No. 2 (2026)

The need to conduct explosive ordnance clearance operations in liberated territories has necessitated a significant expansion of humanitarian demining capabilities. Particular attention is being paid to increasing the number of highly effective robotic systems capable of searching for and destroying various types of explosive ordnance. The relevance of this study is determined by the fact that the Russian Federation is currently actively developing and producing robotic systems capable of solving these tasks with varying degrees of effectiveness. In many cases, these products are manufactured by companies lacking the necessary experience and, consequently, often failing to deliver the required quality of work. At the same time, the increasing demand for these resources and the increasing funding for their production has made it increasingly important to economically justify both their development and production strategies and the technologies for their application. Based on this, studies were conducted to develop an approach to optimize the allocation of financial resources when planning explosive ordnance clearance activities, identifying the most effective areas for mine clearance development, and improving the organizational structure of units equipped with the appropriate robotic equipment. This approach was based on an assessment of the likelihood of completing explosive ordnance clearance tasks with the assigned forces and resources, as well as an assessment of the costs of performing work included in a set of humanitarian demining activities using robotic equipment. The practical significance of the study lies in the fact that the obtained results will help identify the most cost-effective development directions for robotic humanitarian demining systems and prepare proposals for improving the organizational structure of units that will be equipped with these systems.

Robotization of technological processes in rail transport is a priority area of scientific and technical policy for Russian Railways. One of the technological processes being prioritized for robotization is train break-up from hump marshalling yards, primarily due to the need to increase station processing capacity and reduce workplace injury rates. The objective of this study is to determine the state of the art and identify development trends for robotic systems used in train break-up from hump operations. The research methodology involved analysis of scientific literature and patent information. As a result of this study, quantitative characteristics of publication activity in this technological area were determined, and materials from scientific and patent documents were reviewed and systematized. During material processing, it was established that solutions predominantly utilize either: two independent mobile platforms or two manipulators mounted on a single mobile platform carriage frame. These configurations are employed to enable parallel break-up of two trains from the hump and to improve the reliability of railcar cut formation under challenging conditions. A significant limitation in developing robotic systems for train break-up from hump operations is the uneven lengths of railcar cuts. Additionally, integration with external information and control systems is of critical importance. The practical significance of the obtained results lies in identifying development trends for robotic automation of train break-up processes from hump operations. These findings can be utilized in future development of promising robotic system architectures that will enhance marshalling yard processing capacity and operational reserves through close integration with external information and control systems

A key challenge in developing an information and control system for autonomous navigation of unmanned aerial vehicles (UAVs) in the absence of satellite communications is the generation of a local trajectory in the presence of obstacles (trees, power lines, etc.). The goal of this study is to develop heuristic algorithms that optimize the UAV's local trajectory using LiDAR data and construct a feasible local trajectory based on obstacle avoidance patterns. A two-stage concept is proposed: decomposing the LiDAR point cloud into oriented bounding boxes (OBBs) and generating a trajectory for traversing the OBBs using geometric patterns. The first stage implements a classic (geometric) LiDAR data processing pipeline: voxel thinning, ground plane extraction using the RANSAC method, DBSCAN clustering, and constructing bounding boxes around the clusters. This approach is implemented as a Python software module. Simulations were performed for two scenarios. The first scenario contained three obstacles, one of which was isolated, while the second and third were located in a group. The generated trajectory avoided all obstacles, with a trajectory construction time of 0.29 milliseconds. The second scenario was performed for a set of obstructions obtained by point cloud decomposition; the total number of obstacles, including the ground, was 678. [This is a fragment of the original text. The trajectory construction time in this case was 0.377 seconds. This approach provides predictable performance and a linear computational complexity estimate based on the number of obstacles, making it promising for use in autonomous navigation and UAV motion control systems

