Robotics and Autonomous Systems最新文献_第2页

Vision-driven river following of UAV via safe reinforcement learning using semantic dynamics model 基于语义动力学模型的安全强化学习的无人机视觉驱动河流跟踪

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

Robotics and Autonomous Systems

Pub Date : 2026-01-17 DOI: 10.1016/j.robot.2026.105357

Zihan Wang, Nina Mahmoudian

Vision-driven autonomous river following by Unmanned Aerial Vehicles is critical for applications such as rescue, surveillance, and environmental monitoring, particularly in dense riverine environments where GPS signals are unreliable. These safety-critical navigation tasks must satisfy hard safety constraints while optimizing performance. Moreover, the reward in river following is inherently history-dependent (non-Markovian) by which river segment has already been visited, making it challenging for standard safe Reinforcement Learning (SafeRL). To address these gaps, we cast river following as a coverage control problem with a submodular reward that exhibits diminishing returns as more river segments are visited, framing the task as a Submodular Markov Decision Process. Building on the SafeRL paradigm and the First Order Constrained Optimization in Policy Space (FOCOPS) algorithm, we propose three contributions.

First, we introduce the Marginal Gain Advantage Estimation (MGAE), which refines the reward advantage function using a sliding-window baseline calculated from historical episodic returns, aligning the advantage estimate with non-Markovian dynamics. Second, we develop a Semantic Dynamics Model (SDM) based on patchified water semantic masks, offering more interpretable and data-efficient short-term prediction of future observations compared to latent vision dynamics models. Third, we present the Constrained Actor Dynamics Estimator (CADE) architecture, which integrates the actor, cost estimator, and SDM for cost advantage estimation to form a model-based SafeRL framework capable of solving partially observable Constrained Submodular Markov Decision Processes.

The simulation results demonstrate that MGAE achieves faster convergence and superior performance compared to critic-based methods like Generalized Advantage Estimation. SDM provides more accurate short-term state predictions, enabling the cost estimator to better predict potential violations. Overall, CADE effectively integrates safety regulation into model-based RL, with the Lagrangian approach providing a “soft” balance between reward and safety during training, while the safety layer enhances inference by imposing a “hard” action overlay. Our code is publicly available on Github (https://github.com/EdisonPricehan/omnisafe-cade/tree/cade).

无人驾驶飞行器的视觉驱动自动河流跟踪对于救援，监视和环境监测等应用至关重要，特别是在GPS信号不可靠的密集河流环境中。这些对安全至关重要的导航任务必须在优化性能的同时满足硬性安全约束。此外，河流跟踪的奖励本质上是历史依赖的（非马尔可夫的），因为河段已经被访问过，这使得标准安全强化学习（SafeRL）具有挑战性。为了解决这些差距，我们将河流跟踪作为一个具有子模块奖励的覆盖控制问题，随着更多的河段被访问，该奖励显示出递减的回报，并将该任务构建为一个子模块马尔可夫决策过程。基于SafeRL范式和一阶约束优化策略空间（FOCOPS）算法，我们提出了三个贡献。首先，我们引入了边际收益优势估计（MGAE），它使用从历史情景收益计算的滑动窗口基线来改进奖励优势函数，使优势估计与非马尔可夫动态保持一致。其次，我们开发了一个基于斑块水语义掩模的语义动力学模型（SDM），与潜在视觉动力学模型相比，该模型提供了更可解释和数据效率更高的未来观测短期预测。第三，我们提出了约束参与者动态估计器（Constrained Actor Dynamics Estimator， CADE）架构，该架构集成了参与者、成本估计器和用于成本优势估计的SDM，形成了一个基于模型的SafeRL框架，能够求解部分可观察的约束次模马尔可夫决策过程。仿真结果表明，与基于临界的广义优势估计方法相比，MGAE具有更快的收敛速度和更好的性能。SDM提供了更准确的短期状态预测，使成本估算人员能够更好地预测潜在的违规行为。总的来说，CADE有效地将安全规则集成到基于模型的强化学习中，拉格朗日方法在训练过程中提供了奖励和安全之间的“软”平衡，而安全层通过施加“硬”动作覆盖来增强推理。我们的代码在Github上是公开的（https://github.com/EdisonPricehan/omnisafe-cade/tree/cade）。

