model uncertainty still causes safety violations when parameters drift. Purpose. The study developed a robust model predictive controller for a drum boiler in which the safety margin is adjusted online according to the uncertainty level. Methods. А Takagi—Sugeno model with recursive least squares parameter adaptation described the nonlinear plant. Uncertainty was estimated from two sources: the covariance of the identified parameters and a sliding window of prediction errors. The combined estimate tightened output constraints and adjusted control smoothing. The robust layer does not require robust invariant set computation. Results. Numerical experiments on a drum boiler model compared four controller configurations during a 3000-step scenario with three phases of parameter drift. The configuration that combined model adaptation with the robust layer reduced the integral absolute error by 58.3 percent and reduced the number of constraint violations by 47.8 percent relative to the nominal controller. The maximum violation magnitude decreased by 33.8 percent. The configuration with the robust layer alone achieved the smallest number of violations, although it produced less accurate tracking. The average computation time per control step was 2.6 ms. Practical relevance. The proposed algorithm can be integrated into an existing predictive controller as a lightweight safety layer for thermal power equipment. Discussion. The results support the joint use of online identification and adaptive safety margins; future work should address multi-step uncertainty propagation and experimental validation on hardware.

Ensuring the stability of the communication structure in robotic swarms with limited communication range requires the development of effective mechanisms for adapting the interaction topology between agents. In dynamically changing environments, conventional fixed-topology approaches prove inadequate, necessitating adaptive solutions that function without comprehensive network knowledge. This study investigates a dynamic connectivity graph management method grounded in local node k-connectivity metrics, which quantify the minimum number of agents whose removal would isolate a given node from the network. We propose an algorithm for selecting control agents based on degree centrality criteria while accounting for the structure of connected graph components, implemented within a modular software environment for reproducible testing on double integrator and quadcopter models. Computations proceed in a decentralized manner using local measurements, minimizing communication channel load and enabling scalability as group size increases. Experimental evaluation on systems ranging from 10 to 100 agents demonstrates that the adaptive approach improves connectivity preservation probability from 64 % to 88 % under conditions with 20 % malicious agents in sparse topologies, while the average local k-connectivity increases from 2.3 to 3.8 for k-nearest neighbor scenarios with k = 3. Computational complexity analysis confirms the method’s scalability, with graph update times of 15—45 ms for systems up to 100 agents at a 20 Hz sampling frequency. The low latency of topology calculations enables integration into real-time control loops for micro-UAV and ground platform onboard controllers. Future work may incorporate data transmission delay models and kinematic constraints of physical robots to facilitate field testing. These findings advance decentralized control theory for distributed systems and provide a foundation for designing fault-tolerant robotic swarm architectures.

This article is focused on developing of autonomous «Swarm of drones« for monitoring and controlling territories based on multi-agent technology for a group of smart drones with collective decision making and negotiations, operating without operator intervention. In the proposed approach the software agents of drones independently allocate areas of the territory among themselves and create flight mission schedules. Upon detecting designated target objects, the drone agents collectively decide in real time which drone will follow the detected object and which will reorganize their flight missions and continue monitoring the remaining unoccupied territory. Multi-agent models, methods, and tools for solving the problem, as well as related navigation and pattern recognition technologies, are presented. The functions and architecture of the developed laboratory prototype of a hardware and software system for managing a group of drones are discussed. Recognition of mobile objects triggers collective, coordinated decisions by the agents to adaptively modify drone flight plans and routes in real time. The results of the experiments and the advantages of the developed multi-agent technology for unmanned control of groups of drones are discussed, as well as further steps in creating an industrial prototype of the system.

This paper presents the development and implementation of a quadcopter control method using hand gestures within the ROS2 framework. The approach enables intuitive operator-drone interaction in environments where conventional remote controllers are unavailable or impractical. An Intel RealSense D435i depth camera stream provides precise spatial localization of hand keypoints. High-level control functions — motor arming and disarming, takeoff, and landing — are executed through predefined static gestures. Gesture classification is performed by analyzing the spatial arrangement of hand keypoints in the image plane using a fully connected multilayer perceptron comprising two hidden layers with 168 and 546 neurons, respectively, achieving an optimal trade-off between inference speed and accuracy. Continuous quadcopter control is accomplished by computing roll, pitch, yaw angles, and throttle values derived from three-dimensional hand position analysis, utilizing the camera’s intrinsic calibration parameters and depth map data. To mitigate gimbal lock effects near 90° angles, a deviation range constraint is incorporated into the control pipeline. Control signal smoothing employs a combination of a moving average filter with window size 5 and exponential smoothing with coefficient a = 0.5, ensuring rapid system response while effectively suppressing outliers caused by image processing artifacts. All filtering parameters were determined through systematic experimental evaluation. The system achieves gesture classification accuracy exceeding 99 %, with per-frame processing time below 33 ms (~30 FPS), which eliminates control signal latency and ensures high responsiveness. Test flights conducted in the Gazebo simulator using a DJI Tello quadcopter model validated the proposed method and confirmed its practical potential for interactive unmanned aerial vehicle control applications.