The introduction of groups of unmanned aerial vehicles (UAVs) into precision farming is hampered by several problems related to the stability of communications in areas remote from the take-off point and dependence on weather conditions, which can reduce the effectiveness of the technology. Existing planning models are insufficiently adaptive to dynamic changes in the agricultural environment, communication problems, and in most cases assume strict adherence to fixed trajectories. The purpose of this study is to develop conceptual models of seamless operation and a communication system for coordinating a group of agricultural UAVs. The paper presents a diagram of the connectivity of the system elements, which ensures the continuity of processes from initialization to automatic battery replacement and UAV refueling. A communication model of the system with the relay node has been developed that separates traffic, which makes it possible to eliminate collisions and stabilize data exchange over distances of more than 2.5 km. A method for distributing tasks, considering the energy supply and spatial coordinates of the UAV, is presented, which allows dynamically redistributing the load in case of failures. The proposed solutions enhance the autonomy and fault tolerance of the UAV group, minimize operator involvement, and ensure safe mission performance in a non-deterministic environment.
The approbation of the proposed model, performed in laboratory and field conditions, showed its stability when transmitting data over extended distances and in group operation. The absence of collisions during data exchange between UAVs has been separately confirmed, which indicates the correctness of traffic separation and the effectiveness of the chosen communication architecture. The results obtained confirm the possibility of practical application of the developed models to increase the autonomy and continuity of agricultural work.
The developed approach also demonstrates the potential for scaling, which expands the scope of its application in precision farming tasks.

The relevance of the work is driven by the increasing intensity of electronic countermeasures in the tactical command echelon, where traditional relay communication methods with fixed routes fail to ensure the required interference immunity and signal security. The objective of this study is to develop a method for adaptive route control of a UAV relay based on continuous monitoring of the electromagnetic environment in order to improve interference immunity of ground-to-air radio links. The primary method employed is multi-criteria flight path optimization with adaptive weighting coefficients, simultaneously considering three criteria: minimization of interference levels at the relay operating frequencies, ensuring radio accessibility with network correspondents, and reduction of detection probability by enemy electronic warfare assets. To reconstruct the spatial interference pattern from a limited set of onboard measurements, several interpolation and extrapolation methods are examined: inverse distance weighting, radial basis function interpolation, and statistical extrapolation based on a spatial correlation function. A cyclic operational procedure for the adaptive routing system is developed, comprising data collection, construction of a three-dimensional interference map, prediction of its evolution, calculation of the optimal trajectory, and monitoring of the maneuver outcome. Simulation results show that the proposed method increases the signal-to-interference ratio by 1.5–2 dB on average and up to 8 dB in the worst-case scenario compared to fixed-route flight. The practical significance lies in the possibility of implementing the proposed method using existing UAV platforms and radio-electronic equipment without fundamentally new technical solutions.

The relevance of this research stems from the need to create effective control systems for autonomous robots capable of operating in uncertain and dynamically changing environments. An environment representation system must address the challenges of localization, mapping, object detection and classification, dynamic prediction, and semantic interpretation. For an autonomous robot to navigate and perform goal-directed actions in its environment, it must understand its environment—without this, it will be unable to effectively plan movements, avoid obstacles, and reach destinations. Existing environment representation methods have limitations when adapting to unfamiliar, unstructured environments. The aim of this study is to develop the concept and architecture of an environment representation system based on a multi-agent neurocognitive architecture for controlling autonomous robots within a heterogeneous human-machine team. The scientific novelty of this study lies in the development of a multi-agent neurocognitive architecture for representing the state of the environment. The proposed approach enables the creation of flexible world models capable of self-organization and scalability with increasing knowledge. The research methodology is based on the use of multi-agent technologies and neurocognitive models. A multi-agent neurocognitive architecture has been developed, including mechanisms for collecting data from sensors, representing objects and subjects in the environment, forming a mechanism for constructing cause-and-effect relationships, and sharing knowledge between members of a heterogeneous human-machine team. The developed architecture enables the scalability of decision-making systems and facilitates knowledge sharing between team members. A promising direction for further research is improving the system's safety mechanisms. The results of this study can be used in the development of next-generation autonomous robotic systems.

