Path planning in environments with obstacles is addressed through the construction of visibility graphs, known for high accuracy but computationally expensive edge intersection checks. A fully vectorized approach is proposed, in which all candidate visibility edges are processed against all obstacle edges simultaneously, reducing computation time. Polygonal contours are simplified, which decreases the number of vertices without affecting path optimality, thus further accelerating graph construction. Performance improvements are demonstrated through experiments, showing a substantial reduction in graph construction time compared to traditional and partially vectorized methods. Integration with various polygon extraction techniques is explored, and comparisons are made with other path-finding algorithms. In particular, comparisons are made with Theta*, A* and PRM, as well as with modern sampling-based methods such as RRT*, BIT*, FMT*, and Lazy Theta*. Global path optimality is preserved while achieving competitive or superior construction times. Visibility graphs enable replanning by updating only the necessary edges without reconstructing the entire graph. The graph’s sparse structure combined with optimized vectorized construction enables high-speed path planning. Paths were successfully and quickly found on large maps with a large number of obstacles for various navigation tasks of mobile robots. Overall, on small maps the vectorized visibility graph construction is up to 100 × faster than Theta*, and on larger, obstacle-rich maps it still delivers roughly a 5 × speedup. It also outperforms PRM, as increasing PRM’s sampling density to approach optimal paths leads to a rapid growth in computational cost. Similar trends were observed when compared with other sampling-based planners such as RRT*, BIT*, and FMT*. These results confirm the suitability of the vectorized visibility graph approach for high-performance mobile robot navigation.

This paper describes the challenge of trajectory planning for robotic manipulators operating in obstacle-rich environments. The goal is to propose a novel approach that integrates the TrajOpt algorithm with reinforcement learning (RL) to improve the efficiency and accuracy of motion planning. TrajOpt, a numerical optimization-based method, generates successful trajectories, while RL produces high-quality initial trajectory estimates, serving as strong starting points for optimization. This integration enables TrajOpt to find feasible paths more efficiently and reduces collision risks. RL further refines initial trajectories, enhancing navigation in complex environments. Experimental results highlight the benefits of merging machine learning with traditional optimization methods. The RL-based approach achieved an 82

Generalist robot policies must operate safely and reliably in everyday human environments such as homes, offices, and warehouses, where people and objects move unpredictably. We present Dynamic Neural Potential Field (NPField-GPT), a learning-enhanced model predictive control (MPC) framework that couples classical optimization with a Transformer-based predictor of footprint-aware repulsive potentials. Given an occupancy sub-map, robot footprint, and optional dynamic-obstacle cues, our NPField-GPT model forecasts a horizon of differentiable potentials that are injected into a sequential quadratic MPC program via L4CasADi, yielding real-time, constraint-aware trajectory optimization. We additionally study two baselines: NPField-StaticMLP, where a dynamic scene is treated as a sequence of static maps; and NPField-DynamicMLP, which predicts the future potential sequence in parallel with an MLP. In dynamic indoor scenarios from BenchMR and on a Husky UGV in office corridors, NPField-GPT produces more efficient and safer trajectories under motion changes, while StaticMLP/DynamicMLP offer lower latency. We also compare with the CIAO* and MPPI baselines. Across methods, the Transformer+MPC synergy preserves the transparency and stability of model-based planning while learning only the part that benefits from data: spatiotemporal collision risk. Code and trained models are available at https://github.com/CognitiveAISystems/Dynamic-Neural-Potential-Field

Model Predictive Control (MPC) is a powerful tool for planning the local trajectory of autonomous mobile robots. The paper considers a new algorithm for trajectory planning and obstacle avoidance based on the MPC technique known in Artificial Intelligence (AI) planning and robotics. We have proposed an original method for decomposing obstacles to form a potential field, which in turn is used as an additional component in MPC. Thus, we propose a new intelligent trajectory planning method that takes into account the special shape of obstacles, which in turn significantly improves the metrics of intelligent agent movement on the well-known Moving AI benchmark. The challenging aspect of MPC planning is collision avoidance on large and complicated grid maps. We propose the Polygon Segmentation for obtaining Artificial Potential Field (PolySAP). This local planner approximates the obstacles on the map with a set of polygons. We address the question of how to partition a map with polygons to make it fast and effective for a practical MPC planner. We propose a decomposition algorithm based on Straight Skeleton. Our algorithm returns a set of polygons, which are then convexified. Numerical experiments show that our method outperforms basic algorithms in performance and provides sufficient partition quality for effective planning. We propose an artificial potential function calculated for polygonal obstacles and added to the MPC objective for collision avoidance. We evaluate our approach on city map dataset and on a real robotic platform. Numerical experiments show that PolySAP allows for polygon decomposition that is five times faster than Interior Extensions. Our MPC solver provides a fast solution for the MPC task compared to the state-of-the-art MPC planners. Our planner ensured the safe motion of the real mobile robot through a narrow indoor environment. Our code is available at https://github.com/alhaddad-m/PolySAP.

