Robotics and Autonomous Systems最新文献_第10页

A hierarchical planning framework for self-assembly and self-reconfiguration of autonomous modular space robots 自主模块化空间机器人自组装与自重构的分层规划框架

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

Robotics and Autonomous Systems

Pub Date : 2026-03-01 Epub Date: 2025-12-03 DOI: 10.1016/j.robot.2025.105273

Zhen Huang , Yajie Cheng , Rongqiang Liu , Minghe Shan

Owing to the extreme conditions of the space environment and the difficulty of external intervention, autonomous modular robotic systems offer a viable solution for constructing structures in space through self-assembly and self-reconfiguration. This paper presents a comprehensive planning system designed for modular robot self-assembly and self-reconfiguration tasks in space. The proposed system consists of two hierarchical layers: a sequential path planning layer and a task and trajectory planning layer. At the sequential path planning layer, the self-assembly and self-reconfiguration tasks are formulated as time-varying online Multi-Agent Path Finding (MAPF) problems, and are effectively solved using the enhanced Time-Expanded Network (TEN) and Minimum-Cost Maximum-Flow (MCMF) algorithm to generate collision-free and efficient sequential paths. At the task and trajectory planning layer, robot tasks are automatically partitioned based on the connection status and motion characteristics, with corresponding dynamic models generated adaptively according to the current task state. A hybrid iterative Linear Quadratic Regulator (iLQR) algorithm is introduced to achieve rapid response and trajectory optimality while satisfying constraints on the joint angles, actuator torques, and interface forces. Simulations using the HexaFlipBot modular robot platform confirmed that the proposed planning approach can efficiently and accurately complete the construction and reconfiguration of typical structures in space.

由于空间环境的极端条件和外界干预的困难，自主模块化机器人系统通过自组装和自重构为空间结构的构建提供了可行的解决方案。针对模块化机器人在空间中的自组装和自重构任务，提出了一种综合规划系统。该系统由两个层次组成：顺序路径规划层和任务与轨迹规划层。在序列路径规划层，将自组装和自重构任务表述为时变在线多智能体寻路（MAPF）问题，并利用增强型时间扩展网络（TEN）和最小成本最大流量（MCMF）算法有效地求解，生成无碰撞、高效的序列路径。在任务和轨迹规划层，根据连接状态和运动特征自动划分机器人任务，并根据当前任务状态自适应生成相应的动态模型。提出了一种混合迭代线性二次型调节器（iLQR）算法，在满足关节角度、执行机构扭矩和界面力约束的情况下，实现了快速响应和轨迹优化。利用HexaFlipBot模块化机器人平台进行的仿真验证了所提出的规划方法能够高效、准确地完成空间中典型结构的构造和重构。

{"title":"A hierarchical planning framework for self-assembly and self-reconfiguration of autonomous modular space robots","authors":"Zhen Huang , Yajie Cheng , Rongqiang Liu , Minghe Shan","doi":"10.1016/j.robot.2025.105273","DOIUrl":"10.1016/j.robot.2025.105273","url":null,"abstract":"<div><div>Owing to the extreme conditions of the space environment and the difficulty of external intervention, autonomous modular robotic systems offer a viable solution for constructing structures in space through self-assembly and self-reconfiguration. This paper presents a comprehensive planning system designed for modular robot self-assembly and self-reconfiguration tasks in space. The proposed system consists of two hierarchical layers: a sequential path planning layer and a task and trajectory planning layer. At the sequential path planning layer, the self-assembly and self-reconfiguration tasks are formulated as time-varying online Multi-Agent Path Finding (MAPF) problems, and are effectively solved using the enhanced Time-Expanded Network (TEN) and Minimum-Cost Maximum-Flow (MCMF) algorithm to generate collision-free and efficient sequential paths. At the task and trajectory planning layer, robot tasks are automatically partitioned based on the connection status and motion characteristics, with corresponding dynamic models generated adaptively according to the current task state. A hybrid iterative Linear Quadratic Regulator (iLQR) algorithm is introduced to achieve rapid response and trajectory optimality while satisfying constraints on the joint angles, actuator torques, and interface forces. Simulations using the HexaFlipBot modular robot platform confirmed that the proposed planning approach can efficiently and accurately complete the construction and reconfiguration of typical structures in space.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"197 ","pages":"Article 105273"},"PeriodicalIF":5.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145685203","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

A generic task model and control strategy to support learning, robust control, and generalization of contact-rich manipulation tasks 一个通用的任务模型和控制策略，以支持学习、鲁棒控制和丰富的接触操作任务的泛化

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

Robotics and Autonomous Systems

Pub Date : 2026-03-01 Epub Date: 2025-12-01 DOI: 10.1016/j.robot.2025.105270

Ali Mousavi Mohammadi, Maxim Vochten, Erwin Aertbeliën, Joris De Schutter

Learning and controlling contact-rich tasks are based on a task model that is built from motion and wrench data collected from demonstration. In previous work, the procedure for choosing a proper reference frame to express the learned signals is often implicit, requires expert insight, or starts from proposed frame candidates. Furthermore, the exact location of the geometric features in the world or tool that determine this frame may be uncertain during task demonstration or execution. This article presents an automatic method to derive the optimal reference frame for the task model directly from recorded motion and wrench data. This frame maximizes decoupling between the learned signals, enabling the design of a decoupled task controller. The method is founded on screw theory, entirely probabilistic and requires no hyperparameters. The method is generic as it works regardless of whether the task involves translation, rotation, force, or moment, or any combination thereof. While other control approaches could be used, we propose a generic, decoupled, constraint-based and velocity-resolved controller to work in tandem with the derived task model. It involves a small number of task-dependent control parameters that need to be tuned. The derivation of the task model, the robustness of the controller, and the task generalization capabilities of their combination were validated for various tasks involving contact between rigid objects, including surface following and manipulation of articulated objects. The results show a good agreement between derived and assumed expert task frames as well as the effectiveness and versatility of the overall approach.