With the growing demand for mobile robotic complexes with an increased degree of autonomy, the demand for their information awareness increases significantly. Vision systems play a leading role in ensuring this awareness. To support the functioning of groups, as well as each individual robot, it is necessary to form an information field that allows for prompt decision-making in the control systems of both individual robots and the entire group as a whole. The desire to assign an increasing number of independently solved tasks to the robotic complexes requires an increasingly broad and detailed consideration of the field of operation from the VS. Distributed space surveillance and control systems are one of the promising areas for solving this problem, ensuring the formation of a unified information field based on cooperative perception, joint localization and semantic interpretation of the situation. The paper considers the problem of monitoring a given area of space by a distributed vision system in order to detect objects of interest with a known characteristic size in this area. At the same time, it is assumed that video cameras can be located on both ground and air vehicles. The layout of a distributed vision system for solving the formulated problem is described. The core of the described system is a unified information module with an omnidirectional recording unit.
The module acts as an agent of the distributed VS. Several such modules are distributed over the area of space indicated on the map, taking into account the terrain features and other objects located in this area. The problem of choosing the optimal composition of a distributed system according to a number of criteria is solved. A methodology is proposed for planning the location of agents with specified characteristics of recording units on a given 3D terrain map to ensure the detection of objects of interest with known characteristic dimensions. An algorithm is given for calculating the locations of VS with fixed fields of view for monitoring a given area of rough terrain in order to identify static objects with known characteristic dimensions and differences from the surrounding background in reflectivity or moving objects

The relevance of this research stems from the contradiction between the need to increase food production in an urbanized environment and the fragmentation of existing high-tech solutions (hydroponics, aeroponics, IoT), which are being implemented in isolation, without a unified management methodology. The lack of unified approaches to data collection and adaptive control of environmental parameters limits the scalability of vertical farms. The goal of this research is to develop and theoretically substantiate the architecture of a universal adaptive management model for closed-loop agricultural production systems, integrating various cultivation methods based on machine learning algorithms. The methodology is based on a systematic analysis of scientific publications and experimental data on the use of embedded devices and machine learning algorithms in hydroponic, aeroponic, and soil-based vertical greenhouses. Based on this data synthesis, parametric matrices were constructed to standardize technological processes. The main results include the development of a structural diagram of a universal model that enables the integration of disparate systems into a single platform with the ability to continuously monitor and perform predictive analytics. The minimum required sensor set is substantiated: pH, EC, temperature, humidity, CO₂, PAR, pressure, and nutrient solution flow. The proposed architecture enables dynamic switching between hydroponic, aeroponic, and indoor modes within a single phytotron. The conclusions and significance of this work lie in creating a foundation for designing scalable vertical farms with predictable profitability and resource efficiency indicators, while enabling continuous further training of AI algorithms for predictive microclimate management and early plant disease detection in urban environments

The paper considers the problem of constructing a chassis for a research lunar rover with ultra-high traversability over uncertain terrain with soft soil. Direct remote and supervisory control of modern complex robotic systems in non-deterministic environment places an increased workload on the operator, especially in the case of high-traversability mobile platforms with a large number of degrees of freedom (DoF) requiring coordinated control. In this regard, the problem of automating the movement of such a multi-DoF chassis as well as automating the motion planning depending on situation based on sensor feedback becomes relevant. This article proposes a concept for a multi-degree chassis for a research lunar rover, including a design and layout scheme of the chassis and a method for its application, i.e., motion algorithms on various types of rough terrain. A methodology for control algorithms design is proposed, and a brief description of the developed algorithms and the simulation models of the chassis for its preliminary testing is provided. The final outcome of this work is expected to be a number of experimentally obtained characteristics of the laboratory chassis model and the verification of the developed computer models and control algorithms of the multi-DOF chassis. Completing these tasks will provide scientific and technical groundwork for the motion control of wheeled- legged systems and improve the quality of lunar rovers' modeling and design. The results obtained at this stage allow us to move on to the manufacturing an experimental prototype and conducting physical experiments on this prototype to test the developed algorithms and simulation models. Experimental verification of the multi-DoF chassis control algorithms and their design methodology will improve the level of autonomy of future mobile robots designed to operate in extreme off-planet conditions