{"title":"Vision-driven river following of UAV via safe reinforcement learning using semantic dynamics model","authors":"Zihan Wang, Nina Mahmoudian","doi":"10.1016/j.robot.2026.105357","DOIUrl":"10.1016/j.robot.2026.105357","url":null,"abstract":"<div><div>Vision-driven autonomous river following by Unmanned Aerial Vehicles is critical for applications such as rescue, surveillance, and environmental monitoring, particularly in dense riverine environments where GPS signals are unreliable. These safety-critical navigation tasks must satisfy hard safety constraints while optimizing performance. Moreover, the reward in river following is inherently history-dependent (non-Markovian) by which river segment has already been visited, making it challenging for standard safe Reinforcement Learning (SafeRL). To address these gaps, we cast river following as a coverage control problem with a submodular reward that exhibits diminishing returns as more river segments are visited, framing the task as a Submodular Markov Decision Process. Building on the SafeRL paradigm and the First Order Constrained Optimization in Policy Space (FOCOPS) algorithm, we propose three contributions.</div><div>First, we introduce the Marginal Gain Advantage Estimation (MGAE), which refines the reward advantage function using a sliding-window baseline calculated from historical episodic returns, aligning the advantage estimate with non-Markovian dynamics. Second, we develop a Semantic Dynamics Model (SDM) based on patchified water semantic masks, offering more interpretable and data-efficient short-term prediction of future observations compared to latent vision dynamics models. Third, we present the Constrained Actor Dynamics Estimator (CADE) architecture, which integrates the actor, cost estimator, and SDM for cost advantage estimation to form a model-based SafeRL framework capable of solving partially observable Constrained Submodular Markov Decision Processes.</div><div>The simulation results demonstrate that MGAE achieves faster convergence and superior performance compared to critic-based methods like Generalized Advantage Estimation. SDM provides more accurate short-term state predictions, enabling the cost estimator to better predict potential violations. Overall, CADE effectively integrates safety regulation into model-based RL, with the Lagrangian approach providing a “soft” balance between reward and safety during training, while the safety layer enhances inference by imposing a “hard” action overlay. Our code is publicly available on Github (<span><span>https://github.com/EdisonPricehan/omnisafe-cade/tree/cade</span><svg><path></path></svg></span>).</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105357"},"PeriodicalIF":5.2,"publicationDate":"2026-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037691","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 self-supervised learning approach to acquire representation of concave object manipulation with sparse tactile sensing 基于稀疏触觉感知的凹物体操作表征的自监督学习方法

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

Robotics and Autonomous Systems

Pub Date : 2026-01-13 DOI: 10.1016/j.robot.2025.105319

Daiki Takamori , Yuichi Kobayashi , Tomohiro Hayakawa , Kosuke Hara , Dotaro Usui

Tactile sensing is essential for improving robotic manipulation, particularly when handling transparent or deformable objects. However, effectively leveraging tactile observations remains a key challenge. In this study, we propose a semi-self-supervised complementary learning framework that integrates visual input with sparse tactile data collected through probing actions. Unlike previous approaches that rely on high-resolution tactile sensors or detailed 3D reconstructions, our method employs sparse tactile sensing to construct object representations via unsupervised learning. The proposed framework enables both complementary and independent recognition through vision and tactile perception, allowing the robot to perform additional probing actions to verify whether its hand has actually reached inside an object. We trained and evaluated our method on opening exploration tasks involving semi-transparent and deformable objects, using a relatively small real-world dataset collected with a robotic hand equipped with a simple tactile sensor.