学习和控制丰富的接触任务是基于一个任务模型，该模型是由运动和扳手数据从演示中收集而来。在以前的工作中，选择合适的参考框架来表达学习到的信号的过程通常是隐含的，需要专家的洞察力，或者从提出的候选框架开始。此外，在任务演示或执行期间，确定该框架的几何特征在世界或工具中的确切位置可能是不确定的。本文提出了一种直接从记录的运动和扳手数据中自动导出任务模型最优参考系的方法。该框架最大限度地解耦了学习到的信号，使解耦任务控制器的设计成为可能。该方法建立在螺旋理论的基础上，完全是概率性的，不需要超参数。该方法是通用的，因为无论任务是否涉及平移、旋转、力或力矩，或它们的任何组合，它都可以工作。虽然可以使用其他控制方法，但我们提出了一种通用的、解耦的、基于约束的和速度分辨的控制器来与派生的任务模型协同工作。它涉及少量需要调优的任务相关控制参数。任务模型的推导、控制器的鲁棒性以及它们组合的任务泛化能力在涉及刚性物体接触的各种任务中得到了验证，包括表面跟踪和铰接物体的操作。结果表明，导出的专家任务框架和假设的专家任务框架之间有很好的一致性，以及整体方法的有效性和通用性。

{"title":"A generic task model and control strategy to support learning, robust control, and generalization of contact-rich manipulation tasks","authors":"Ali Mousavi Mohammadi, Maxim Vochten, Erwin Aertbeliën, Joris De Schutter","doi":"10.1016/j.robot.2025.105270","DOIUrl":"10.1016/j.robot.2025.105270","url":null,"abstract":"<div><div>Learning and controlling contact-rich tasks are based on a task model that is built from motion and wrench data collected from demonstration. In previous work, the procedure for choosing a proper reference frame to express the learned signals is often implicit, requires expert insight, or starts from proposed frame candidates. Furthermore, the exact location of the geometric features in the world or tool that determine this frame may be uncertain during task demonstration or execution. This article presents an automatic method to derive the optimal reference frame for the task model directly from recorded motion and wrench data. This frame maximizes decoupling between the learned signals, enabling the design of a decoupled task controller. The method is founded on screw theory, entirely probabilistic and requires no hyperparameters. The method is generic as it works regardless of whether the task involves translation, rotation, force, or moment, or any combination thereof. While other control approaches could be used, we propose a generic, decoupled, constraint-based and velocity-resolved controller to work in tandem with the derived task model. It involves a small number of task-dependent control parameters that need to be tuned. The derivation of the task model, the robustness of the controller, and the task generalization capabilities of their combination were validated for various tasks involving contact between rigid objects, including surface following and manipulation of articulated objects. The results show a good agreement between derived and assumed expert task frames as well as the effectiveness and versatility of the overall approach.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"197 ","pages":"Article 105270"},"PeriodicalIF":5.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145685204","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

Deterministic delay-aware reinforcement learning 确定性延迟感知强化学习

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

Robotics and Autonomous Systems

Pub Date : 2026-03-01 Epub Date: 2025-12-02 DOI: 10.1016/j.robot.2025.105271

Sathira Dilshan Bataduwaarachchi , Zoran Najdovski , Hieu Trinh , Chee Peng Lim , Van Thanh Huynh

Reinforcement Learning (RL) has effectively paved the way in achieving robotic control during the past decade. As a result, the avenue of integrating RL-powered robotic control and teleoperation has caught the attention of researchers. Every RL framework involves the basis of suitable observation and action communication between the environment and the agent, and the involvement of teleoperation can introduce random time delays within the said communication process. Achieving robotic control under such constraints remains an untapped area in the domain of reinforcement learning. We take the initiative to achieve the goal of robotic control while handling delays in the RL setting based on a fitting Markov Decision Process (MDP) structure. Our algorithm will learn a deterministic policy and can tackle control environments, especially robotic manipulation environments, using observations with proprioceptive information. We methodically present the theoretical adjustments based on an existing dominant off-policy algorithm to express the algorithm’s competency with proof of convergence. We perform experimentations with DeepMind Control Suite, illustrating significant results showing the algorithm’s capabilities in learning complex environments powered by delay-aware RL.