This article proposes an approach for path planning in environments with obstacles. The methodology integrates recent neural obstacle polygon extraction with visibility graph path planning, complemented by the integration of vectorization techniques. That significantly enhances path planning efficiency. The modular design of the neural network method, encompassing contour detection, vertex identification, and polygon approximation modules, facilitates improved performance compared to traditional methods. Furthermore, the investigation into vectorization’s impact on the intersection operation accelerates algorithm speed, contributing to faster path planning processes. Experimental results validate the efficacy of the integrated approach, showcasing notable improvements in path planning efficiency, especially with the utilization of vectorization techniques. The study systematically addresses challenges such as slow graph construction and inaccurate obstacle detection, providing a robust solution for optimizing path planning processes. Moreover, the implementation of multiple modules in the methodology enables its versatility for testing with various environments. This versatility allows researchers to assess the method’s performance across diverse scenarios and visualize the results effectively. Overall, the integrated approach offers a comprehensive solution for optimizing path planning in complex environments, demonstrating its potential to streamline path planning processes and improve mobile robot navigation.

In this work, we develop an optimization-based solution for a manipulation planning among obstacles. This task is particularly challenging for collaborative manipulators when the operations include movements through the singular configurations. Trajectory optimization for this case requires a collision model, which allows converging to a valid solution from an invalid initial guess. By improving the initial trajectory approximation and optimizing collisions with advanced obstacle representation, it significantly enhances trajectory planning accuracy and efficiency. In our work, we propose a method based on the combining of obstacles. We use Octomap as the baseline (obstacles are represented as a set of cubes). This creates the need to calculate each cube when calculating the trajectory. Enlarging cubes to parallelepipeds allows one to obtain a collision-free trajectory faster. We have conducted a set of experiments with Octomap and with the representation of obstacles in the form of enlarged cubes. Experiments have shown that enlarging cubes to parallelepipeds reduces planning time and increases the success rate.

Model predictive control (MPC) may provide local motion planning for mobile robotic platforms. The challenging aspect is the analytic representation of collision cost for the case when both the obstacle map and robot footprint are arbitrary. We propose a Neural Potential Field: a neural network model that returns a differentiable collision cost based on robot pose, obstacle map, and robot footprint. The differentiability of our model allows its usage within the MPC solver. It is computationally hard to solve problems with a very high number of parameters. Therefore, our architecture includes neural image encoders, which transform obstacle maps and robot footprints into embeddings, which reduce problem dimensionality by two orders of magnitude. The reference data for network training are generated based on algorithmic calculation of a signed distance function. Comparative experiments showed that the proposed approach is comparable with existing local planners: it provides trajectories with outperforming smoothness, comparable path length, and safe distance from obstacles.

The application of learning-based control methods in robotics presents significant challenges. One is that model-free reinforcement learning algorithms use observation data with low sample efficiency. To address this challenge, a prevalent approach is model-based reinforcement learning, which involves employing an environment dynamics model. We suggest approximating transition dynamics with symbolic expressions, which are generated via symbolic regression. Approximation of a mechanical system with a symbolic model has fewer parameters than approximation with neural networks, which can potentially lead to higher accuracy and quality of extrapolation. We use a symbolic dynamics model to generate trajectories in model-based policy optimization to improve the sample efficiency of the learning algorithm. We evaluate our approach across various tasks within simulated environments. Our method demonstrates superior sample efficiency in these tasks compared to model-free and model-based baseline methods.