触觉感知对于提高机器人的操作能力至关重要，尤其是在处理透明或可变形的物体时。然而，有效地利用触觉观察仍然是一个关键的挑战。在这项研究中，我们提出了一种半自监督的互补学习框架，该框架将视觉输入与通过探测动作收集的稀疏触觉数据相结合。与以往依赖高分辨率触觉传感器或详细3D重建的方法不同，我们的方法采用稀疏触觉感知通过无监督学习来构建对象表示。所提出的框架通过视觉和触觉感知实现互补和独立识别，允许机器人执行额外的探测动作，以验证其手是否真的到达了物体内部。我们使用配备简单触觉传感器的机器人手收集的相对较小的真实世界数据集，对涉及半透明和可变形物体的开放探索任务进行了训练和评估。

{"title":"A self-supervised learning approach to acquire representation of concave object manipulation with sparse tactile sensing","authors":"Daiki Takamori , Yuichi Kobayashi , Tomohiro Hayakawa , Kosuke Hara , Dotaro Usui","doi":"10.1016/j.robot.2025.105319","DOIUrl":"10.1016/j.robot.2025.105319","url":null,"abstract":"<div><div>Tactile sensing is essential for improving robotic manipulation, particularly when handling transparent or deformable objects. However, effectively leveraging tactile observations remains a key challenge. In this study, we propose a semi-self-supervised complementary learning framework that integrates visual input with sparse tactile data collected through probing actions. Unlike previous approaches that rely on high-resolution tactile sensors or detailed 3D reconstructions, our method employs sparse tactile sensing to construct object representations via unsupervised learning. The proposed framework enables both complementary and independent recognition through vision and tactile perception, allowing the robot to perform additional probing actions to verify whether its hand has actually reached inside an object. We trained and evaluated our method on opening exploration tasks involving semi-transparent and deformable objects, using a relatively small real-world dataset collected with a robotic hand equipped with a simple tactile sensor.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105319"},"PeriodicalIF":5.2,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978329","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

Task allocation for heterogeneous multi-AUV system with rechargeable docking stations: A multitask bundling auction approach 具有可充电坞站的异构多auv系统任务分配：一种多任务捆绑拍卖方法

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

Robotics and Autonomous Systems

Pub Date : 2026-01-12 DOI: 10.1016/j.robot.2026.105356

Song Han , Jiaao Zhao , Xinbin Li , Liwen Jia , Zhixin Liu

This study proposes a multitask bundling auction task allocation algorithm for heterogeneous multiple autonomous underwater vehicle (AUV) system with rechargeable docking stations. First, a distributed multitask bundling auction model, where multiple AUVs are allowed to win the bid in one auction round, is constructed. Meanwhile, the constructed model allows each winning AUV to achieve multiple tasks, thereby greatly improving the auction efficiency. In the bidding phase, each AUV can autonomously generate a multitask bundle, where the continuity of task execution can be effectively considered. Therefore, the utility of the multi-AUV system and the distributed allocation efficiency can be greatly improved. Second, a Crab Trap Artificial Intelligence (CTAI) algorithm, which mimics the process of catching crabs with crab traps, is proposed to effectively solve the particular constructed multitask bundle generation problem. Meanwhile, the continuity of task execution and the recharging timing for the AUV are comprehensively optimized by the proposed CTAI algorithm, which can efficiently generate the most competitive multitask bundle for each AUV. Moreover, a competition balance mechanism, that can effectively avoid the extra auction rounds caused by popular and unpopular tasks, is proposed to further improve the auction efficiency. The numerical results validate the superiority of the proposed algorithm.

针对具有可充电坞的异构多自主水下航行器系统，提出了一种多任务捆绑拍卖任务分配算法。首先，构建了允许多个auv在一轮拍卖中中标的分布式多任务捆绑拍卖模型；同时，所构建的模型允许每个中标AUV同时完成多个任务，从而大大提高了拍卖效率。在投标阶段，每个AUV可以自主生成一个多任务包，可以有效地考虑任务执行的连续性。因此，可以大大提高多auv系统的实用性和分布式分配效率。其次，提出了一种螃蟹陷阱人工智能（CTAI）算法，利用螃蟹陷阱模拟捕捉螃蟹的过程，有效解决了特定构造的多任务束生成问题。同时，本文提出的CTAI算法对AUV的任务执行连续性和充电时间进行了全面优化，能够有效地为每个AUV生成最具竞争力的多任务束。此外，提出了一种竞争平衡机制，可以有效避免因热门任务和冷门任务而导致的额外拍卖回合，进一步提高拍卖效率。数值结果验证了该算法的优越性。