在过去的十年中，强化学习（RL）有效地为实现机器人控制铺平了道路。因此，将rl驱动的机器人控制和远程操作相结合的途径引起了研究人员的注意。每个强化学习框架都涉及到环境与智能体之间适当的观察和行动通信的基础，遥操作的参与会在上述通信过程中引入随机的时间延迟。在这种约束下实现机器人控制仍然是强化学习领域的一个未开发的领域。我们主动实现机器人控制的目标，同时处理基于拟合马尔可夫决策过程（MDP）结构的RL设置中的延迟。我们的算法将学习一个确定性策略，并可以处理控制环境，特别是机器人操作环境，使用本体感知信息的观察。我们系统地提出了基于现有的优势离策略算法的理论调整，以表达算法的收敛性证明。我们使用DeepMind Control Suite进行了实验，结果显示了该算法在学习由延迟感知强化学习驱动的复杂环境中的能力。

{"title":"Deterministic delay-aware reinforcement learning","authors":"Sathira Dilshan Bataduwaarachchi , Zoran Najdovski , Hieu Trinh , Chee Peng Lim , Van Thanh Huynh","doi":"10.1016/j.robot.2025.105271","DOIUrl":"10.1016/j.robot.2025.105271","url":null,"abstract":"<div><div>Reinforcement Learning (RL) has effectively paved the way in achieving robotic control during the past decade. As a result, the avenue of integrating RL-powered robotic control and teleoperation has caught the attention of researchers. Every RL framework involves the basis of suitable observation and action communication between the environment and the agent, and the involvement of teleoperation can introduce random time delays within the said communication process. Achieving robotic control under such constraints remains an untapped area in the domain of reinforcement learning. We take the initiative to achieve the goal of robotic control while handling delays in the RL setting based on a fitting Markov Decision Process (MDP) structure. Our algorithm will learn a deterministic policy and can tackle control environments, especially robotic manipulation environments, using observations with proprioceptive information. We methodically present the theoretical adjustments based on an existing dominant off-policy algorithm to express the algorithm’s competency with proof of convergence. We perform experimentations with DeepMind Control Suite, illustrating significant results showing the algorithm’s capabilities in learning complex environments powered by delay-aware RL.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"197 ","pages":"Article 105271"},"PeriodicalIF":5.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145738062","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

A Multi-UAV unfixed scale task assignment method with uncertainty based on deep Q network 基于深度Q网络的不确定多无人机不固定尺度任务分配方法

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

Robotics and Autonomous Systems

Pub Date : 2026-03-01 Epub Date: 2025-12-02 DOI: 10.1016/j.robot.2025.105269

Wenqi Ding, Jia Song, Jingwei Yu

Task assignment is a fundamental aspect in optimizing the efficiency of multiple unmanned aerial vehicle (Multi-UAV) systems. This paper presents a data-driven approach for addressing the Multi-UAV unfixed scale task assignment problem (MUSTA) in the presence of uncertain conditions. Firstly, we propose a task allocation model that incorporates uncertainty, thereby simulating the penetration process that may result in losses for unmanned aerial vehicles, and comprehensively evaluating the algorithm’s effectiveness. Secondly, we refine the task allocation strategy in network processing to effectively handle scenarios with variable-scale configurations of both UAVs and targets. By devising appropriate allocation rules, a Q-network can be trained to approximate allocation strategies suitable for diverse UAV-task scales, eliminating the need for network structure reinitialization and training for variable scale problems. Leveraging the characteristics of the allocation problem under study, we design a state–action space and a reward function, and simulation results confirm that this approach not only efficiently provides relatively optimal solutions for different large-scale problems but also acquires allocation strategies that closely resemble real-world application scenarios through training in uncertain environments.

任务分配是优化多无人机系统效率的一个基本方面。提出了一种数据驱动的方法来解决不确定条件下多无人机不固定尺度任务分配问题。首先，提出了一种考虑不确定性的任务分配模型，模拟了无人机突防可能造成损失的过程，并对算法的有效性进行了综合评价。其次，我们改进了网络处理中的任务分配策略，以有效地处理无人机和目标都具有变尺度配置的场景。通过设计适当的分配规则，q -网络可以训练出适合不同无人机任务规模的近似分配策略，从而消除了网络结构重新初始化和对变规模问题的训练。利用所研究分配问题的特点，设计了状态-动作空间和奖励函数，仿真结果表明，该方法不仅能有效地为不同大规模问题提供相对最优的解决方案，而且通过不确定环境下的训练，获得了与现实应用场景非常相似的分配策略。

{"title":"A Multi-UAV unfixed scale task assignment method with uncertainty based on deep Q network","authors":"Wenqi Ding, Jia Song, Jingwei Yu","doi":"10.1016/j.robot.2025.105269","DOIUrl":"10.1016/j.robot.2025.105269","url":null,"abstract":"<div><div>Task assignment is a fundamental aspect in optimizing the efficiency of multiple unmanned aerial vehicle (Multi-UAV) systems. This paper presents a data-driven approach for addressing the Multi-UAV unfixed scale task assignment problem (MUSTA) in the presence of uncertain conditions. Firstly, we propose a task allocation model that incorporates uncertainty, thereby simulating the penetration process that may result in losses for unmanned aerial vehicles, and comprehensively evaluating the algorithm’s effectiveness. Secondly, we refine the task allocation strategy in network processing to effectively handle scenarios with variable-scale configurations of both UAVs and targets. By devising appropriate allocation rules, a Q-network can be trained to approximate allocation strategies suitable for diverse UAV-task scales, eliminating the need for network structure reinitialization and training for variable scale problems. Leveraging the characteristics of the allocation problem under study, we design a state–action space and a reward function, and simulation results confirm that this approach not only efficiently provides relatively optimal solutions for different large-scale problems but also acquires allocation strategies that closely resemble real-world application scenarios through training in uncertain environments.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"197 ","pages":"Article 105269"},"PeriodicalIF":5.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145738060","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