The possibility of taking account of the granulometric composition of coke in assessing its metallurgical properties is considered. In practice, the main coke characteristics vary with forced or deliberate change in the coal batch. The research is based on operational data for blast furnaces and the characteristics of coke in periods when its hot strength and granulometric composition change. A priori data regarding the influence of the hot strength CSR on the decrease in piece size of the coke in the reaction zone are used to formulate an equation taking account of both the equivalent diameter of the coke pieces and their CSR values. The proposed index takes better account of the change in blast-furnace operating conditions with change in coke quality.

75 % of the total carbon dioxide emissions by ferrous metallurgy enterprises is generated in the blast furnace process. One of the directions of CO2 emission reduction in pig iron production is partial replacement of carbon monoxide with hydrogen as a reducing agent. It is shown that such a replacement can lead to a decrease in the total carbon consumption due to reduction of heat consumption for the direct reduction of iron oxides. Using a mathematical model of the blast furnace process, the efficiency of partial replacement of process fuel (coke, natural gas, pulverized coal) with a hydrogen additive was evaluated. Calculations were performed for the operating conditions of blast furnaces of EVRAZ NTMK JSC, which melt vanadium-containing titanomagnetites. The coefficients of process fuel replacement with hydrogen and the coefficients of the influence of the replacement of process fuel with hydrogen on the change in CO2 emissions are calculated. The dependence of the change in the productivity of the furnace on the consumption of hydrogen at a constant minute flow rate of the blast and its adjustment to maintain the pressure drop has been established. It is shown that in the absence of gas dynamics reserves, the replacement of process fuel with hydrogen will be accompanied by a decrease in furnace productivity. The smelting of pig iron from titanomagnetites with the replacement of technological fuel with hydrogen will complicate the refining of melting products due to an increase in the formation of titanium carbides and carbonitrides. The reduction in CO2 formation at hydrogen entering the blast furnace with natural gas is 0.35 kg/m3 compared to pure hydrogen replacement of 0.73 kg/m3, not taking into account CO2 emission during hydrogen production

Исследованы образцы титансодержащего шлака АО «ЕВРАЗ НТМК» от доменного передела железорудного концентрата уральских титаномагнетитовых руд Качканарского месторождения с целью изучения возможности извлечения титана. Исходные образцы отличаются способом кристаллизации. Один образец получен при послойном сливе шлака в траншею и охлаждении его водой (далее – шлаки с быстрой кристаллизацией), второй образец – шлаковые корки, оставшиеся в чаше после слива жидкой части (далее – шлаки с медленной кристаллизацией). Химический и фазовый составы всех исследованных образцов аналогичные, содержание TiO2 составляет ~10,5%. Основной фазой (матрицей) является акерманит-геленит, которая составляет ~65%. Титансодержащей фазой является перовскит ~26%. Шлаки с быстрой кристаллизацией характеризуются мелкой структурой титансодержащей фазы со средним размером частиц ~13 мкм. Титансодержащая фаза состоит на 88% из частиц площадью более 100 мкм2 со средним размером частиц ~17 мкм. Шлаки с медленной кристаллизацией характеризуются более крупной структурой титансодержащей фазы, размеры таких частиц ~30 мкм. Титансодержащая фаза состоит на 92% из крупных частиц площадью более 500 мкм2 со средним размером частиц ~40 мкм. Для механического обогащения предпочтительно использовать шлаки с медленной кристаллизацией, характеризующиеся крупными включениями титансодержащей фазы – перовскита. Размеры титансодержащей фазы и соотношение плотностей перовскита и акерманит-геленит (матрицы) на уровне 1,35 позволяют исследовать возможность механического выделения титансодержащего концентрата по плотности на концентрационном столе. The samples of the titanium-containing slag of EVRAZ NTMK JSC from the blast furnace process of concentrate of titanomagnetite ore the Ural Kachkanar deposit field in order to study the possibility of extracting titanium were investigated. The initial samples differ in crystallization method. One sample was obtained with a layer slag in a trench and cooling with water (hereinafter referred to as slags with fast crystallization), the second sample is slag crusts that remain in the cup after draining the liquid part (hereinafter referred to as the slags with slow crystallization). The chemical and phase compositions of all studied samples are similar the content of TiO2 is ~10.5%. The main phase (matrix) is akermanite-gehlenite, which is ~65%. The titanium-containing phase is perovskite ~26%. The fast crystallization slags are characterized by a fine structure of the titanium-containing phase with an average particle size of ~13 μm. The titaniumcontaining phase consists of 88% of particles of more than 100 μm2 with an average particle size of ~17 μm. A larger structure of the titanium-containing phase, the dimensions of such particles ~30 μm characterize the slow crystallization slags. The titaniumcontaining phase consists of 92% of large particles with an area of more than 500 μm2 with an average particle size of ~40 μm. For mechanical enrichment, it is preferable to use slags with slow crystallization, characterized by large inclusions of the titanium-containing phase - perovskite. The size of the titanium-containing phase and the ratio of perovskite density and akermanite-gehlenite (matrix) at the level of 1.35 make it possible to investigate the possibility of mechanical separation of the titanium-containing concentrate on the density on the concentrating table.