The execution of inspection and manipulation operations by autonomous unmanned underwater vehicles (AUVs) under operator supervision is limited by the low bandwidth of the hydroacoustic communication channel, which does not allow the transmission of video streams and high-quality photographic images in real time. At the same time, existing methods for automatic recognition of target objects by onboard machine vision systems of AUVs do not guarantee accurate determination of the object’s position and shape when defects are present on its surface (siltation, biofouling, mechanical damage). To address this problem, the paper formulates the task of developing a method that enables information interaction between the operator and the AUV via a low-bandwidth hydroacoustic communication channel. This interaction allows the operator to evaluate the quality of target object recognition and subsequently generate target designations for the vehicle and its onboard manipulator. The proposed method is based on an algorithm for identifying objects with defects, which employs a modified ICP point cloud registration approach with the exclusion of defective regions and characteristic features (CFs). The detected CFs and defect regions are transformed into a compact set of geometric primitives transmitted to the operator through the hydroacoustic communication channel with the AUV. Based on the received information, incoming telemetry from the AUV, and a reference three-dimensional model of the object, the graphical interface visualizes the scanned scene, allowing the operator to assess the recognition accuracy. Full-scale experimental tests of the method were conducted using a hydroacoustic modem manufactured by the Institute of Marine Technology Problems, Far Eastern Branch of the Russian Academy of Sciences (IMTP FEB RAS) and an onboard computer NVIDIA Jetson TX2. The information packet transmitted within
1 minute contained the object type, its position, a defect map, and the coordinates of the characteristic features, with a data volume three orders of magnitude smaller than that required for transmitting the full point cloud of the scanned object. Based on the received information, the operator confirmed successful object identification even when up to 30% of the surface area was covered by defects. The practical significance of the proposed method lies in enabling the execution of critical missions under operator supervision in conditions of environmental uncertainty

When developing algorithms for real-time robot path planning, the problem of performance limitations of the corresponding classical algorithms arises. This paper considers a method for planning robot movements in a two-dimensional complex conflict environment. For planning in complex environments, a hybrid planning algorithm is proposed, based on a combination and synthesis of the classical cellular decomposition algorithm and a recently proposed algorithm based on the characteristic visibility graph. This algorithm involves a preliminary analysis of the complexity of the obstacle scene, based on the results of which one of the two specified particular algorithms is selected. It is shown that this approach can significantly overcome the limitations of both of these algorithms. A disturbance avoidance method based on the apparatus of characteristic probability functions is described in a compact form, and its relationship with planning methods in complex environments is demonstrated when solving corresponding problems of global optimization of the probability of successful completion of a target trajectory. The developed approach examines the relationship between the probability of successful path completion in a source field and the corresponding risk function. To solve global robot motion planning problems in complex conflict environments, the proposed hybrid algorithm is first proposed for constructing a family of initial curves within the appropriate feasible motion corridors, ignoring sources. A family of local optimization problems is then solved within the feasible motion corridors, taking sources into account. Next, the trajectory with the maximum probability of successful completion or the normalized safe motion function is selected

This paper proposes a methodology and method for constructing a model of the study area as a finite set of zones (sections) covering it, characterized by the fact that all sections are rectangular. The paper examines methods for forming a minimal set of sections on a large field that completely cover the accessible territory surveyed by unmanned aerial vehicle (UAV) sensors. The size, orientation, and relative positions of the sections are aimed at minimizing their survey time. In general, a route M is a sequential set of linear segments. Route M is divided into segments using a set of control points P. A methodology and algorithm for constructing an optimal UAV route based on the ant colony method have been developed. Mechanisms for controlling the UAV's movements along the route have been developed. In general, a route M is a sequential set of linear segments. A segment of the UAV's path (trajectory) over a surveyed area of territory being scanned (explored) is called a working segment, while a segment of the UAV's path (trajectory) over a non-scanned (explored) area is called a dummy segment. Generally, a route is an alternating sequence of working and dummy segments, replacing each other. A methodology and algorithm for moving an UAV between reference points of a segment corresponding to reference points have been developed. Two algorithms represent the solution search procedure: Algorithm 1, which describes the behavior of an ant colony; Algorithm 2, which describes the behavior of an agent. An adaptation unit supports the process of moving an UAV along a route in real-world conditions. The adaptation unit's task is to control the UAV's movement along a reference line along the route. The control method involves replanning the motion parameters of an unmanned aerial vehicle (UAV) moving parallel to a reference vector at each moment during flight. Adaptation of the UAV consists of adapting the control parameter values. A structure of maneuvers performed by UAVs to correct parameter deviations is proposed. The adaptation task consists of generating a sequence of adaptive actions in the adaptation machine that extremize the quality indicators of the resulting solutions (adaptation criteria). The adaptation object is a set of continuous flight control parameters: the UAV's deviation from the reference line; the angle between the UAV's motion vector and the reference line of the current segment; and the UAV's flight altitude above the current segment. Adaptation of the UAV consists of adapting the control parameter values