Adaptive path planning method for multi-AUVs in complex environments 复杂环境下多auv自适应路径规划方法

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

Robotics and Autonomous Systems

Pub Date : 2026-03-01 Epub Date: 2025-12-08 DOI: 10.1016/j.robot.2025.105290

Jianhao Wu , Chang Liu , Vladimir Filaretov , Dmitry Yukhimets

In modern marine scientific exploration, environmental protection, and resource investigation missions, autonomous underwater vehicles (AUV) have seen increasingly widespread applications. However, complex underwater environments present challenges such as multi-objective optimization conflicts and communication constraints, particularly in scenarios with high-density obstacles and dynamic ocean current disturbances. This paper proposes a Distributed Learning-Enhanced Double Deep Q-Network (DLE-DDQN) algorithm for adaptive path planning, integrating deep reinforcement learning with distributed collaborative optimization. The algorithm employs an improved Double Deep Q-Network (DDQN) architecture and a dynamically adaptive reward function. These components guide AUVs in path planning within unknown environments. Meanwhile, distributed learning mechanisms are leveraged to enhance training efficiency under limited communication bandwidth. Experimental results demonstrate that DLE-DDQN achieves an 18.7% improvement in task success rate, a 29.6% increase in path efficiency, and a 51.4% reduction in response time compared to traditional DDQN and Multi-Agent Deep Deterministic Policy Gradient (MADDPG) methods in high-obstacle-density scenarios. Through a cross-attention mechanism that fuses environmental perception with navigational states, the algorithm maintains path curvature stability with a standard deviation below 0.15 rad/m even under abrupt current disturbances. These results validate the algorithm’s robustness and real-time performance in unstructured dynamic environments, confirming its capability to overcome both static obstacles and current interference for rapid, precise path planning.

在现代海洋科学探测、环境保护和资源调查任务中，自主水下航行器（AUV）得到了越来越广泛的应用。然而，复杂的水下环境面临着多目标优化冲突和通信约束等挑战，特别是在高密度障碍物和动态洋流干扰的情况下。将深度强化学习与分布式协同优化相结合，提出了一种分布式学习增强双深度Q-Network （DLE-DDQN）自适应路径规划算法。该算法采用改进的双深度q网络（DDQN）结构和动态自适应的奖励函数。这些组件指导auv在未知环境中进行路径规划。同时，利用分布式学习机制，在有限的通信带宽下提高训练效率。实验结果表明，在高障碍密度场景下，与传统的DDQN和Multi-Agent Deep Deterministic Policy Gradient （MADDPG）方法相比，DLE-DDQN的任务成功率提高了18.7%，路径效率提高了29.6%，响应时间缩短了51.4%。通过融合环境感知和导航状态的交叉注意机制，即使在突发电流干扰下，该算法也能保持低于0.15 rad/m的路径曲率稳定性。这些结果验证了该算法在非结构化动态环境中的鲁棒性和实时性，证实了该算法能够克服静态障碍物和电流干扰，实现快速、精确的路径规划。

{"title":"Adaptive path planning method for multi-AUVs in complex environments","authors":"Jianhao Wu , Chang Liu , Vladimir Filaretov , Dmitry Yukhimets","doi":"10.1016/j.robot.2025.105290","DOIUrl":"10.1016/j.robot.2025.105290","url":null,"abstract":"<div><div>In modern marine scientific exploration, environmental protection, and resource investigation missions, autonomous underwater vehicles (AUV) have seen increasingly widespread applications. However, complex underwater environments present challenges such as multi-objective optimization conflicts and communication constraints, particularly in scenarios with high-density obstacles and dynamic ocean current disturbances. This paper proposes a Distributed Learning-Enhanced Double Deep Q-Network (DLE-DDQN) algorithm for adaptive path planning, integrating deep reinforcement learning with distributed collaborative optimization. The algorithm employs an improved Double Deep Q-Network (DDQN) architecture and a dynamically adaptive reward function. These components guide AUVs in path planning within unknown environments. Meanwhile, distributed learning mechanisms are leveraged to enhance training efficiency under limited communication bandwidth. Experimental results demonstrate that DLE-DDQN achieves an 18.7% improvement in task success rate, a 29.6% increase in path efficiency, and a 51.4% reduction in response time compared to traditional DDQN and Multi-Agent Deep Deterministic Policy Gradient (MADDPG) methods in high-obstacle-density scenarios. Through a cross-attention mechanism that fuses environmental perception with navigational states, the algorithm maintains path curvature stability with a standard deviation below 0.15 rad/m even under abrupt current disturbances. These results validate the algorithm’s robustness and real-time performance in unstructured dynamic environments, confirming its capability to overcome both static obstacles and current interference for rapid, precise path planning.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"197 ","pages":"Article 105290"},"PeriodicalIF":5.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145737592","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

A sufficient condition for feature agnostic video stabilization for surveillance robots: A state space approach 监控机器人特征不可知视频稳定的充分条件：一种状态空间方法

