Optical Memory and Neural Networks最新文献

Monocular RGB Metric-Semantic SLAM with 3D Reconstruction–Based Odometry 基于三维重构的单目RGB度量语义SLAM

IF 0.9

Optical Memory and Neural Networks Pub Date : 2026-10-16 DOI: 10.3103/S1060992X2660357X

I. Podmarev, N. Menshikov, O. Bulichev

{"title":"Monocular RGB Metric-Semantic SLAM with 3D Reconstruction–Based Odometry","authors":"I. Podmarev, N. Menshikov, O. Bulichev","doi":"10.3103/S1060992X2660357X","DOIUrl":"10.3103/S1060992X2660357X","url":null,"abstract":"Incorporating semantic scene information has been shown to improve the performance of classical SLAM algorithms. While recent advances in 3D scene reconstruction enable the generation of dense point clouds using only an RGB camera, without relying on additional sensors. Motivated by this, we propose DA3-Khronos, a monocular RGB metric-semantic SLAM pipeline that combines Depth Anything 3 (DA3), a neural 3D reconstruction model for depth and camera pose estimation, with Khronos, a spatio-temporal metric-semantic SLAM system for scene-graph construction and global map optimization. The initial vision-inertial odometry module is replaced with a new learning-based solution, eliminating the need for an inertial measurement unit (IMU) and depth camera. Dense point clouds generated from the 3D reconstruction are then processed with SLAM system, resulting information rich map. Our architecture enables metric-semantic SLAM using only a monocular RGB camera, providing an efficient and sensor-light solution for detailed scene understanding. This approach demonstrates competitive results with state-of-the-art RGB methods. Code is available at https://github.com/IakovPodmarev/da3-khronos.","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 editor—Dmitry","pages":"S101 - S111"},"PeriodicalIF":0.9,"publicationDate":"2026-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176635","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

VICONav: Vectorized Isaac Configurable Object Navigation Benchmark Vectorized Isaac可配置对象导航基准

IF 0.9

Optical Memory and Neural Networks Pub Date : 2026-10-16 DOI: 10.3103/S1060992X26603738

N. Oskolkov, D. Makarov

{"title":"VICONav: Vectorized Isaac Configurable Object Navigation Benchmark","authors":"N. Oskolkov, D. Makarov","doi":"10.3103/S1060992X26603738","DOIUrl":"10.3103/S1060992X26603738","url":null,"abstract":"This paper introduces VICONav, a vectorized configurable object-navigation benchmark for mobile robots in Isaac Lab. The benchmark is built on Isaac Sim and uses USD (Universal Scene Description) assets, physically simulated robot motion, and rendered indoor scenes executed in parallel environment instances. All instances share a common potential object pool, while each reset independently samples the target object, object arrangement, obstacle configuration, and initial robot state. The tensor shapes and object indices remain fixed across episodes, which supports efficient vectorized training and direct comparison of policies under a stable observation interface. Scene generation is specified by a two-level configuration scheme. The first level defines object classes, physical parameters, USD assets, and semantic object tags. The second level defines compositional scene blocks that represent functional zones and object groups. This design separates the object vocabulary from the spatial organization of the scene and allows controlled variation of object composition, target distribution, graph construction, observation modality, and action parameterization. VICONav supports discrete and continuous control, visual and graph-based observations, curriculum-dependent initial-state distributions, and simulator-state access for reward computation and auxiliary training objectives. We evaluate the benchmark through a series of experiments covering learned navigation policies, classical reference methods, and robustness tests under controlled visual and graph perturbations. The results show that VICONav can be used as a configuration-driven experimental substrate for comparing object-goal navigation approaches under controlled changes of scene structure, observation composition, and control interface.","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 editor—Dmitry","pages":"S112 - S125"},"PeriodicalIF":0.9,"publicationDate":"2026-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176739","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Sensor Placement Optimization for Cooperative Active Sensing in Multi-Robot Systems 多机器人系统中协同主动传感传感器布局优化

IF 0.9

Optical Memory and Neural Networks Pub Date : 2026-10-16 DOI: 10.3103/S1060992X26603519