The results of a comprehensive study of the material microstructure of refractory samples after service in the furnace of a blast furnace with a useful volume of 2200 m3 for 14 years are presented. Structural and genetic analysis established a characteristic sequence of processes of degeneration and wear of the ceramic furnace stack, including 6 stages: the formation of micro-and macro-cracks; condensation of vaporous zinc; zinc oxidation with the formation of fire-resistant ZnO zincite and the deposition of carbon black by the; chemical interaction of zinkite with mullite and corundum to form ZnAl2O4 ganite and 2ZnO·SiO2 willemite; partial oxidation of silicon carbide with the release of silica glass SiO2, eutectic melt of complex composition, and wellimite. Due to intensive accumulation of capillary pores, infiltration of the slag melt in the volume of refractories does not occur. Wear of refractories in the furnace has a complex, mainly thermochemical mechanism and a low speed due to the formation of a garnish containing refractory compounds. Ill. 5. Ref. 14.

Applying learning-based control methods to real robots presents hard challenges, including the low sample efficiency of model-free reinforcement learning algorithms. The widely adopted approach to tackling this problem uses an environment dynamics model. We propose to use the Neural Ordinary Differential Equations to approximate transition dynamics as this allows for finer control of a trajectory generation process. NODE offers a continuous-time formulation that captures the temporal dependencies. We evaluate our approach on various tasks from simulation environment including learning 6-DoF robotic arm to open the door, which represents particular challenges for policy search. The NODE model is trained to predict movement of the arm and the door, and is used to generate trajectories for the model-based policy optimization. Our method shows better sample efficiency on this task comparing to the model-free and model-based baseline. It also shows comparable results on several other tasks. The application of NODE to model-based reinforcement learning enables more precise modeling of robotic system dynamics and enhances the sample efficiency of learning-based control methods. The empirical evaluation on various tasks demonstrates the efficacy of our approach, offering promising prospects for improving the performance and efficiency of real-world robotic systems.

The task of local trajectory planning for an autonomous wheeled robotic platform in cluttered indoor environment is considered. Such environment might include narrow passages, which width is less than the length of the platform. Therefore, it is not possible to apply standard approach, when the obstacles are inflated with the maximum radius of the platform. We propose a novel approach based on numerical solution of nonlinear model predictive control task. Oblong shape of the platform is approximated with a high-order ellipse. We define differentiable sigmoid po-tential function, which may be computed for any point of the workspace given position and orientation of the plat-form. This function is small far from the platform, and very high inside the platform; it increases when moving to-wards the robot. The value of this potential function is computed for the set of the support obstacle points and add-ed to the cost function. This function serve as a penalty for collision with obstacles or coming too close to them. We develop an algorithm for the mapping support points onto occupancy grid, which provide collision avoidance. We apply Acados open library, which implement numerical solution of nonlinear model predictive task with sequential quadratic programming. Our approach is implemented as a local planner for the collaborative mobile platform. The experiments were made in artificial maze and in real office environment with narrow passages. Proposed ap-proached allowed the robot to come through the passages that were 10-20 cm wider than the platform. Computa-tion time was around 20 milliseconds.

Results are presented for a comprehensive study of the chemical and material composition, as well as the macrostructure, of refractory samples after over 14 years service in a blast furnace hearth with a useful volume of 2200 m(3). Sample carbon content is determined using a comprehensive method. Zinc-containing phase formation is established in both the ceramic nozzle area oxide refractories and within the carbon lining of the furnace tuyere zone and hearth. Furnace refractory wear is complex, predominantly by a thermochemical mechanism, and a slow rate due to garnish formation containing refractory compounds.