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

Robotics and Autonomous Systems

Pub Date : 2026-03-01 Epub Date: 2025-12-03 DOI: 10.1016/j.robot.2025.105275

Adwaith Vijayakumar, Abhishek Gupta, Leena Vachhani

A camera is a ubiquitous sensor in surveillance robots. Often the camera is subjected to unintended motion from either external factors or inherent dynamics. The intended camera motion induces constant flow vectors for short segments of a video sequence. Video stabilization is the removal of the effect of unintended camera motion from a video sequence. The surveillance robots are often made to operate indoors, where features are scarce for motion estimation rendering traditional video stabilization methods unreliable. Unlike feature/motion extraction methods, investigation in this work is to explore the possibility of stabilizing video while remaining agnostic to the scene. One such possibility is stabilizing the video by manipulating the coefficients of a basis of the signal subspace, the smallest subspace containing the video. We show a sufficient condition for which stabilization can be performed in the signal subspace of an unstabilized video. The sufficient condition is derived by developing a novel state-space model that describes a short-length video sequence as a discrete-time state-space model. The properties of matrices corresponding to the state model are analyzed, and the corresponding results are utilized in deriving the sufficient condition. Simulation videos and experimental videos collected from a UAV validate the proposed sufficient condition.

摄像头是监控机器人中无处不在的传感器。通常，相机会受到来自外部因素或内在动力的意外运动。预期的摄像机运动为视频序列的短片段诱导恒定的流矢量。视频稳定是从视频序列中去除非预期的摄像机运动的影响。监控机器人通常在室内工作，在室内运动估计的特征很少，使得传统的视频稳定方法不可靠。与特征/运动提取方法不同，这项工作的研究是探索稳定视频的可能性，同时保持对场景的不可知论。其中一种可能性是通过操纵信号子空间（包含视频的最小子空间）的基的系数来稳定视频。给出了在不稳定视频的信号子空间中实现稳定的一个充分条件。通过开发一种新的状态空间模型，将短视频序列描述为离散时间状态空间模型，推导出充分条件。分析了状态模型所对应的矩阵的性质，并利用相应的结果推导了状态模型的充分条件。从无人机上采集的仿真视频和实验视频验证了所提出的充分条件。

{"title":"A sufficient condition for feature agnostic video stabilization for surveillance robots: A state space approach","authors":"Adwaith Vijayakumar, Abhishek Gupta, Leena Vachhani","doi":"10.1016/j.robot.2025.105275","DOIUrl":"10.1016/j.robot.2025.105275","url":null,"abstract":"<div><div>A camera is a ubiquitous sensor in surveillance robots. Often the camera is subjected to unintended motion from either external factors or inherent dynamics. The intended camera motion induces constant flow vectors for short segments of a video sequence. Video stabilization is the removal of the effect of unintended camera motion from a video sequence. The surveillance robots are often made to operate indoors, where features are scarce for motion estimation rendering traditional video stabilization methods unreliable. Unlike feature/motion extraction methods, investigation in this work is to explore the possibility of stabilizing video while remaining agnostic to the scene. One such possibility is stabilizing the video by manipulating the coefficients of a basis of the signal subspace, the smallest subspace containing the video. We show a sufficient condition for which stabilization can be performed in the signal subspace of an unstabilized video. The sufficient condition is derived by developing a novel state-space model that describes a short-length video sequence as a discrete-time state-space model. The properties of matrices corresponding to the state model are analyzed, and the corresponding results are utilized in deriving the sufficient condition. Simulation videos and experimental videos collected from a UAV validate the proposed sufficient condition.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"197 ","pages":"Article 105275"},"PeriodicalIF":5.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145685197","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

Prior-Backbone-Prediction cascade depth completion framework for robotic grasp oriented to transparent objects 面向透明目标的机器人抓取先验-主干预测级联深度补全框架

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

Robotics and Autonomous Systems

Pub Date : 2026-03-01 Epub Date: 2025-11-28 DOI: 10.1016/j.robot.2025.105268

Hongkun Tian , Shuo Liu , Shuai Liu , Jun Li , Ling Tong , Yi Man

Transparent objects are frequent manipulation objects for industrial production and manufacturing systems. The texture-less characteristics of transparent objects lead to insignificant RGB image features and incomplete and inaccurate Depth images, which significantly challenge robotic grasping detection and potentially dangerous for independent and safe manipulation in situations such as industrial production and manufacturing system. Reconstruction of depth data by fusing RGB-Depth (RGB-D) data through depth completion is the mainstream solution as a pre-processing robot-acquired data. Therefore, the quality, as well as the efficiency of depth completion, determines the effectiveness of the robot grasping transparent objects. However, the no-prior processing used in existing studies leads to an unimpressive depth completion. In addition, existing studies employ one-way prediction results and do not address the issue of image distortion. In general, current studies are not ideal in balancing accuracy and efficiency. To this end, we propose a Prior-Backbone-Prediction (PBP) cascade lightweight depth completion framework, named PBP framework. Prior: the proposed Reverse Texture Prior Guidance strategy guides the backbone network to focus on texture-less regions better. Backbone: the proposed Pyramid Interactive Residual Backbone network is used for in-depth feature learning. Prediction: the proposed consistency-difference Joint Prediction strategy maximizes the accuracy of prediction results and reduces distortion. The dual validation is performed on both public datasets and robotic platforms. PBP achieves state-of-the-art performance on the well-known TransCG dataset with a model size of only 17.2 M. The success rate of the Aubo i5 robot platform is improved by 44 % based on GG-CNN to grasp transparent objects.

