Robotics and Autonomous Systems最新文献_第3页

Novel prescribed performance control for a kind of back-support exoskeleton with time delays 一种具有时滞的背托式外骨骼的新型规定性能控制

IF 5.2 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Robotics and Autonomous Systems

Pub Date : 2026-01-08 DOI: 10.1016/j.robot.2026.105350

Xiaogang Li , Yuan Ke

Back-support exoskeletons (BSEs) offer significant potential for reducing lumbar load and preventing lower-back injuries; however, their control is complicated by strong inter-channel coupling, time delays in actuation and sensing, and uncertainties induced by human–robot interaction. This paper develops a unified dynamic–kinematic model for a multi-input multi-output BSE that explicitly accounts for coupling effects, compliance, and delay characteristics, providing a control-oriented and physically consistent foundation. Based on this model, a prescribed performance control (PPC) framework is proposed to guarantee bounded tracking errors with predefined transient and steady-state behaviour in the presence of multidimensional, time-varying delays, without requiring model decoupling. To enhance robustness against lumped uncertainties, a hybrid observer integrating a linear extended state observer and a sliding mode observer is designed for real-time disturbance estimation and compensation. Simulation results obtained on a biomechanically realistic BSE platform demonstrate that the proposed PPC–LESO–SMO scheme achieves superior tracking accuracy, robustness, and convergence speed compared with conventional PPC and existing observer-based control approaches.

背部支撑外骨骼（bse）提供显著的潜力，以减少腰椎负荷和防止腰背部损伤；然而，由于通道间的强耦合、驱动和传感的时间延迟以及人机交互引起的不确定性，它们的控制变得复杂。本文为多输入多输出BSE建立了一个统一的动态运动学模型，该模型明确地考虑了耦合效应、顺应性和延迟特性，为面向控制和物理一致提供了基础。在此模型的基础上，提出了一种规定性能控制（PPC）框架，以保证在存在多维时变延迟的情况下，具有预定义的瞬态和稳态行为的有界跟踪误差，而不需要模型解耦。为了增强对集总不确定性的鲁棒性，设计了一种集成线性扩展状态观测器和滑模观测器的混合观测器，用于实时干扰估计和补偿。仿真结果表明，与传统的PPC和现有的基于观测器的控制方法相比，所提出的PPC - leso - smo方案具有更好的跟踪精度、鲁棒性和收敛速度。

{"title":"Novel prescribed performance control for a kind of back-support exoskeleton with time delays","authors":"Xiaogang Li , Yuan Ke","doi":"10.1016/j.robot.2026.105350","DOIUrl":"10.1016/j.robot.2026.105350","url":null,"abstract":"<div><div>Back-support exoskeletons (BSEs) offer significant potential for reducing lumbar load and preventing lower-back injuries; however, their control is complicated by strong inter-channel coupling, time delays in actuation and sensing, and uncertainties induced by human–robot interaction. This paper develops a unified dynamic–kinematic model for a multi-input multi-output BSE that explicitly accounts for coupling effects, compliance, and delay characteristics, providing a control-oriented and physically consistent foundation. Based on this model, a prescribed performance control (PPC) framework is proposed to guarantee bounded tracking errors with predefined transient and steady-state behaviour in the presence of multidimensional, time-varying delays, without requiring model decoupling. To enhance robustness against lumped uncertainties, a hybrid observer integrating a linear extended state observer and a sliding mode observer is designed for real-time disturbance estimation and compensation. Simulation results obtained on a biomechanically realistic BSE platform demonstrate that the proposed PPC–LESO–SMO scheme achieves superior tracking accuracy, robustness, and convergence speed compared with conventional PPC and existing observer-based control approaches.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105350"},"PeriodicalIF":5.2,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978326","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

RoCeDiRNet-3DoF: Robust center direction network for cloth grasping point localization in complex scenes RoCeDiRNet-3DoF：基于鲁棒中心方向网络的复杂场景布料抓取点定位

IF 5.2 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Robotics and Autonomous Systems