A large number of different interfaces can be used to control robots, from traditional remotes to augmented reality technologies. However, all such interfaces have a number of limitations, which are particularly acute in service robotics. They are associated with long-term training of a human operator, non-intuitive control for humans, and the need for full human involvement. On the other hand, a new direction has emerged today, related to large language models that are capable of processing natural language and then translating it into robot control commands. There are a number of works demonstrating the possibility of using language models in tasks of planning robot actions. Based on the analysis of existing work, a new method of controlling a mobile robot is proposed, combining the advantages of other methods. The method allows you to plan scenarios for the robot, receiving a natural language mission, the robot's TOP, and information from its sensors. The article also describes the sequence of configuring the system using a large language model to solve this problem. Three variants of instructions for the neural network are presented, which gradually improve the achievability of the generated scenarios. After that, various missions are described, which are set as part of experimental studies - a total of 100 missions were tested, divided into 4 levels of difficulty in equal proportions. The complexity of the missions ranged from describing objects in the robot's field of view to interacting with complex missions involving synonyms of objects and implicitly defined goals. At the end of the work, the results of the evaluation of the algorithm and three variants of the instruction are presented. The conclusion can be considered that the use of language models to assign scenarios to robots is possible, including with a sufficiently high achievement. The model with the most advanced instruction reached 91% of correctly formed scenarios, which suggests the applicability of the developed method for controlling a mobile robot in natural language

The relevance of the problem of adaptive control of nonlinear dynamic objects is ensured by the ever-increasing demands on the quality and operating conditions of technical systems. Automation and robotics lead to an increase in the complexity of the problems being solved and the need to adapt to structural and parametric uncertainty and external disturbances. In recent years, the use of reinforcement learning methods for the synthesis of adaptive control systems has gained popularity. These methods demonstrate effectiveness in controlling uncertain dynamic objects. However, there are two fundamental problems with the application of machine learning methods to control systems for dynamic objects. First, the need to ensure asymptotic stability of the desired trajectory of a closed-loop system limits the application of deep learning methods. Second, during the learning process, it is necessary that the intelligent controller does not generate controls that lead to state variables exceeding specified limits. The purpose of this article is to review and analyze recent advances in the application of reinforcement learning methods to the synthesis of adaptive control systems for nonlinear dynamic objects. Particular attention is given to Actor-Critic methods, which are structurally similar to adaptive control systems with self-adjusting parameters. Based on the analysis, the structure of an adaptive system with two-component control, including nominal and adaptive controllers, is proposed. An adaptive control algorithm is proposed, distinguished by the use of a modified Actor-Critic algorithm, distinguished by a new form of the delta error and the absence of a singularity in the neighborhood of zero. The proposed algorithm allows for a reduction in the number of adjustable parameters. The article presents conditions for Lyapunov stability of the zero-equilibrium position of a closed-loop system and an example of the synthesis and modeling of the proposed adaptive control algorithm