透明对象是工业生产和制造系统中常用的操作对象。透明物体的无纹理特征导致RGB图像特征不明显，深度图像不完整和不准确，这给机器人抓取检测带来了极大的挑战，并对工业生产和制造系统等情况下的独立安全操作带来了潜在的危险。通过深度完井融合RGB-Depth （RGB-D）数据重建深度数据是机器人采集数据预处理的主流解决方案。因此，深度完成的质量和效率决定了机器人抓取透明物体的有效性。然而，现有研究中使用的无先验处理导致深度完成效果不佳。此外，现有的研究采用单向预测结果，没有解决图像失真的问题。总的来说，目前的研究在平衡准确性和效率方面并不理想。为此，我们提出了一个Prior-Backbone-Prediction （PBP）级联轻量级深度完井框架，命名为PBP框架。先验：提出的反向纹理先验指导策略可以更好地引导骨干网关注无纹理的区域。骨干网：提出的金字塔交互残差骨干网用于深度特征学习。预测：提出的一致性-差异联合预测策略使预测结果的准确性最大化，减少了失真。双重验证在公共数据集和机器人平台上进行。PBP在著名的TransCG数据集上达到了最先进的性能，模型大小仅为17.2 M.基于GG-CNN的Aubo i5机器人平台抓取透明物体的成功率提高了44%。

{"title":"Prior-Backbone-Prediction cascade depth completion framework for robotic grasp oriented to transparent objects","authors":"Hongkun Tian , Shuo Liu , Shuai Liu , Jun Li , Ling Tong , Yi Man","doi":"10.1016/j.robot.2025.105268","DOIUrl":"10.1016/j.robot.2025.105268","url":null,"abstract":"<div><div>Transparent objects are frequent manipulation objects for industrial production and manufacturing systems. The texture-less characteristics of transparent objects lead to insignificant RGB image features and incomplete and inaccurate Depth images, which significantly challenge robotic grasping detection and potentially dangerous for independent and safe manipulation in situations such as industrial production and manufacturing system. Reconstruction of depth data by fusing RGB-Depth (RGB-D) data through depth completion is the mainstream solution as a pre-processing robot-acquired data. Therefore, the quality, as well as the efficiency of depth completion, determines the effectiveness of the robot grasping transparent objects. However, the no-prior processing used in existing studies leads to an unimpressive depth completion. In addition, existing studies employ one-way prediction results and do not address the issue of image distortion. In general, current studies are not ideal in balancing accuracy and efficiency. To this end, we propose a Prior-Backbone-Prediction (PBP) cascade lightweight depth completion framework, named PBP framework. <strong>Prior:</strong> the proposed Reverse Texture Prior Guidance strategy guides the backbone network to focus on texture-less regions better. <strong>Backbone:</strong> the proposed Pyramid Interactive Residual Backbone network is used for in-depth feature learning. <strong>Prediction:</strong> the proposed consistency-difference Joint Prediction strategy maximizes the accuracy of prediction results and reduces distortion. The dual validation is performed on both public datasets and robotic platforms. PBP achieves state-of-the-art performance on the well-known TransCG dataset with a model size of only 17.2 M. The success rate of the Aubo i5 robot platform is improved by 44 % based on GG-CNN to grasp transparent objects.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"197 ","pages":"Article 105268"},"PeriodicalIF":5.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145685201","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

Physics-informed LSTM-based delay compensation framework for high-fidelity teleoperation of wheeled mobile robots 基于物理信息lstm的轮式移动机器人高保真遥操作延迟补偿框架

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

Robotics and Autonomous Systems

Pub Date : 2026-03-01 Epub Date: 2025-11-26 DOI: 10.1016/j.robot.2025.105265

Ahmad Abubakar , Yahya Zweiri , Abdel Gafoor Haddad , Mubarak Yakubu , Ruqayya Alhammadi , Lakmal Seneviratne

Bilateral teleoperation of wheeled mobile robots (WMRs) traversing soft terrains with interaction force feedback is crucial for applications like lunar exploration, as it help regulates travel reduction and path deviation. However, network-induced communication delays present a challenge in achieving high-fidelity closed-loop integration, degrading slippage awareness and command-tracking performance. To address this, a physics-informed long short-term memory (PiLSTM)-based predictor framework is proposed to predicts intended commands and anticipated slippage feedback in WMR teleoperation systems under large delays. Conventional model-free predictors often demonstrate significant prediction errors when compensating for large delays in complex dynamics systems due to their limited adaptability and generalizability. In contrast, the proposed PiLSTM predictor integrates an LSTM network with time-delay dynamics and a feedback mechanism to better capture strong nonlinear and temporal dynamics for improved adaptability and generalizability while ensuring accurate prediction. A series of human-in-the-loop experiments for a track-following task were conducted to validate the effectiveness of the PiLSTM framework in compensating for the large delays, similar to those encountered in lunar exploration. Its performance was compared against first-order time-delay (FOD) and wave variable transformation (WV) approaches for delay compensation. The results demonstrate the robustness of the PiLSTM framework across various test scenarios, achieving up to 29.7% improvement in prediction accuracy compared to the FOD framework. Furthermore, the proposed PiLSTM-based teleoperation system achieved significantly improved tracking performance and force transparency compared to FOD- and WV-based systems, ensuring higher fidelity and more robust closed-loop integration. These findings highlight the PiLSTM framework’s adaptability and generalizability for complex teleoperation systems. A supplementary video is available at https://youtu.be/dMVLEV7qTvE.