Pub Date : 2026-01-08 DOI: 10.1016/j.robot.2026.105328

Jiaxiang Luo, Yufan Hu

Cloth grasping poses a fundamental challenge in robotics and computer vision, and constitutes a key capability for service robots. Accurate and robust localization of fabric grasping points is a critical prerequisite for enabling efficient and dexterous fabric manipulation. While recent deep learning approaches have achieved promising results on standard benchmarks, their performance still degrades substantially in complex scenarios involving severe folding, occlusion, and unknown backgrounds. This limitation primarily arises from insufficient global contextual modeling in existing dense regression frameworks, which rely heavily on local feature extraction while lacking the ability to capture global dependencies crucial for understanding fabric deformation patterns. To address these challenges, we propose RoCeDiRNet-3DoF, a novel framework built upon an InceptionNeXt encoder and a Wavelet Multiscale Convolutional Attention Decoder (WMCAD). WMCAD adopts a three-stage hierarchical architecture, incorporating Wavelet Convolution Blocks for global feature extraction, Dynamic Wavelet Upsampling Blocks to preserve semantic details during interpolation, and Multi-scale Mixed Attention Gates for effective cross-layer feature fusion. By leveraging WMCAD’s enhanced global feature modeling, RoCeDiRNet-3DoF achieves superior performance in challenging scenarios, achieving state-of-the-art (SOTA) results on the ViCoS dataset with an F1 score of 82.6%. Furthermore, across various complex scenario configurations, RoCeDiRNet-3DoF consistently outperforms competing methods, representing the current optimal solution for this task. The source code is available at: https://github.com/hyf381752569-stack/RoCeDiRNet-3DoF.

布料抓取是机器人技术和计算机视觉领域的一个基本挑战，也是服务机器人的一项关键能力。织物抓握点的准确和鲁棒定位是实现高效灵巧织物操作的关键前提。虽然最近的深度学习方法在标准基准上取得了可喜的结果，但在涉及严重折叠、遮挡和未知背景的复杂场景下，它们的性能仍然会大幅下降。这种限制主要源于现有密集回归框架中缺乏全局上下文建模，这些框架严重依赖于局部特征提取，而缺乏捕获全局依赖关系的能力，这对于理解织物变形模式至关重要。为了解决这些挑战，我们提出了RoCeDiRNet-3DoF，这是一个基于InceptionNeXt编码器和小波多尺度卷积注意力解码器（WMCAD）的新框架。WMCAD采用三阶段分层结构，采用小波卷积块进行全局特征提取，动态小波上采样块在插值过程中保留语义细节，多尺度混合注意门进行有效的跨层特征融合。通过利用WMCAD增强的全局特征建模，RoCeDiRNet-3DoF在具有挑战性的场景中取得了卓越的性能，在ViCoS数据集上获得了最先进的（SOTA）结果，F1得分为82.6%。此外，在各种复杂的场景配置中，RoCeDiRNet-3DoF始终优于竞争对手的方法，代表了该任务的当前最佳解决方案。源代码可从https://github.com/hyf381752569-stack/RoCeDiRNet-3DoF获得。

{"title":"RoCeDiRNet-3DoF: Robust center direction network for cloth grasping point localization in complex scenes","authors":"Jiaxiang Luo, Yufan Hu","doi":"10.1016/j.robot.2026.105328","DOIUrl":"10.1016/j.robot.2026.105328","url":null,"abstract":"<div><div>Cloth grasping poses a fundamental challenge in robotics and computer vision, and constitutes a key capability for service robots. Accurate and robust localization of fabric grasping points is a critical prerequisite for enabling efficient and dexterous fabric manipulation. While recent deep learning approaches have achieved promising results on standard benchmarks, their performance still degrades substantially in complex scenarios involving severe folding, occlusion, and unknown backgrounds. This limitation primarily arises from insufficient global contextual modeling in existing dense regression frameworks, which rely heavily on local feature extraction while lacking the ability to capture global dependencies crucial for understanding fabric deformation patterns. To address these challenges, we propose <strong>RoCeDiRNet-3DoF</strong>, a novel framework built upon an InceptionNeXt encoder and a <strong>Wavelet Multiscale Convolutional Attention Decoder (WMCAD)</strong>. WMCAD adopts a three-stage hierarchical architecture, incorporating Wavelet Convolution Blocks for global feature extraction, Dynamic Wavelet Upsampling Blocks to preserve semantic details during interpolation, and Multi-scale Mixed Attention Gates for effective cross-layer feature fusion. By leveraging WMCAD’s enhanced global feature modeling, RoCeDiRNet-3DoF achieves superior performance in challenging scenarios, achieving state-of-the-art (SOTA) results on the ViCoS dataset with an F1 score of 82.6%. Furthermore, across various complex scenario configurations, RoCeDiRNet-3DoF consistently outperforms competing methods, representing the current optimal solution for this task. The source code is available at: <span><span>https://github.com/hyf381752569-stack/RoCeDiRNet-3DoF</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105328"},"PeriodicalIF":5.2,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145940038","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Collision cone based time-efficient method for 3D collision avoidance for UAVs: A purely heading-based solution 基于碰撞锥的无人机三维避碰方法：一种纯粹基于航向的解决方案