M. Sofin, N. Menshikov, O. Bulichev

{"title":"Sensor Placement Optimization for Cooperative Active Sensing in Multi-Robot Systems","authors":"M. Sofin, N. Menshikov, O. Bulichev","doi":"10.3103/S1060992X26603519","DOIUrl":"10.3103/S1060992X26603519","url":null,"abstract":"Effective perception in multi-robot systems depends both on the physical placement of sensors and on their active control during operation. However, existing approaches typically address sensor placement and active sensing separately, neglecting the dependency between hardware configuration and sensing strategies. In this work, we propose a two-stage framework for cooperative active sensing in robot teams equipped with multiple steerable sensors. In the first stage, sensor placement and base orientations are optimized using the Non-dominated Sorting Genetic Algorithm II (NSGA-II) multi-objective genetic algorithm, balancing field-of-view overlap and angular sensing density under platform constraints. In the second stage, cooperative sensor control is formulated as a reinforcement learning problem. Centralized policies trained with three algorithms—Proximal Policy Optimization (PPO), Advantage Actor-Critic (A2C), and Soft Actor-Critic (SAC)—learn to rotate sensors within bounded ranges around the optimized orientations to maximize environmental coverage while maintaining visual observation of teammates, with SAC achieving the highest reward improvement of up to 11.5%. Simulation experiments with teams of three robots show that the proposed approach significantly improves sensing coverage compared to static configurations, demonstrating the benefit of jointly considering sensor placement and cooperative active sensing.","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 editor—Dmitry","pages":"S59 - S70"},"PeriodicalIF":0.9,"publicationDate":"2026-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176634","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

PixelNav: Towards Model-Based Vision-Only Navigation with Topological Graphs PixelNav：迈向基于模型的视觉导航与拓扑图

IF 0.9

Optical Memory and Neural Networks Pub Date : 2026-10-16 DOI: 10.3103/S1060992X26604082

S. Bakulin, T. Akhtyamov, D. Fatykhov, G. Devchich, Gonzalo Ferrer

引用次数: 0

Enhancing Trajectory Planning with Vision-Language-Action Models and Perception Guidance 用视觉-语言-行动模型和感知引导加强轨迹规划

IF 0.9

Optical Memory and Neural Networks Pub Date : 2026-10-16 DOI: 10.3103/S1060992X26603854

L. Ivanov, D. Yudin

{"title":"Enhancing Trajectory Planning with Vision-Language-Action Models and Perception Guidance","authors":"L. Ivanov, D. Yudin","doi":"10.3103/S1060992X26603854","DOIUrl":"10.3103/S1060992X26603854","url":null,"abstract":"Vision-Language-Action (VLA) models are emerging as a flexible interface for autonomous driving, but directly mapping camera observations to future trajectories requires a general-purpose Vision-Language Model (VLM) to implicitly solve perception, motion prediction, and planning at once. This can limit spatial grounding and safety in metric trajectory generation. We propose PG-VLA (Perception-Guided VLA), a planner that keeps task-specific perception and motion prediction outside the VLM while using their outputs as structured planning context. PG-VLA receives a front-view camera image together with detected agents, predicted agent motion, ego-state, historical ego trajectory, and a high-level navigation command, and generates future ego waypoints as vehicle actions autoregressively. We instantiate PG-VLA with an open-weight Qwen3-VL-2B VLM backbone and fine-tune it using supervised learning on expert trajectories from the nuScenes dataset. Under the ST-P3 and UniAD open-loop planning evaluation benchmarks, PG-VLA achieves competitive L2 trajectory error and substantially lower collision rates than the GPT-Driver baseline. Compared with recent VLA planners, it obtains strong safety-oriented performance without reinforcement fine-tuning, explicit reasoning supervision, or dedicated bird’s-eye-view (BEV) token projection modules. An open-source method implementation is available at https://github.com/LIvanoff/PG-VLA.","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 editor—Dmitry","pages":"S139 - S149"},"PeriodicalIF":0.9,"publicationDate":"2026-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176696","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Odometry-Guided Mobile Robot Homing with Neural Slip Compensation in Complex Dynamic Environments 复杂动态环境下基于神经滑移补偿的里程制导移动机器人寻的研究

IF 0.9

Optical Memory and Neural Networks Pub Date : 2026-10-16 DOI: 10.3103/S1060992X2670013X