{"title":"Task allocation for heterogeneous multi-AUV system with rechargeable docking stations: A multitask bundling auction approach","authors":"Song Han , Jiaao Zhao , Xinbin Li , Liwen Jia , Zhixin Liu","doi":"10.1016/j.robot.2026.105356","DOIUrl":"10.1016/j.robot.2026.105356","url":null,"abstract":"<div><div>This study proposes a multitask bundling auction task allocation algorithm for heterogeneous multiple autonomous underwater vehicle (AUV) system with rechargeable docking stations. First, a distributed multitask bundling auction model, where multiple AUVs are allowed to win the bid in one auction round, is constructed. Meanwhile, the constructed model allows each winning AUV to achieve multiple tasks, thereby greatly improving the auction efficiency. In the bidding phase, each AUV can autonomously generate a multitask bundle, where the continuity of task execution can be effectively considered. Therefore, the utility of the multi-AUV system and the distributed allocation efficiency can be greatly improved. Second, a Crab Trap Artificial Intelligence (CTAI) algorithm, which mimics the process of catching crabs with crab traps, is proposed to effectively solve the particular constructed multitask bundle generation problem. Meanwhile, the continuity of task execution and the recharging timing for the AUV are comprehensively optimized by the proposed CTAI algorithm, which can efficiently generate the most competitive multitask bundle for each AUV. Moreover, a competition balance mechanism, that can effectively avoid the extra auction rounds caused by popular and unpopular tasks, is proposed to further improve the auction efficiency. The numerical results validate the superiority of the proposed algorithm.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105356"},"PeriodicalIF":5.2,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978325","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

AdaRRT: A novel adaptive path planning algorithm for mobile robots in complex terrain environments AdaRRT：一种复杂地形环境下移动机器人自适应路径规划算法

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

Robotics and Autonomous Systems

Pub Date : 2026-01-12 DOI: 10.1016/j.robot.2026.105354

Yuqing Chen , Haotong Yan , Shiyu Li , Shunqing Yang , Huosheng Hu

Path planning for mobile robots in complex environments has attracted increasing attention. However, the classical RRT and its variants still suffer from low sampling efficiency, unsafe paths close to obstacles, and susceptibility to local optima. Recently proposed methods, such as Adapted-RRT, QS-RRT*, and PQ-RRT*, partially address these issues but rely on complex operators and parameter tuning, leading to high computational costs and poor real-time suitability. To address these limitations, we propose an Adaptive RRT (AdaRRT). Its qualitative advantages include adaptive exploration via dynamic sampling adjustment, improved safety through obstacle-avoiding growth, enhanced robustness with local-optima escape, and smoother paths by tangent-based optimization. Quantitatively, simulations show that AdaRRT reduces search time by up to 91.14%, shortens path length by 20-33%, and decreases node count by over 70%. Real-world experiments further validate its efficiency and safety. Overall, AdaRRT outperforms existing methods in efficiency, safety, and robustness, offering a practical solution for autonomous navigation in complex environments.

复杂环境下移动机器人的路径规划问题越来越受到人们的关注。然而，经典的RRT及其变体仍然存在采样效率低、靠近障碍物的路径不安全、易受局部最优的影响等问题。最近提出的方法，如adaptive - rrt、QS-RRT*和PQ-RRT*，部分解决了这些问题，但依赖于复杂的算子和参数调优，导致计算成本高，实时性差。为了解决这些限制，我们提出了自适应RRT （AdaRRT）。它的定性优势包括通过动态采样调整进行自适应探索，通过避障生长提高安全性，通过局部最优逃逸增强鲁棒性，以及通过切线优化实现路径平滑。定量仿真结果表明，AdaRRT算法的搜索时间缩短了91.14%，路径长度缩短了20-33%，节点数减少了70%以上。实际实验进一步验证了该方法的有效性和安全性。总的来说，AdaRRT在效率、安全性和鲁棒性方面优于现有方法，为复杂环境下的自主导航提供了实用的解决方案。