IF 5.2 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Robotics and Autonomous Systems

Pub Date : 2026-01-07 DOI: 10.1016/j.robot.2026.105332

Manaram Gnanasekera, Jay Katupitiya

The increasing deployment of unmanned aerial vehicles (UAVs) across various fields, from agriculture to disaster management, has raised critical concerns about mid-air collisions in increasingly congested airspaces. While previous research has extensively explored collision avoidance techniques, most solutions focus either on static or low-density dynamic environments, leaving a gap in addressing UAV navigation in densely cluttered, dynamic 3D environments. This paper introduces a novel collision cone-based approach designed to enhance time-efficiency and precision in 3D UAV collision avoidance scenarios, particularly in complex and dynamic environments with multiple obstacles. Through both simulation and real-world experiments, the method demonstrates superior time-efficiency compared to a benchmark method, while maintaining robust performance in unpredictable environments. The contributions of this work include the development of a real-time adaptable algorithm that recalculates optimal paths based on dynamic changes and its practical validation in realistic, high-density scenarios. This work fills a significant research gap by addressing the limitations of previous 2D approaches and static obstacle methods, providing a comprehensive solution for UAVs operating in highly dynamic 3D spaces.

从农业到灾害管理等各个领域越来越多地部署无人驾驶飞行器（uav），这引起了人们对日益拥挤的空域中空中碰撞的严重担忧。虽然之前的研究已经广泛探索了避碰技术，但大多数解决方案都集中在静态或低密度动态环境上，这在解决密集混乱的动态3D环境下的无人机导航问题上留下了空白。本文介绍了一种新的基于碰撞锥的方法，旨在提高三维无人机避碰场景的时效性和精度，特别是在具有多个障碍物的复杂动态环境中。通过仿真和实际实验，与基准方法相比，该方法具有更高的时间效率，同时在不可预测的环境中保持稳健的性能。这项工作的贡献包括开发了一种基于动态变化重新计算最优路径的实时自适应算法，并在现实的高密度场景中进行了实际验证。这项工作通过解决以前的2D方法和静态障碍方法的局限性，填补了一个重要的研究空白，为无人机在高动态3D空间中运行提供了一个全面的解决方案。

{"title":"Collision cone based time-efficient method for 3D collision avoidance for UAVs: A purely heading-based solution","authors":"Manaram Gnanasekera, Jay Katupitiya","doi":"10.1016/j.robot.2026.105332","DOIUrl":"10.1016/j.robot.2026.105332","url":null,"abstract":"<div><div>The increasing deployment of unmanned aerial vehicles (UAVs) across various fields, from agriculture to disaster management, has raised critical concerns about mid-air collisions in increasingly congested airspaces. While previous research has extensively explored collision avoidance techniques, most solutions focus either on static or low-density dynamic environments, leaving a gap in addressing UAV navigation in densely cluttered, dynamic 3D environments. This paper introduces a novel collision cone-based approach designed to enhance time-efficiency and precision in 3D UAV collision avoidance scenarios, particularly in complex and dynamic environments with multiple obstacles. Through both simulation and real-world experiments, the method demonstrates superior time-efficiency compared to a benchmark method, while maintaining robust performance in unpredictable environments. The contributions of this work include the development of a real-time adaptable algorithm that recalculates optimal paths based on dynamic changes and its practical validation in realistic, high-density scenarios. This work fills a significant research gap by addressing the limitations of previous 2D approaches and static obstacle methods, providing a comprehensive solution for UAVs operating in highly dynamic 3D spaces.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105332"},"PeriodicalIF":5.2,"publicationDate":"2026-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978331","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Unified motion control of 4WIS-4WID WMR with unlimited steering and load transfer consideration 4wi - 4wid WMR的统一运动控制，具有无限转向和负载转移考虑

IF 5.2 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Robotics and Autonomous Systems

Pub Date : 2026-01-07 DOI: 10.1016/j.robot.2026.105341

Dongwoo Seo, Seokki Moon, Jaeyoung Kang

This paper presents a unified motion control framework for a four-wheel independent steering (4WIS) and four-wheel independent drive (4WID) wheeled mobile robot (WMR) equipped with an unlimited steering angle system. Unlike conventional methods that rely on mode-specific kinematic controllers, the proposed controller does not require prior classification of driving modes. The controller defines the necessary forces at the tire positions to track the desired velocity profile in the body-fixed frame, considering tire slip dynamics, load transfer effects, and friction constraints based on the Magic Formula. The steering angles and in-wheel motor torques are then determined to generate the required forces. Due to the difficulty of directly measuring tire forces, a disturbance observer (DOB) is used to estimate these forces in real-time. Simulation results demonstrate that the proposed controller outperforms conventional kinematics-based approaches in velocity tracking accuracy, while maintaining stable tire force distribution and vertical load limits, ensuring the robot’s stability and effective maneuverability even under highly dynamic conditions characterized by large lateral accelerations and the resulting tire slip and load transfer.