轮式移动机器人（WMRs）在软地形上的双向遥操作与交互力反馈对于月球探测等应用至关重要，因为它有助于调节行程减少和路径偏差。然而，网络引起的通信延迟对实现高保真闭环集成提出了挑战，降低了滑动感知和命令跟踪性能。为了解决这个问题，提出了一种基于物理信息的长短期记忆（PiLSTM）预测框架，用于预测大延迟下WMR远程操作系统的预期命令和预期滑动反馈。传统的无模型预测器由于其有限的适应性和泛化性，在补偿复杂动力学系统的大延迟时往往表现出显著的预测误差。相比之下，本文提出的PiLSTM预测器集成了具有时滞动力学和反馈机制的LSTM网络，可以更好地捕获强非线性和时间动力学，从而在保证准确预测的同时提高适应性和泛化性。为了验证PiLSTM框架在补偿大延迟（类似于月球探测中遇到的延迟）方面的有效性，进行了一系列轨道跟踪任务的人在环实验。将其性能与一阶时延（FOD）和波动变量变换（WV）方法进行了比较。结果证明了PiLSTM框架在各种测试场景中的鲁棒性，与FOD框架相比，预测精度提高了29.7%。此外，与基于FOD和wv的系统相比，所提出的基于pilstm的远程操作系统实现了显著改善的跟踪性能和力透明度，确保了更高的保真度和更强大的闭环集成。这些发现突出了PiLSTM框架对复杂远程操作系统的适应性和通用性。补充视频可在https://youtu.be/dMVLEV7qTvE上获得。

{"title":"Physics-informed LSTM-based delay compensation framework for high-fidelity teleoperation of wheeled mobile robots","authors":"Ahmad Abubakar , Yahya Zweiri , Abdel Gafoor Haddad , Mubarak Yakubu , Ruqayya Alhammadi , Lakmal Seneviratne","doi":"10.1016/j.robot.2025.105265","DOIUrl":"10.1016/j.robot.2025.105265","url":null,"abstract":"<div><div>Bilateral teleoperation of wheeled mobile robots (WMRs) traversing soft terrains with interaction force feedback is crucial for applications like lunar exploration, as it help regulates travel reduction and path deviation. However, network-induced communication delays present a challenge in achieving high-fidelity closed-loop integration, degrading slippage awareness and command-tracking performance. To address this, a physics-informed long short-term memory (PiLSTM)-based predictor framework is proposed to predicts intended commands and anticipated slippage feedback in WMR teleoperation systems under large delays. Conventional model-free predictors often demonstrate significant prediction errors when compensating for large delays in complex dynamics systems due to their limited adaptability and generalizability. In contrast, the proposed PiLSTM predictor integrates an LSTM network with time-delay dynamics and a feedback mechanism to better capture strong nonlinear and temporal dynamics for improved adaptability and generalizability while ensuring accurate prediction. A series of human-in-the-loop experiments for a track-following task were conducted to validate the effectiveness of the PiLSTM framework in compensating for the large delays, similar to those encountered in lunar exploration. Its performance was compared against first-order time-delay (FOD) and wave variable transformation (WV) approaches for delay compensation. The results demonstrate the robustness of the PiLSTM framework across various test scenarios, achieving up to 29.7% improvement in prediction accuracy compared to the FOD framework. Furthermore, the proposed PiLSTM-based teleoperation system achieved significantly improved tracking performance and force transparency compared to FOD- and WV-based systems, ensuring higher fidelity and more robust closed-loop integration. These findings highlight the PiLSTM framework’s adaptability and generalizability for complex teleoperation systems. A supplementary video is available at <span><span>https://youtu.be/dMVLEV7qTvE</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"197 ","pages":"Article 105265"},"PeriodicalIF":5.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145685202","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

Hybrid and trustworthy AI approach to anomaly detection in realistic human–robot collaboration scenarios 基于混合可信人工智能的人机协作异常检测方法

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

Robotics and Autonomous Systems

Pub Date : 2026-03-01 Epub Date: 2025-12-08 DOI: 10.1016/j.robot.2025.105286

Rafał Kozik , Szymon Buś , Jakub Główka , Aleksandra Pawlicka , Marek Pawlicki , Rafał Renk , Michał Choraś

DNN (Deep Neural Networks) have recently demonstrated superior performance in various difficult and challenging tasks, such as natural language processing (NLP) and computer vision (CV). However, the researchers often stress the need for more practical cognitive and reasoning abilities of ML systems. In contrast to DNN, symbolic machine learning demonstrates exceptional reasoning capabilities. It is an obvious and beneficial approach to hybridize these two research domains to merge the advantages of both methodologies in order to achieve neuro-symbolic learning systems that exhibit effective perception and reasoning capabilities. In this paper, the authors propose a neuro-symbolic approach for anomaly detection in a real human–robot collaboration scenario. The emphasis in our research is put on constrained and industrial environments where the so-called cobots assist humans in the assembly process. The proposed approach encompasses several symbolic programs which allow enforcing safety guidelines in the human–robot collaboration as well as make the neural system aware of the physical and environmental constraints. The experiments and the ablation studies show each symbolic program plays an important role and contributes to improved anomaly detection capabilities.