{"title":"AdaRRT: A novel adaptive path planning algorithm for mobile robots in complex terrain environments","authors":"Yuqing Chen , Haotong Yan , Shiyu Li , Shunqing Yang , Huosheng Hu","doi":"10.1016/j.robot.2026.105354","DOIUrl":"10.1016/j.robot.2026.105354","url":null,"abstract":"<div><div>Path planning for mobile robots in complex environments has attracted increasing attention. However, the classical RRT and its variants still suffer from low sampling efficiency, unsafe paths close to obstacles, and susceptibility to local optima. Recently proposed methods, such as Adapted-RRT, QS-RRT*, and PQ-RRT*, partially address these issues but rely on complex operators and parameter tuning, leading to high computational costs and poor real-time suitability. To address these limitations, we propose an Adaptive RRT (AdaRRT). Its qualitative advantages include adaptive exploration via dynamic sampling adjustment, improved safety through obstacle-avoiding growth, enhanced robustness with local-optima escape, and smoother paths by tangent-based optimization. Quantitatively, simulations show that AdaRRT reduces search time by up to 91.14%, shortens path length by 20-33%, and decreases node count by over 70%. Real-world experiments further validate its efficiency and safety. Overall, AdaRRT outperforms existing methods in efficiency, safety, and robustness, offering a practical solution for autonomous navigation in complex environments.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105354"},"PeriodicalIF":5.2,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037684","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

Spiking control of dielectric elastomer actuators 介电弹性体致动器的尖峰控制

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

Robotics and Autonomous Systems

Pub Date : 2026-01-09 DOI: 10.1016/j.robot.2026.105353

Lukas Sohlbach , Fernando Pérez-Peña , Karsten Schmidt

Rigid robots are highly specialised and can perform tasks with incredible precision. In contrast, soft robots provide a promising solution for creating robotic systems that are inherently better suited for unstructured and dynamic environments. Artificial muscles comprise one of the core components of soft robots. Among them, dielectric elastomer actuators (DEAs) represent the technology that comes closest to the capabilities of a natural muscle. However, their viscoelastic effects may limit the applicability and represent the main reason why suitable control methods are required. Thus, the objective of this work is to have a look at bioinspired spiking closed-loop control systems. By doing so, the research attempts to take a step towards creating true soft robots, which are bioinspired in all systems. A spiking neural network (SNN) is developed that comprised the main part of the controller and whose output is used as the control value. All information inside the controller was represented via spikes and the controller was implemented on neuromorphic hardware. During the validation, the general functionality was proven and a frequency-dependent tracking performance was observed. In a frequency range comparable to other works (≤ 0.5 Hz), the qualitative evaluation shows a good tracking performance even with a sinusoidal input.

刚性机器人是高度专业化的，可以以令人难以置信的精度执行任务。相比之下，软机器人为创建机器人系统提供了一个很有前途的解决方案，它本质上更适合于非结构化和动态环境。人造肌肉是软机器人的核心部件之一。其中，介电弹性体致动器（dea）代表了最接近天然肌肉能力的技术。然而，它们的粘弹性效应可能会限制其适用性，这是需要适当控制方法的主要原因。因此，这项工作的目的是有一个生物启发脉冲闭环控制系统。通过这样做，该研究试图朝着创造真正的软体机器人迈出一步，这些机器人在所有系统中都是受生物启发的。设计了一种脉冲神经网络（SNN），该网络由控制器的主要部分组成，其输出作为控制值。控制器内部的所有信息都通过尖峰表示，控制器在神经形态硬件上实现。在验证期间，证明了一般功能，并观察到频率相关的跟踪性能。在与其他作品相当的频率范围内（≤0.5 Hz），定性评估显示即使在正弦输入下也具有良好的跟踪性能。