提出了一种具有无限转向角系统的四轮独立转向（4WIS）和四轮独立驱动（4WID）轮式移动机器人（WMR）的统一运动控制框架。与依赖于特定模式的运动学控制器的传统方法不同，所提出的控制器不需要事先对驱动模式进行分类。考虑轮胎滑移动力学、载荷传递效应和基于Magic Formula的摩擦约束，控制器定义轮胎位置所需的力以跟踪车身固定框架中所需的速度轮廓。然后确定转向角度和轮内电机扭矩以产生所需的力。由于难以直接测量轮胎受力，采用扰动观测器（DOB）实时估计轮胎受力。仿真结果表明，该控制器在速度跟踪精度方面优于传统的基于运动学的方法，同时保持稳定的轮胎力分布和垂直负载限制，即使在具有大横向加速度和由此导致的轮胎打滑和负载转移的高动态条件下，也能确保机器人的稳定性和有效的机动性。

{"title":"Unified motion control of 4WIS-4WID WMR with unlimited steering and load transfer consideration","authors":"Dongwoo Seo, Seokki Moon, Jaeyoung Kang","doi":"10.1016/j.robot.2026.105341","DOIUrl":"10.1016/j.robot.2026.105341","url":null,"abstract":"<div><div>This paper presents a unified motion control framework for a four-wheel independent steering (4WIS) and four-wheel independent drive (4WID) wheeled mobile robot (WMR) equipped with an unlimited steering angle system. Unlike conventional methods that rely on mode-specific kinematic controllers, the proposed controller does not require prior classification of driving modes. The controller defines the necessary forces at the tire positions to track the desired velocity profile in the body-fixed frame, considering tire slip dynamics, load transfer effects, and friction constraints based on the Magic Formula. The steering angles and in-wheel motor torques are then determined to generate the required forces. Due to the difficulty of directly measuring tire forces, a disturbance observer (DOB) is used to estimate these forces in real-time. Simulation results demonstrate that the proposed controller outperforms conventional kinematics-based approaches in velocity tracking accuracy, while maintaining stable tire force distribution and vertical load limits, ensuring the robot’s stability and effective maneuverability even under highly dynamic conditions characterized by large lateral accelerations and the resulting tire slip and load transfer.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105341"},"PeriodicalIF":5.2,"publicationDate":"2026-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978321","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Vision-guided gripping process with minimizing folding for flexible fabric materials by integrating a sequential optimization algorithm and FEM analysis 结合序列优化算法和有限元分析的视觉引导柔性织物材料最小折叠夹持过程

IF 5.2 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Robotics and Autonomous Systems

Pub Date : 2026-01-07 DOI: 10.1016/j.robot.2026.105349

Minh Khang Ngo , Chanhee Won , HyunKyo Lim , Dae Young Lim , Than Trong Khanh Dat , Jonghun Yoon

Automatic fabric handling in garment manufacturing presents significant challenges due to the soft, deformable, and highly variable nature of textile materials. This paper proposes an integrated robotic fabric gripping system capable of reliably identifying and manipulating fabric parts with minimal folding. The system comprises an industrial robotic arm, four needle grippers mounted on an adjustable jig mechanism, and a high-precision 3D vision camera for real-time fabric detection. A key contribution of this work is a sequential optimization model that determines four optimal gripping points for each fabric pattern, based on material characterization, deformation criteria, and folding definitions derived from experiments and finite element simulations. These points are mapped relative to CAD data and stored for retrieval. A vision-based matching algorithm then aligns real-time image inputs with CAD templates to localize the fabric piece and recover the precomputed optimal gripping points, which are transmitted to the robot for autonomous execution. Quantitative evaluations demonstrate that the proposed approach significantly reduces fabric folding and enhances the reliability of robotic garment handling, representing a substantial step toward fully automated garment production.