The search for designated bottom objects is one of the most difficult tasks solved by the AUV, due to a number of factors, the main of which are: the variety of search objects (sunken submarines, surface ships, airplanes, helicopters, mines, underwater pipelines and communication cables, various underwater infrastructure), the need for integrated use for search for various bottom monitoring tools that differ in their physical principles of operation, resolution, and search performance. The purpose of the work, the results of which are presented in the article, was to develop an algorithm for managing the AUV when searching for a designated bottom object that meets these requirements, and to verify it using a digital polygon and a digital twin of the AUV. The probability of correctly attributing the detected bottom object to the search object was chosen as a criterion for choosing a bottom monitoring tool in each specific case. As a result, the logic of searching for a designated bottom object is as follows. The search for bottom objects is carried out using a tool with maximum search performance. When a bottom object is detected, the probability of its belonging to the search object is determined. If it exceeds the set high threshold, a decision is made to locate the designated bottom object. If it is less than the specified low threshold, it is decided that an extraneous object has been detected. In other cases, a decision is made on the need to examine the object with a higher resolution. The technology of classification of bottom objects based on the training of an artificial neural network trained using synthesized training material is described. The results of checking the effectiveness of the developed algorithm using a digital polygon and a digital twin of AUV are presented. The simulation of the developed algorithm showed that the integrated use of bottom monitoring tools increases the likelihood of a successful solution to the problem and reduces the time needed to solve it.

This paper examines a class of telecommunication networks with mobile nodes, an important subclass within which is the so-called geography-aware networks. Their distinguishing feature is the availability of information about the geographic coordinates of all nodes in the network to each individual node, whereby each node is aware of the complete topology of the network graph, which enables rapid identification of the required number of information transmission routes between any two nodes. The purpose of this article is to investigate multipath routing algorithms. To this end, we propose a method for the automatic synthesis of adequate test models capable of representing networks of virtually unlimited complexity. The proposed solution is based on a compositional approach, in which a relatively simple network fragment that satisfies given constraints is first formed, and then the resulting model is constructed as a composition of copies of this fragment. To investigate the efficiency of multipath routing algorithms, we propose a compositional method for the random synthesis of test models of complex networks that satisfy constraints on distances between the vertices. This method was applied to the investigation of two routing algorithms, resulting in a large body of illustrative model data. The obtained results, presented in the form of graphs, demonstrate an increase in the gain in message transmission time as the queue length in the target flow grows. While the efficiency of multipath routing is insignificant under low network load conditions, its usefulness increases with growing network load. A similar increase in efficiency is demonstrated when transitioning from an algorithm that does not allow intersecting paths to an algorithm that allows such intersections.

The article addresses the relevant task of ensuring the autonomy of mobile robots in complex conditions, where the use of traditional navigation methods based on global coordinate systems and satellite data is impossible or ineffective. To solve this problem, an approach based on cognitive (interpretive) navigation is proposed, where semantic understanding of the environment plays the central role. The key feature of the method is the construction of a cognitive map – a semantically oriented graph whose vertices correspond to landmark objects (or groups of homogeneous landmarks), and whose edges correspond to fixed sets of information-motor actions (elementary conditioned behavioral patterns). Thus, the robot's route while moving along the cognitive map is reduced to a fixed set of information-motor actions. The map construction process is carried out automatically based on a pre-obtained semantically segmented image of the terrain, which allows the mobile robot to acquire a priori information about the relative positions and shapes of the landmarks. To formalize the navigation process and manage the robot's behavior based on the cognitive map, the authors propose a specially developed formal language, LRNB (Language of Robot Navigation Behavior). This language allows the decomposition of complex missions into elementary information-motor actions, the specification of their completion conditions, and the description of interaction scenarios with dynamic and static objects. The work details the principles of building a cognitive map, the syntax of the LRNB language, and the mechanism for forming a route as a sequence of commands. The practical part includes the results of verifying the approach in a simulation environment using a specially developed emulator, as well as preliminary field tests on a laboratory tracked mobile robot, which confirmed the fundamental feasibility of the proposed approach. The obtained results indicate the potential of the method for application in critically important scenarios, such as disaster zones, areas of electronic warfare, and other environments with a high degree of uncertainty. Further work plans are proposed, related to bringing experimental conditions closer to the real-world conditions of potential operation.