{"title":"Spiking control of dielectric elastomer actuators","authors":"Lukas Sohlbach , Fernando Pérez-Peña , Karsten Schmidt","doi":"10.1016/j.robot.2026.105353","DOIUrl":"10.1016/j.robot.2026.105353","url":null,"abstract":"<div><div>Rigid robots are highly specialised and can perform tasks with incredible precision. In contrast, soft robots provide a promising solution for creating robotic systems that are inherently better suited for unstructured and dynamic environments. Artificial muscles comprise one of the core components of soft robots. Among them, dielectric elastomer actuators (DEAs) represent the technology that comes closest to the capabilities of a natural muscle. However, their viscoelastic effects may limit the applicability and represent the main reason why suitable control methods are required. Thus, the objective of this work is to have a look at bioinspired spiking closed-loop control systems. By doing so, the research attempts to take a step towards creating true soft robots, which are bioinspired in all systems. A spiking neural network (SNN) is developed that comprised the main part of the controller and whose output is used as the control value. All information inside the controller was represented via spikes and the controller was implemented on neuromorphic hardware. During the validation, the general functionality was proven and a frequency-dependent tracking performance was observed. In a frequency range comparable to other works (≤ 0.5 Hz), the qualitative evaluation shows a good tracking performance even with a sinusoidal input.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105353"},"PeriodicalIF":5.2,"publicationDate":"2026-01-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978330","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 attention-guided RRT*: Learning spatial sampling priors for accelerated path planning 混合注意引导RRT*：学习空间采样先验加速路径规划

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

Robotics and Autonomous Systems

Pub Date : 2026-01-09 DOI: 10.1016/j.robot.2026.105338

Asmaa Loulou , Mustafa Unel

Sampling-based planners such as RRT* are widely used for motion planning in high-dimensional and complex environments. However, their reliance on uniform sampling often leads to slow convergence and inefficiency, especially in scenarios with narrow passages or long-range dependencies. To address this, we propose HAGRRT*, a Hybrid Attention-Guided RRT* algorithm that learns to generate spatially informed sampling priors. Our method introduces a new neural architecture that fuses multi-scale convolutional features with a lightweight cross-attention mechanism, explicitly conditioned on the start and goal positions. These features are decoded via a DPT-inspired module to produce 2D probability maps that guide the sampling process. Additionally, we propose an obstacle-aware loss function that penalizes disconnected and infeasible predictions which further encourages the network to focus on traversable, goal-directed regions. Extensive experiments on both structured (maze) and unstructured (forest) environments show that HAGRRT* achieves significantly faster convergence and improved path quality compared to both classical RRT* and recent deep-learning guided variants. Our method consistently requires fewer iterations and samples and is able to generalize across varying dataset types. On structured scenarios, our method achieves an average reduction of 39.6% in the number of samples and an average of 24.4% reduction in planning time compared to recent deep learning methods. On unstructured forest maps, our method reduces the number of samples by 71.5%, and planning time by 81.7% compared to recent deep learning methods, and improves the success rate from 67% to 93%. These results highlight the robustness, efficiency, and generalization ability of our approach across a wide range of planning environments.

基于采样的规划器，如RRT*，广泛用于高维和复杂环境中的运动规划。然而，它们对统一采样的依赖往往导致缓慢的收敛和低效率，特别是在狭窄通道或长期依赖的情况下。为了解决这个问题，我们提出了HAGRRT*，这是一种混合注意引导RRT*算法，它可以学习生成空间信息采样先验。我们的方法引入了一种新的神经结构，它融合了多尺度卷积特征和轻量级的交叉注意机制，明确地以起点和目标位置为条件。这些特征通过dpt启发的模块解码，以产生指导采样过程的二维概率图。此外，我们提出了一个障碍感知损失函数，该函数惩罚断开和不可行的预测，从而进一步鼓励网络关注可遍历的目标导向区域。在结构化（迷宫）和非结构化（森林）环境中进行的大量实验表明，与经典RRT*和最近的深度学习引导变体相比，HAGRRT*实现了显著更快的收敛和更好的路径质量。我们的方法始终需要更少的迭代和样本，并且能够跨不同的数据集类型进行推广。在结构化场景中，与最近的深度学习方法相比，我们的方法平均减少了39.6%的样本数量，平均减少了24.4%的规划时间。在非结构化森林地图上，与目前的深度学习方法相比，我们的方法减少了71.5%的样本数量，减少了81.7%的规划时间，并将成功率从67%提高到93%。这些结果突出了我们的方法在广泛的规划环境中的鲁棒性、效率和泛化能力。