由于纺织材料的柔软，可变形和高度可变的性质，服装制造中的自动织物处理提出了重大挑战。提出了一种集成的机器人织物抓取系统，该系统能够以最小的折叠量可靠地识别和操纵织物零件。该系统包括一个工业机械臂、安装在可调夹具机构上的四个抓针器和一个用于实时织物检测的高精度3D视觉相机。这项工作的一个关键贡献是一个顺序优化模型，该模型基于材料特性、变形标准和从实验和有限元模拟中得出的折叠定义，确定每种织物图案的四个最佳夹紧点。这些点相对于CAD数据进行映射并存储以供检索。然后，基于视觉的匹配算法将实时图像输入与CAD模板对齐，以定位织物片并恢复预先计算的最佳抓取点，这些点将传输给机器人进行自主执行。定量评估表明，所提出的方法显著减少了织物折叠，提高了机器人服装处理的可靠性，代表着向全自动服装生产迈出了实质性的一步。

{"title":"Vision-guided gripping process with minimizing folding for flexible fabric materials by integrating a sequential optimization algorithm and FEM analysis","authors":"Minh Khang Ngo , Chanhee Won , HyunKyo Lim , Dae Young Lim , Than Trong Khanh Dat , Jonghun Yoon","doi":"10.1016/j.robot.2026.105349","DOIUrl":"10.1016/j.robot.2026.105349","url":null,"abstract":"<div><div>Automatic fabric handling in garment manufacturing presents significant challenges due to the soft, deformable, and highly variable nature of textile materials. This paper proposes an integrated robotic fabric gripping system capable of reliably identifying and manipulating fabric parts with minimal folding. The system comprises an industrial robotic arm, four needle grippers mounted on an adjustable jig mechanism, and a high-precision 3D vision camera for real-time fabric detection. A key contribution of this work is a sequential optimization model that determines four optimal gripping points for each fabric pattern, based on material characterization, deformation criteria, and folding definitions derived from experiments and finite element simulations. These points are mapped relative to CAD data and stored for retrieval. A vision-based matching algorithm then aligns real-time image inputs with CAD templates to localize the fabric piece and recover the precomputed optimal gripping points, which are transmitted to the robot for autonomous execution. Quantitative evaluations demonstrate that the proposed approach significantly reduces fabric folding and enhances the reliability of robotic garment handling, representing a substantial step toward fully automated garment production.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105349"},"PeriodicalIF":5.2,"publicationDate":"2026-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978333","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Path planning of bionic robotic fish based on enhanced SAC algorithm combined with HER and CQL 基于改进SAC算法结合HER和CQL的仿生机器鱼路径规划

IF 5.2 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Robotics and Autonomous Systems

Pub Date : 2026-01-06 DOI: 10.1016/j.robot.2026.105336

Ming Wang, Tinglong Zhao, Lingchen Zuo, Yanling Gong, Guangxin Lv, Ruilong Wang

Underwater robots play a crucial role in ocean exploration and resource development. As a type of underwater robot, bionic robotic fish can mimic the swimming patterns of real fish, offering unique advantages in complex underwater environments. However, traditional path planning methods perform poorly in unknown or dynamic environments. This paper proposes a path planning algorithm based on deep reinforcement learning, which improves the path planning capabilities of bionic robotic fish in unknown environments by combining the Soft Actor-Critic (SAC) algorithm with the Hindsight Experience Replay (HER) mechanism. Furthermore, the use of the Conservative Q-Learning (CQL) algorithm enhances the algorithm’s exploration capability and robustness. The effectiveness and practicality of the proposed algorithm are validated through simulation results.

水下机器人在海洋勘探和资源开发中发挥着至关重要的作用。仿生机器鱼作为一种水下机器人，能够模仿真鱼的游动模式，在复杂的水下环境中具有独特的优势。然而，传统的路径规划方法在未知或动态环境中表现不佳。本文提出了一种基于深度强化学习的路径规划算法，通过将软Actor-Critic （SAC）算法与后见之明经验回放（HER）机制相结合，提高了仿生机器鱼在未知环境下的路径规划能力。此外，使用保守Q-Learning （CQL）算法增强了算法的探索能力和鲁棒性。仿真结果验证了该算法的有效性和实用性。

引用次数: 0

A self-supervised exploratory guided push and grasp approach to enhance robotic manipulation in clutter 一种自监督探索性引导推抓方法提高机器人在杂波环境下的操作能力

IF 5.2 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Robotics and Autonomous Systems

Pub Date : 2026-01-06 DOI: 10.1016/j.robot.2026.105339

Chiat Pin Tay , Shijun Yan , Chong Chen , Wei Qi Toh , Miaolong Yuan , Dongkyu Choi