This paper addresses the pressing challenges of collaborative autonomous video navigation of unmanned aerial vehicles and ground robots in urban environments, including dense urban development and buildings, as well as in rugged terrain, including mountainous and wooded areas, where, as in urban environments, the use of traditional remote control and navigation tools may be limited by the presence of shielded areas. It is proposed to solve group navigation problems using data from an onboard vision system during operational reconnaissance of the work area by an unmanned aerial vehicle. The results ensure autonomous movement and flight of both individual heterogeneous robotic systems and in a group. The navigation algorithms are based on the methods and algorithms for processing data from the onboard vision system, consisting of a complex of mutually adjusted lidar, television camera and thermal imager, which form the geometry of the surrounding space in the form of a point cloud with the distribution of color and temperature fields on it, allowing for the effective solution of the SLAM problem (determination of the current coordinates of the control object with the formation of a geometric model of the external environment) and the classification of the working area according to the criteria of geometric and support cross-country ability, which ensures autonomous flight and movement of robots for air and ground use in urbanized environments and on rough terrain. It is proposed to use an information and navigation field, represented as a visibility graph, to organize the autonomous operation of aerial and ground robots, including in a group, and a set of reference images, allowing for the correct execution of planned trajectories, taking into account errors in the video navigation task. This information and navigation field allows for the compact presentation of information necessary and sufficient for the autonomous operation of unmanned aerial vehicles and ground robots and facilitates its exchange between group members. The results of experimental studies in real-world conditions of urbanized environments and rugged terrain are presented, confirming the effectiveness of the proposed methods, algorithms, and corresponding software and hardware

The relevance of this research is driven by the growing need for safe validation of navigation algorithms for autonomous mobile robots operating in cluttered and dynamically changing environments, where the use of physical equipment entails risks of damage and high costs. The aim of the work is to develop a rigorous methodology for software emulation of a technical vision system based on complementary virtual infrared and ultrasonic sensors. To achieve this aim, the following tasks were solved: formalization of the kinematic model of a differential drive with a state vector [x, y, θ]ᵀ; mathematical description of nonlinear triangulation for IR sensors and the physics of ultrasound propagation using the time-of-flight method d = c·t/2; integration of additive Gaussian noises with parameters σ_us = 0.005 m, σ_ir = 0.002 m; implementation of heterogeneous data fusion using an Unscented Kalman Filter. The navigation controllers employed were the artificial potential field method with attractive and repulsive components, and fuzzy logic controllers. Experimental validation in a Python simulation environment of a maze with static obstacles demonstrated an average positioning error of 0.2 m in spherical configurations and an obstacle detection accuracy of 89.61%. The novelty of the proposed approach lies in providing a deterministic link between theoretical trajectory planning and physical implementation through the synergistic use of optical and acoustic sensory modalities. The practical significance of the work consists in a substantial reduction in the development and testing time for intelligent robotic systems, owing to the possibility of preliminary debugging of perception, data fusion, and navigation algorithms in controlled emulation conditions without the need for expensive hardware

The aim of this study is to develop and experimentally evaluate a method for improving the quality of marine vessel detection based on the Attention RPN architecture under conditions of limited training data, with the integration of inter-class hard negative mining. To achieve this goal, the following were implemented: model adaptation in a fine-tuning mode based on five examples, creation of a specialized dataset consisting of 9 vessel categories (637 training and 175 validation images), and an algorithm for selecting hard negative pairs using cosine similarity. The methodological foundation includes deep learning approaches for object detection based on Siamese neural networks, analysis of feature distributions in embedding space, and algorithms for generating hard negative samples by computing inter-class distances using the cosine metric. Experimental evaluation was carried out on standard few-shot object detection benchmarks. The integration of cross-class hard negative mining reduced the proportion of false positive detections by approximately 18% and increased the mean Average Precision (mAP) by about 6% on average compared to the baseline model without hard negative mining. The greatest improvement (up to 9.1%) was observed for classes with large target objects. The results demonstrate that targeted selection of inter-class hard negative examples significantly improves the robustness of metric-based detectors under conditions of strong class imbalance and high intra-class similarity. The practical significance of the work lies in the applicability of the proposed approach for deploying video surveillance systems in tasks where large training datasets and extensive annotation are unavailable