{"title":"Hybrid attention-guided RRT*: Learning spatial sampling priors for accelerated path planning","authors":"Asmaa Loulou , Mustafa Unel","doi":"10.1016/j.robot.2026.105338","DOIUrl":"10.1016/j.robot.2026.105338","url":null,"abstract":"<div><div>Sampling-based planners such as RRT* are widely used for motion planning in high-dimensional and complex environments. However, their reliance on uniform sampling often leads to slow convergence and inefficiency, especially in scenarios with narrow passages or long-range dependencies. To address this, we propose HAGRRT*, a Hybrid Attention-Guided RRT* algorithm that learns to generate spatially informed sampling priors. Our method introduces a new neural architecture that fuses multi-scale convolutional features with a lightweight cross-attention mechanism, explicitly conditioned on the start and goal positions. These features are decoded via a DPT-inspired module to produce 2D probability maps that guide the sampling process. Additionally, we propose an obstacle-aware loss function that penalizes disconnected and infeasible predictions which further encourages the network to focus on traversable, goal-directed regions. Extensive experiments on both structured (maze) and unstructured (forest) environments show that HAGRRT* achieves significantly faster convergence and improved path quality compared to both classical RRT* and recent deep-learning guided variants. Our method consistently requires fewer iterations and samples and is able to generalize across varying dataset types. On structured scenarios, our method achieves an average reduction of <strong>39.6%</strong> in the number of samples and an average of <strong>24.4%</strong> reduction in planning time compared to recent deep learning methods. On unstructured forest maps, our method reduces the number of samples by <strong>71.5%</strong>, and planning time by <strong>81.7%</strong> compared to recent deep learning methods, and improves the success rate from <strong>67%</strong> to <strong>93%</strong>. These results highlight the robustness, efficiency, and generalization ability of our approach across a wide range of planning environments.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105338"},"PeriodicalIF":5.2,"publicationDate":"2026-01-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978328","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

Power in numbers: Primitive algorithm for swarm robot navigation in unknown environments 数量的力量：未知环境下群体机器人导航的原始算法

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.2026.105329

Yusuke Tsunoda , Shoken Otsuka , Kazuki Ito , Runze Xiao , Keisuke Naniwa , Yuichiro Sueoka , Koichi Osuka

Recently, the navigation of mobile robots in unknown environments has become a particularly significant research topic. Previous studies have primarily adopted real-time environmental mapping using cameras and LiDAR, along with self-localization and path generation based on those maps. However, there are many areas in the natural environment, such as soft ground and depressions hidden by dead leaves, where distance and vision sensors cannot determine whether the robot can move or not. It is also difficult to rigorously observe and model an unknown environment that changes unpredictably from moment to moment. Therefore, we accept that it is inevitable that robots get stuck in unknown environments, and rather advocate an approach that takes advantage of stucked robots. This study proposes a simple navigation algorithm for traversing unknown environments by utilizes the number of swarm robots. The proposed algorithm assumes that the robot has only the simple function of sensing the direction of the goal and the relative positions of the surrounding robots. The robots can navigate an unknown environment by simply continuing towards the goal while bypassing surrounding robots. In this method, each robot senses only two things: 1. its direction toward the goal, and 2. its relative position to the surrounding robots, and does not need to determine whether it or the surrounding robots are stuck, nor does it need complex robot communication. We mathematically validate the proposed navigation algorithm, present numerical simulations based on the potential field method, and conduct experimental demonstrations using developed robots based on the sound fields for navigation.

近年来，移动机器人在未知环境中的导航成为一个特别重要的研究课题。之前的研究主要采用的是使用摄像头和激光雷达的实时环境地图，以及基于这些地图的自定位和路径生成。然而，在自然环境中，有许多区域，如柔软的地面和被枯叶隐藏的洼地，距离和视觉传感器无法确定机器人是否可以移动。严格地观察和模拟一个每时每刻都在不可预测地变化的未知环境也是很困难的。因此，我们接受机器人被困在未知环境中是不可避免的，并主张一种利用被困机器人的方法。本研究提出一种利用群机器人数量穿越未知环境的简单导航算法。该算法假设机器人仅具有感知目标方向和周围机器人相对位置的简单功能。机器人可以绕过周围的机器人，继续朝着目标前进，从而在未知的环境中导航。在这种方法中，每个机器人只感知两件事：1。它指向目标的方向，2。它与周围机器人的相对位置，不需要判断它与周围机器人是否卡住，也不需要复杂的机器人通信。我们在数学上验证了所提出的导航算法，给出了基于势场方法的数值模拟，并使用基于声场的机器人进行了实验演示。