Robot manipulation of cluttered and stacked objects remains a crucial yet challenging task with immense real-world applications. Integrating push actions into grasp strategies has been proposed to address this challenge. However, existing deep learning approaches, particularly those based on reinforcement learning, often suffer from lengthy training times and inefficient action selection. This paper introduces a novel exploratory guided push approach that leverages self-supervised learning and a memory buffer enhancement strategy to accelerate efficient data sampling for model training. We investigated and analysed various network architectures, data collection methods, and learning strategies to understand their impacts on push-and-grasp tasks. Extensive experiments conducted in both simulated and real environments demonstrate the effectiveness of our proposed solution.

机器人操纵杂乱和堆叠的物体仍然是一个具有巨大现实应用的关键但具有挑战性的任务。为了应对这一挑战，建议将推动行动整合到把握策略中。然而，现有的深度学习方法，尤其是那些基于强化学习的方法，往往存在训练时间长、动作选择效率低的问题。本文介绍了一种新的探索性引导推送方法，该方法利用自监督学习和记忆缓冲增强策略来加速模型训练的有效数据采样。我们调查并分析了各种网络架构、数据收集方法和学习策略，以了解它们对推抓任务的影响。在模拟和真实环境中进行的大量实验证明了我们提出的解决方案的有效性。

引用次数: 0

Enhancing sampling-based planning with a library of paths 使用路径库增强基于抽样的规划

IF 5.2 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Robotics and Autonomous Systems

Pub Date : 2026-01-05 DOI: 10.1016/j.robot.2026.105334

Michal Minařík, Vojtěch Vonásek, Robert Pěnička

Path planning for 3D solid objects is a challenging problem, requiring a search in a six-dimensional configuration space, which is, nevertheless, essential in many robotic applications such as bin-picking and assembly. The commonly used sampling-based planners, such as Rapidly-exploring Random Trees, struggle with narrow passages where the sampling probability is low, increasing the time needed to find a solution. In scenarios like robotic bin-picking, various objects must be transported through the same environment. However, traditional planners start from scratch each time, losing valuable information gained during the planning process. We address this by using a library of past solutions, allowing the reuse of previous experiences even when planning for a new, previously unseen object. Paths for a set of objects are stored, and when planning for a new object, we find the most similar one in the library and use its paths as approximate solutions, adjusting for possible mutual transformations. The configuration space is then sampled along the approximate paths. Our method is tested in various narrow passage scenarios and compared with state-of-the-art methods from the OMPL library. Results show significant speed improvements (up to

85 %

decrease in the required time) of our method, often finding a solution in cases where the other planners fail. Our implementation of the proposed method is released as an open-source package.

三维实体物体的路径规划是一个具有挑战性的问题，需要在六维构型空间中进行搜索，然而，这在许多机器人应用中是必不可少的，例如拾取和组装。常用的基于抽样的计划，如快速探索随机树，在抽样概率低的狭窄通道中挣扎，增加了找到解决方案所需的时间。在像机器人拾取垃圾箱这样的场景中，不同的物体必须通过相同的环境进行运输。然而，传统的规划者每次都从零开始，失去了在规划过程中获得的宝贵信息。我们通过使用过去解决方案的库来解决这个问题，即使在规划新的，以前未见过的对象时，也可以重用以前的经验。存储一组对象的路径，当规划一个新对象时，我们在库中找到最相似的对象，并使用其路径作为近似解，调整可能的相互转换。然后沿着近似路径对构型空间进行采样。我们的方法在各种狭窄通道场景中进行了测试，并与OMPL库中最先进的方法进行了比较。结果表明，我们的方法显著提高了速度（所需时间减少了85%），通常在其他计划器失败的情况下找到解决方案。我们提出的方法的实现作为一个开源包发布。