Effective solving of manipulation tasks is significant for collaborative robots to act within human-oriented environments. It may be executed using classical or learning-based control methods. Classical methods are accurate; however, they require complicated tuning of the regulators. Learning-based control provide obtaining the parameters of the process model while training, but this model is rough. We apply a combined approach to solving manipulation tasks, where the robot moves to the target vicinity under learning-based control and then operates the target under simplified classical control. On the first stage, control system generates reference trajectory for execution using dynamic movement primitives (DMP). The parameters of the DMP are determined by output of a neural network and trained via policy optimization. On the second stage the forcing term of the DMP is set to zero, while goal is defined by the simplified predictive control model. We evaluate our approach on the tasks of reaching target point by end effector and pushing elevator button with UR5 collaborative manipulator. Evaluation is made in Isaac and URSim simulation taking in mind dynamics and functionality of the robot. The approach is successfully reproduced on a real robot.

We address the task of robotic door opening in office environments. This task is important for providing indoor mobility for collaborative mobile manipulators. In our work, we mainly focus on the use of high-level control opportunities and identification of the door parameters from visual and lidar data. We develop a solution, which includes handle recognition, handle twisting, and opening. The position of the handle is identified from stereo images by a neural-network-based method. We divide the opening procedure into two stages: first, handle twisting and slightly opening, and second, wide opening. The first stage is implemented via high-level task-space control of the robotic arm, while the platform is static. The position of the door axis is identified during the slight opening by fitting lidar data to the kinematic model. At the second stage, both the platform and the arm are active. The trajectory of the platform is defined by the model predictive planner in such a way that it avoids pushing the arm into a singular configuration, while the manipulator is operated via high-level impedance control. In our experiments, a mobile manipulator composed from the wheeled platform and the robotic arm was able to open office doors using the proposed approach.

The growing requirements for the quality of finished steel products oblige to obtain steel with a low content of harmful impurities, especially sulfur. The main charge for steel production in the conditions of JSC EVRAZ NTMK is vanadium iron, the external desulfurization of which leads to large losses of valuable vanadium. Therefore, the smelting of iron with a low sulfur content is an urgent task. The main way to reduce sulfur in iron is to increase the basicity of the slag (B2), but MgO is the best desulfurizer. The paper evaluates the effectiveness of the use of various magnesium-containing materials in the conditions of the blast-furnace stage of JSC EVRAZ NTMK, taking into account CO2 emissions. The possibility of replacing limestone in the iron flux charge with dolomite, as well as with steel converter slag, is considered. Both materials will increase the proportion of MgO in blast-furnace slag, however, the use of dolomite is accompanied by a two-fold increase in CO2 emissions compared to limestone. Converter steel slag (CSS) does not contain carbonates, thus eliminating the formation of CO2 during the limestone decomposition. An additional advantage of CSS is its high content of valuable vanadium and manganese. The effect of an increase in MgO in slag on its melting point is considered. It has been established that an increase in the proportion of MgO to 14% will not cause difficulties in the processing of blast-furnace smelting products and will reduce the sulfur content in iron and the formation of titanium carbides and carbonitrides.

Comparison of mining algorithms in the problem of detecting malicious network activity based on machine learning models is performed. A structural diagram of a system for analyzing network traffic in an industrial network based on machine learning methods has been developed. On one of the known datasets (CICIDS17), a series of experiments was carried out on preliminary analysis and preprocessing of features, highlighting the most significant features and building final models of classifiers. The f1-measure score for the committee of classifiers on the test sample is 0.967.

The complex of research works carried out by the NTMK team and the leading research teams of UIM, IMET UB RAS and Ural Federal University expanded the theoretical foundations of processing vanadium-containing titanomagnetites. The blast furnaces have achieved a specific productivity of over 3.0 t/m 3 , which is the best indicator in Russia. Specific fuel and carbon consumption is the best in Russia. The conditions for the operation of blast furnaces, in which the development of carbide formation processes is significantly reduced, have been established. Measures have been implemented that ensure the smelting of cast iron with a low content of silicon and titanium. The main measures are: maximum forcing of the furnace; increased pressure under the mouth; oxygen enrichment of the blast; as well as increasing the rate of filtration through the coke packing by optimizing the properties of the slag and increasing the hot strength of the coke. The introduction of blast-furnace smelting technology with the use of staflux and the injection of pulverized coal has made it possible to reduce the consumption of total carbon in the fuel to 418 kg/t of cast iron.