A. V. Kozyr, A. A. Barsukova

{"title":"Odometry-Guided Mobile Robot Homing with Neural Slip Compensation in Complex Dynamic Environments","authors":"A. V. Kozyr, A. A. Barsukova","doi":"10.3103/S1060992X2670013X","DOIUrl":"10.3103/S1060992X2670013X","url":null,"abstract":"In the absence of global navigation signals and loss of communication, ensuring the autonomous return of a mobile robot to the starting point becomes a critically important task. In this work, an algorithm is proposed for the autonomous return of a differential-drive mobile robot to the starting point with the ability to bypass dynamic obstacles, functioning exclusively based on wheel rotation odometry. For the formation of the return trajectory, a multi-level processing of accumulated odometric data is implemented, including preliminary filtering and intelligent segmentation of the trajectory into homogeneous segments (straight movement, turns, movement at constant speed) using the GRASP-UTS method. Based on the identified segments, the resulting displacement vector is calculated, the inversion of which allows constructing a return vector field along the shortest path. Navigation and obstacle avoidance are carried out based on the potential fields method, while the distance to objects is measured using lidar. The main problem with using odometric path calculation in real conditions is the progressive error accumulated when integrating encoder signals. To compensate for this, a neural network algorithm for wheel slip detection was developed. The LSTM network was trained on a specialized dataset [1], collected at a test site with a motion capture system (MoCap). The developed algorithm and control system are implemented in the ROS 2 environment. The dataset includes numerous runs over surfaces of various types with precise alignment of odometric and reference trajectories. Experimental validation confirmed the high accuracy and stability of the developed homing algorithm in a complex dynamic environment, providing reliable operation solely based on wheel odometry.","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 editor—Dmitry","pages":"S71 - S86"},"PeriodicalIF":0.9,"publicationDate":"2026-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176738","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

CoT Oracle: Dominant Gait Parameters Identification for Cost of Transport Prediction in Learned Quadrupedal Locomotion Policies 学习四足运动策略中运输成本预测的优势步态参数识别

IF 0.9

Optical Memory and Neural Networks Pub Date : 2026-10-16 DOI: 10.3103/S1060992X26603477

A. Tsaritsin, Ammar Issa, Anubhav Singh, S. Kolyubin

{"title":"CoT Oracle: Dominant Gait Parameters Identification for Cost of Transport Prediction in Learned Quadrupedal Locomotion Policies","authors":"A. Tsaritsin, Ammar Issa, Anubhav Singh, S. Kolyubin","doi":"10.3103/S1060992X26603477","DOIUrl":"10.3103/S1060992X26603477","url":null,"abstract":"Energy efficiency is a central objective in the design of reinforcement learning (RL)-based controllers for legged robots. Yet, the influence of individual gait parameters on energy consumption in learned locomotion policies remains underexplored. This paper presents a systematic sensitivity analysis of the cost of transport (CoT) across an 11-dimensional gait-parameter space for a learned quadrupedal locomotion policy deployed on the Unitree AlienGo platform. Gait commands are varied over predefined ranges, and the resulting CoT is evaluated at multiple locomotion velocities. To ensure experimental validity, we assess the policy’s ability to reliably track the commanded gait parameters. Our analysis reveals that only a subset of five parameters exerts a dominant influence on CoT, with the effect strongly modulated by locomotion speed, while the remaining six parameters contribute negligibly within the examined operating envelope. Furthermore, we show that a data-driven surrogate model–specifically, a multilayer perceptron–can accurately predict CoT directly from commanded gait parameters and velocity. These findings offer practical guidance for reducing the dimensionality of energy-aware gait optimization and enable rapid, model-based screening of efficient locomotion strategies in learned quadruped controllers.","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 editor—Dmitry","pages":"S1 - S14"},"PeriodicalIF":0.9,"publicationDate":"2026-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176647","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Comparative Analysis of Baseline and Filter-Guided Hybrid Architectures for Non-Stationary Noise Suppression 用于非平稳噪声抑制的基线和滤波制导混合架构的比较分析

IF 0.9

Optical Memory and Neural Networks Pub Date : 2026-10-16 DOI: 10.3103/S1060992X26603507