{"title":"Enhancing sampling-based planning with a library of paths","authors":"Michal Minařík, Vojtěch Vonásek, Robert Pěnička","doi":"10.1016/j.robot.2026.105334","DOIUrl":"10.1016/j.robot.2026.105334","url":null,"abstract":"<div><div>Path planning for 3D solid objects is a challenging problem, requiring a search in a six-dimensional configuration space, which is, nevertheless, essential in many robotic applications such as bin-picking and assembly. The commonly used sampling-based planners, such as Rapidly-exploring Random Trees, struggle with narrow passages where the sampling probability is low, increasing the time needed to find a solution. In scenarios like robotic bin-picking, various objects must be transported through the same environment. However, traditional planners start from scratch each time, losing valuable information gained during the planning process. We address this by using a library of past solutions, allowing the reuse of previous experiences even when planning for a new, previously unseen object. Paths for a set of objects are stored, and when planning for a new object, we find the most similar one in the library and use its paths as approximate solutions, adjusting for possible mutual transformations. The configuration space is then sampled along the approximate paths. Our method is tested in various narrow passage scenarios and compared with state-of-the-art methods from the OMPL library. Results show significant speed improvements (up to <span><math><mrow><mn>85</mn><mtext>%</mtext></mrow></math></span> decrease in the required time) of our method, often finding a solution in cases where the other planners fail. Our implementation of the proposed method is released as an open-source package.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105334"},"PeriodicalIF":5.2,"publicationDate":"2026-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145940041","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Reinforcement learning in robotic systems : A review on sim-to-real transfer 机器人系统中的强化学习：模拟到真实迁移研究综述

IF 5.2 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Robotics and Autonomous Systems

Pub Date : 2026-01-05 DOI: 10.1016/j.robot.2025.105327

Rajesh Tiwari, Shailesh Khapre, Avantika Singh

Sim-to-real transfer reinforcement learning has become a pivotal approach for narrowing the gap between simulation environments and real-world robotic applications. While reinforcement learning methods achieve remarkable success in simulation, their direct deployment in real environments remains challenging due to the inherent reality gap caused by mismatches in physical dynamics, sensory inputs, and environmental variability. This paper presents a comprehensive review of recent advances in sim-to-real transfer, emphasizing improved simulation fidelity, actuator-level modeling, and domain randomization encompassing both environmental and robotic parameters. A unified framework is proposed that outlines key processes, including simulation model optimization, strategy transfer, and iterative policy refinement across virtual and real domains. The study also discusses emerging developments such as Progressive Neural Networks for policy migration and modern benchmarking platforms. The review concludes with open challenges and future directions, including automated simulator tuning, adaptive domain adaptation, and establishing theoretical guarantees for robust and generalizable transfer learning.

模拟到真实的迁移强化学习已经成为缩小模拟环境和现实世界机器人应用之间差距的关键方法。虽然强化学习方法在模拟中取得了显著的成功，但由于物理动力学、感官输入和环境可变性不匹配导致的固有现实差距，它们在真实环境中的直接部署仍然具有挑战性。本文全面回顾了模拟到真实转移的最新进展，强调了改进的仿真保真度，执行器级建模以及包含环境和机器人参数的域随机化。提出了一个统一的框架，概述了关键过程，包括仿真模型优化、策略转移和跨虚拟和现实领域的迭代策略优化。该研究还讨论了新兴的发展，如用于政策迁移的渐进式神经网络和现代基准平台。最后提出了未来的挑战和发展方向，包括自动模拟器调优、自适应领域适应以及建立鲁棒和可推广迁移学习的理论保证。

{"title":"Reinforcement learning in robotic systems : A review on sim-to-real transfer","authors":"Rajesh Tiwari, Shailesh Khapre, Avantika Singh","doi":"10.1016/j.robot.2025.105327","DOIUrl":"10.1016/j.robot.2025.105327","url":null,"abstract":"<div><div>Sim-to-real transfer reinforcement learning has become a pivotal approach for narrowing the gap between simulation environments and real-world robotic applications. While reinforcement learning methods achieve remarkable success in simulation, their direct deployment in real environments remains challenging due to the inherent reality gap caused by mismatches in physical dynamics, sensory inputs, and environmental variability. This paper presents a comprehensive review of recent advances in sim-to-real transfer, emphasizing improved simulation fidelity, actuator-level modeling, and domain randomization encompassing both environmental and robotic parameters. A unified framework is proposed that outlines key processes, including simulation model optimization, strategy transfer, and iterative policy refinement across virtual and real domains. The study also discusses emerging developments such as Progressive Neural Networks for policy migration and modern benchmarking platforms. The review concludes with open challenges and future directions, including automated simulator tuning, adaptive domain adaptation, and establishing theoretical guarantees for robust and generalizable transfer learning.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105327"},"PeriodicalIF":5.2,"publicationDate":"2026-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978324","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

System identification and model reference adaptive control of bipedal locomotion with neural networks 基于神经网络的双足运动系统辨识与模型参考自适应控制