DNN（深度神经网络）最近在各种困难和具有挑战性的任务中表现出优异的性能，例如自然语言处理（NLP）和计算机视觉（CV）。然而，研究人员经常强调机器学习系统需要更实用的认知和推理能力。与深度神经网络相比，符号机器学习展示了卓越的推理能力。将这两个研究领域结合起来，融合两种方法的优势，以实现具有有效感知和推理能力的神经符号学习系统，是一种明显而有益的方法。在本文中，作者提出了一种在真实人机协作场景中进行异常检测的神经符号方法。我们的研究重点放在受限和工业环境中，所谓的协作机器人在装配过程中协助人类。提出的方法包括几个象征性的程序，这些程序允许在人机协作中执行安全指导方针，并使神经系统意识到物理和环境约束。实验和烧蚀研究表明，每个符号程序都发挥了重要作用，有助于提高异常检测能力。

{"title":"Hybrid and trustworthy AI approach to anomaly detection in realistic human–robot collaboration scenarios","authors":"Rafał Kozik , Szymon Buś , Jakub Główka , Aleksandra Pawlicka , Marek Pawlicki , Rafał Renk , Michał Choraś","doi":"10.1016/j.robot.2025.105286","DOIUrl":"10.1016/j.robot.2025.105286","url":null,"abstract":"<div><div>DNN (Deep Neural Networks) have recently demonstrated superior performance in various difficult and challenging tasks, such as natural language processing (NLP) and computer vision (CV). However, the researchers often stress the need for more practical cognitive and reasoning abilities of ML systems. In contrast to DNN, symbolic machine learning demonstrates exceptional reasoning capabilities. It is an obvious and beneficial approach to hybridize these two research domains to merge the advantages of both methodologies in order to achieve neuro-symbolic learning systems that exhibit effective perception and reasoning capabilities. In this paper, the authors propose a neuro-symbolic approach for anomaly detection in a real human–robot collaboration scenario. The emphasis in our research is put on constrained and industrial environments where the so-called cobots assist humans in the assembly process. The proposed approach encompasses several symbolic programs which allow enforcing safety guidelines in the human–robot collaboration as well as make the neural system aware of the physical and environmental constraints. The experiments and the ablation studies show each symbolic program plays an important role and contributes to improved anomaly detection capabilities.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"197 ","pages":"Article 105286"},"PeriodicalIF":5.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145738061","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

A composite motion planning scheme based on time-varying recurrent neural network for mobile robot manipulators 基于时变递归神经网络的移动机器人复合运动规划方案

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

Robotics and Autonomous Systems

Pub Date : 2026-03-01 Epub Date: 2025-12-12 DOI: 10.1016/j.robot.2025.105298

Xitong Gao , Luwen Yang , Zhijun Zhang

In order to solve the double ended (end-effector and mobile platform) motion planning problem of mobile robot manipulator, a composite motion planning (CMP) scheme based on time-varying recurrent neural network is proposed and analyzed. In traditional schemes, motion planning of the end-effector is common, the route of the platform is generated from the end-trajectory calculation. However, in realistic tasks, the end trajectory and platform route are often independent to each other. To do so, kinematic models of the double ended are first derived in detail and formulated as equality constraints, respectively. Secondly, the posture constraint and combined physical constraint are designed and formulated as an equation and inequality constraint, respectively. Then, the CMP scheme is proposed and formulated as a constrained quadratic programming problem. Thirdly, the optimal solution of the quadratic programming problem is obtained by the designed time-varying recurrent neural network. Finally, experiments verify that the proposed CMP scheme can simultaneously plan the double ended of the mobile manipulator.

为解决移动机器人机械臂双端（末端执行器和移动平台）运动规划问题，提出并分析了一种基于时变递归神经网络的复合运动规划方案。在传统的方案中，末端执行器的运动规划是常见的，平台的路线是由末端轨迹计算生成的。然而，在现实任务中，终点轨迹和平台路线往往是相互独立的。为此，首先详细推导了双端机构的运动模型，并分别将其表述为等式约束。其次，将姿态约束和组合物理约束分别设计为方程约束和不等式约束；然后，提出了CMP格式，并将其表述为约束二次规划问题。第三，利用设计的时变递归神经网络求出二次规划问题的最优解。最后，实验验证了所提出的CMP方案能够同时规划移动机械臂的双端。

{"title":"A composite motion planning scheme based on time-varying recurrent neural network for mobile robot manipulators","authors":"Xitong Gao , Luwen Yang , Zhijun Zhang","doi":"10.1016/j.robot.2025.105298","DOIUrl":"10.1016/j.robot.2025.105298","url":null,"abstract":"<div><div>In order to solve the double ended (end-effector and mobile platform) motion planning problem of mobile robot manipulator, a composite motion planning (CMP) scheme based on time-varying recurrent neural network is proposed and analyzed. In traditional schemes, motion planning of the end-effector is common, the route of the platform is generated from the end-trajectory calculation. However, in realistic tasks, the end trajectory and platform route are often independent to each other. To do so, kinematic models of the double ended are first derived in detail and formulated as equality constraints, respectively. Secondly, the posture constraint and combined physical constraint are designed and formulated as an equation and inequality constraint, respectively. Then, the CMP scheme is proposed and formulated as a constrained quadratic programming problem. Thirdly, the optimal solution of the quadratic programming problem is obtained by the designed time-varying recurrent neural network. Finally, experiments verify that the proposed CMP scheme can simultaneously plan the double ended of the mobile manipulator.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"197 ","pages":"Article 105298"},"PeriodicalIF":5.2,"publicationDate":"2026-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145738065","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