Import substitution of hardware–software control systems is a key enabler for the development of Russia’s domestic robotics industry. The limited software ecosystem of Russian computing platforms still constrains their use in onboard computing systems (OCS) of robotic systems (RS), even though modern domestic multicore processors—such as the Elbrus—offer competitive performance and energy efficiency. ROS (Robot Operating System) and its successor ROS 2 are widely adopted for robotics software development, but official support is largely restricted to x86 and ARM architectures running mainstream Linux distributions. At the same time, a substantial amount of deployed application software remains ROS-based, while ROS 2 continues to evolve rapidly. This creates a practical need to support both ROS and ROS 2 on domestic hardware–software platforms. This work investigates approaches to unifying robotics middleware based on ROS and ROS 2 for the Elbrus platform. We implemented two porting strategies:
(1) application-level binary translation and (2) a fully native build. The native approach included adapting architecture-specific dependencies, modifying architecture-dependent code, and producing Debian (deb) packages. We then evaluated both approaches under multiple runtime configurations, including scenarios where ROS and ROS 2 operate together. Experimental results confirm that the Elbrus platform can be integrated into the modern robotics software ecosystem. They also show that the native implementation delivers higher performance than the binary-translated solution. The proposed work expands the toolchain available for robotics development on domestic platforms and reduces technological dependence when building control systems for robotic systems.

This paper addresses the problem of ensuring the motion safety of a mobile robotic platform in an environment with known object classes, based on object detection results from a surround-view camera system. While solutions based on optical flow or other methods for detecting moving objects in a camera's field of view are well-known, this work proposes to use the results from a dedicated object detector for safety assurance. Here, we propose to utilize object positions, calculated from the output of a neural network detector, for safety purposes. The surround-view camera system (SVS) consists of 4 cameras with wide-angle lenses. Several approaches for feeding objects into the neural network detector are considered. Differences in quality and performance for these approaches are demonstrated, and methodological recommendations for their use are formulated. To calculate object positions in the camera coordinate system and, through camera positions, in the robot's coordinate frame, 2 solutions based on camera calibration and certain assumptions are proposed. In the first case, the robot's position on the surface is assumed to be horizontal, and the camera's height above the surface is assumed constant and known. In the second case, one of the metric dimensions of objects for each class are assumed to be known in advance. The proposed solution has been deployed and tested on a Rockchip 3588-based computing platform, demonstrating high performance in terms of detection count (from 8 to 19 objects per frame on average, depending on settings) and processing speed (0.29 s for 8 objects per frame and about 1.05 s for the result of 19 objects per frame). Regarding the accuracy of distance estimation to objects, for the first method, the standard deviation ranged from 4.2 to 7.9 mm, and for the second method, from 3.8 to 7.6 mm. The standard deviation of object size estimation for the first method ranged from 0.64 to 2.02 mm, and for the second method from 1.63 to 2.22 mm, respectively. The obtained results allow us to confidently state that the proposed object detection algorithm is applicable for ensuring the safety and navigation of a mobile robot using surround-view system cameras

This paper presents a fully learnable system for solving the problem of matching two images. All the main elements of the system are trainable, i.e. their final form corresponds to the target dataset on which the training was carried out. The fact that the system is trainable, the methods used in training and the architecture of the system allow using the system to solve a large number of various computer vision problems. The system consists of a convolutional neural network that serves both to extract key points and their descriptors, as well as a trainable matcher of the extracted key points based on their description and mutual arrangement in the observed scene. The used convolutional neural network processes full-size images and calculates both the location of interest points with pixel accuracy and the descriptors associated with them in a single forward pass. Matching key points is a separate step and is performed after the forward pass of the neural network. In the process of training the model for calculating the positions of key points and their descriptors, a method for forming a training sample is used, called homographic adaptation - an approach that helps to increase the repeatability and accuracy of detecting key points. The process of training the feature point detection model consists of obtaining new weights in the process of additional training of the base detector, which represents the initialization weights of the model. The final feature point detection model, trained on the universal MS-COCO image set using homographic adaptation, repeatedly outperforms the original base detector in terms of the number, reliability and repeatability of feature points, and also outperforms any other traditional corner detector based on classical approaches