D. Dudolin, A. Nosov, K. Mukhin, Y. Kalinin

{"title":"Comparative Analysis of Baseline and Filter-Guided Hybrid Architectures for Non-Stationary Noise Suppression","authors":"D. Dudolin, A. Nosov, K. Mukhin, Y. Kalinin","doi":"10.3103/S1060992X26603507","DOIUrl":"10.3103/S1060992X26603507","url":null,"abstract":"Low-light imaging conditions introduce signal-dependent Poisson noise that degrades structural detail and compromises the reliability of downstream computer vision algorithms. The present work investigates resource-efficient denoising strategies for preserving morphological image features under non-stationary noise regimes. A class of architectures designated Filter-Guided Hybrid Cascades (FGHC) is proposed and evaluated, in which lightweight deterministic filters supply structural priors to compact neural restoration models. The study systematically compares classical filtering approaches, standalone neural baselines, and four hybrid integration strategies, namely Early Fusion, Residual Addition, Spatial Attention, and Spatial Feature Transform, with respect to restoration fidelity and computational overhead. Experiments conducted on a fluorescence microscopy denoising benchmark demonstrate that the strongest hybrid cascade configurations improve restoration quality relative to isolated baselines. The best-performing overall configuration, consisting of an Anisotropic Diffusion Zhang filter coupled with NAFNet under Spatial Feature Transform modulation, achieves a PSNR gain of 5.46 dB and an SSIM of 0.941. For deployment under resource-constrained conditions, the configuration combining a Gradient-Guided Bilateral Filter with Noise2Detail under Spatial Feature Transform modulation attains the most favorable efficiency trade-off, delivering a PSNR gain of 5.32 dB, an SSIM of 0.938, a throughput of 10.6 frames per second, and a VRAM footprint of 79.4 MB. The reported results indicate that the incorporation of deterministic structural priors into lightweight neural denoising pipelines improves restoration performance while preserving the latency and memory characteristics required for edge and robotic perception deployment.","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 editor—Dmitry","pages":"S42 - S58"},"PeriodicalIF":0.9,"publicationDate":"2026-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176695","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

SGN-CIRL: Scene Graph-Based Navigation with Curriculum, Imitation, and Reinforcement Learning SGN-CIRL：基于场景图的导航课程、模仿和强化学习

IF 0.9

Optical Memory and Neural Networks Pub Date : 2026-10-16 DOI: 10.3103/S1060992X26603556

N. Oskolkov, D. Yudin, D. Makarov, A. Panov, Huzhenyu Zhang

引用次数: 0

Hybrid Neural Network-Based Real Time Inverse Kinematics Method for a Six-Axis Anthropomorphic Leg without a Spherical Joint 基于混合神经网络的无球面六轴拟人腿实时运动学逆解方法

IF 0.9

Optical Memory and Neural Networks Pub Date : 2026-10-16 DOI: 10.3103/S1060992X26603520

Dotti Pablo, N. Poberezkin

{"title":"Hybrid Neural Network-Based Real Time Inverse Kinematics Method for a Six-Axis Anthropomorphic Leg without a Spherical Joint","authors":"Dotti Pablo, N. Poberezkin","doi":"10.3103/S1060992X26603520","DOIUrl":"10.3103/S1060992X26603520","url":null,"abstract":"In humanoid robot control, the inverse kinematics solver constitutes a critical real-time component, as any latency may compromise stable operation. This study presents the development and evaluation of two hybrid methods for solving the real-time inverse kinematics of a six-axis anthropomorphic leg lacking a spherical joint. The absence of this joint makes closed-form analytical solutions not possible, necessitating fast and accurate numerical approaches for real-time applications. To accelerate convergence and reduce the overall number of required iterations, we propose two distinct strategies for generating an initial guess for a Damped Least Squares numerical solver. The first is a novel geometry-based method designed for the specific kinematic constraints and reachable workspace of the leg structure. The second employs a custom neural network trained on forward kinematics data. Comprehensive testing was conducted to evaluate convergence behavior based on iteration count. Execution time measurements obtained directly on a microcontroller showed that from the proposed methods, the quickest one takes an average of 3.24 ms of calculation time with a maximum of 3.9 ms. This validates the real-time viability of both proposed approaches, demonstrating a 37% increase in computational speed compared to established alternatives when considering the platform on which they were calculated.","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 editor—Dmitry","pages":"S15 - S28"},"PeriodicalIF":0.9,"publicationDate":"2026-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149176693","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0