IF 5.2 2区计算机科学 Q1 AUTOMATION & CONTROL SYSTEMS

Robotics and Autonomous Systems

Pub Date : 2026-01-05 DOI: 10.1016/j.robot.2026.105331

Burak Çatalbaş , Bahadır Çatalbaş , Ömer Morgül

Biped robots have enormous potential to transform our lives with the usability of the infrastructure already prepared for humans, thanks to our morphological similarities. Unfortunately, the realization of this potential requires solutions to inherited difficulties of bipedal locomotion dynamics. Controlling these systems poses challenges due to their high degree of freedom, small support polygon, hybrid dynamics, etc. Nervous system ensures the locomotion of biological counterparts of biped robots. Similarly, artificial neural networks show competence to control bipedal locomotion. Unfortunately, complexity of neural network-based controllers (NNBCs) makes it difficult to adapt them to changes in robot model dynamics. In this paper, we propose a model reference adaptive control algorithm for bipedal locomotion, together with neural networks consisting of recurrent and feedforward layers in the controller and system identification tasks to overcome this drawback. Controller weights are updated via error gradient calculated through system identification neural network to force the controlled system output to desired behavior in an iterative manner. Moreover, we examine the effectiveness of our algorithm in decreasing the speed of steady-state error of neural controllers under different simulated scenarios. It is shown that recurrent and feedforward layers are beneficial for walking control and system identification with neural networks and implementable for real-time applications on various Jetson single-board computers. Results suggest that our method is capable of adapting neural controllers without requiring training from scratch. Under different scenarios, up to 33.6%, 34%, 31.8% in the target speed tracking error decrease is acquired with our algorithm on training, validation, and test sets.

由于我们在形态上的相似性，两足机器人有巨大的潜力来改变我们的生活，利用已经为人类准备好的基础设施。不幸的是，实现这种潜力需要解决两足运动动力学的遗传困难。由于这些系统具有高自由度、小支撑多边形、混合动力学等特点，对其控制提出了挑战。神经系统保证了双足机器人的生物对应物的运动。同样，人工神经网络显示出控制两足运动的能力。然而，基于神经网络的控制器（nnbc）的复杂性使其难以适应机器人模型动力学的变化。在本文中，我们提出了一种模型参考自适应控制算法，并在控制器和系统识别任务中使用由循环层和前馈层组成的神经网络来克服这一缺点。通过系统辨识神经网络计算误差梯度来更新控制器权值，以迭代的方式迫使被控系统输出达到期望行为。此外，我们还检验了该算法在不同仿真场景下降低神经控制器稳态误差速度的有效性。结果表明，循环层和前馈层有利于神经网络的行走控制和系统辨识，并可在各种Jetson单板计算机上实现实时应用。结果表明，我们的方法能够适应神经控制器，而不需要从头开始训练。在不同的场景下，我们的算法在训练集、验证集和测试集上的目标速度跟踪误差分别降低了33.6%、34%和31.8%。

{"title":"System identification and model reference adaptive control of bipedal locomotion with neural networks","authors":"Burak Çatalbaş , Bahadır Çatalbaş , Ömer Morgül","doi":"10.1016/j.robot.2026.105331","DOIUrl":"10.1016/j.robot.2026.105331","url":null,"abstract":"<div><div>Biped robots have enormous potential to transform our lives with the usability of the infrastructure already prepared for humans, thanks to our morphological similarities. Unfortunately, the realization of this potential requires solutions to inherited difficulties of bipedal locomotion dynamics. Controlling these systems poses challenges due to their high degree of freedom, small support polygon, hybrid dynamics, etc. Nervous system ensures the locomotion of biological counterparts of biped robots. Similarly, artificial neural networks show competence to control bipedal locomotion. Unfortunately, complexity of neural network-based controllers (NNBCs) makes it difficult to adapt them to changes in robot model dynamics. In this paper, we propose a model reference adaptive control algorithm for bipedal locomotion, together with neural networks consisting of recurrent and feedforward layers in the controller and system identification tasks to overcome this drawback. Controller weights are updated via error gradient calculated through system identification neural network to force the controlled system output to desired behavior in an iterative manner. Moreover, we examine the effectiveness of our algorithm in decreasing the speed of steady-state error of neural controllers under different simulated scenarios. It is shown that recurrent and feedforward layers are beneficial for walking control and system identification with neural networks and implementable for real-time applications on various Jetson single-board computers. Results suggest that our method is capable of adapting neural controllers without requiring training from scratch. Under different scenarios, up to 33.6%, 34%, 31.8% in the target speed tracking error decrease is acquired with our algorithm on training, validation, and test sets.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105331"},"PeriodicalIF":5.2,"publicationDate":"2026-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978327","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0