{"title":"Power in numbers: Primitive algorithm for swarm robot navigation in unknown environments","authors":"Yusuke Tsunoda , Shoken Otsuka , Kazuki Ito , Runze Xiao , Keisuke Naniwa , Yuichiro Sueoka , Koichi Osuka","doi":"10.1016/j.robot.2026.105329","DOIUrl":"10.1016/j.robot.2026.105329","url":null,"abstract":"<div><div>Recently, the navigation of mobile robots in unknown environments has become a particularly significant research topic. Previous studies have primarily adopted real-time environmental mapping using cameras and LiDAR, along with self-localization and path generation based on those maps. However, there are many areas in the natural environment, such as soft ground and depressions hidden by dead leaves, where distance and vision sensors cannot determine whether the robot can move or not. It is also difficult to rigorously observe and model an unknown environment that changes unpredictably from moment to moment. Therefore, we accept that it is inevitable that robots get stuck in unknown environments, and rather advocate an approach that takes advantage of stucked robots. This study proposes a simple navigation algorithm for traversing unknown environments by utilizes the number of swarm robots. The proposed algorithm assumes that the robot has only the simple function of sensing the direction of the goal and the relative positions of the surrounding robots. The robots can navigate an unknown environment by simply continuing towards the goal while bypassing surrounding robots. In this method, each robot senses only two things: 1. its direction toward the goal, and 2. its relative position to the surrounding robots, and does not need to determine whether it or the surrounding robots are stuck, nor does it need complex robot communication. We mathematically validate the proposed navigation algorithm, present numerical simulations based on the potential field method, and conduct experimental demonstrations using developed robots based on the sound fields for navigation.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105329"},"PeriodicalIF":5.2,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037692","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

An Efficient Geometry-Informed Inverse Kinematics of a 7 DOF Cable-Driven Manipulator with Non-Sphere Shoulder and Wrist 具有非球面肩腕的7自由度缆索驱动机械臂的高效几何信息逆运动学

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.2026.105335

Zhiwei Wu , Lei Yan , Tianhong Cheng , Wenfu Xu , Bin Liang

The cable-driven redundant manipulator (CDRM) is characterized by its lightweight, low inertia, and inherent compliance, enabling a wide range of applications in fields such as home services and medical rehabilitation. However, due to its complicated cable drive and transmission mechanism, compared with traditional redundant manipulator, the additional coupling cable kinematics is introduced into the inverse kinematics. Further, as the number of coupled equivalent joints of CDRM increases, it becomes challenging to obtain an efficient as well as stable inverse kinematics solution. In this paper, we propose an efficient geometry-informed inverse kinematics method by combining the geometry-based analytical approach and gradient-based numerical approach. First, the CDRM with 11 equivalent kinematic joints is reconstructed into a 7-DOF manipulator without joint offset. Based on the geometric characteristics, the analytical inverse kinematics of the reconstructed offset-free manipulator is derived to provide physically explainable iterative initial values in approximate solution space for numerical approach. Several numerical calculation results demonstrate that our method inherits the advantages of analytical approach, achieving accurate IK solutions, and improving the computational efficiency and the number of feasible solutions. Additionally, it also addresses the divergence issue resulting from irrational selection of initial values in numerical approach. Furthermore, the solution space can be comprehensively exploited by intuitively adjusting the arm-shape parameters and optimizing the manipulator’s configuration, in order to avoid surrounding obstacles, and optimize cable-tension distribution.

电缆驱动冗余机械手（CDRM）具有重量轻、惯性小、固有顺应性强等特点，在家庭服务、医疗康复等领域有着广泛的应用。然而，由于其缆索驱动和传动机构复杂，与传统的冗余度机械手相比，将附加的耦合缆索运动学引入了逆运动学中。此外，随着CDRM耦合等效关节数量的增加，获得高效且稳定的逆运动学解变得具有挑战性。本文将基于几何的解析方法和基于梯度的数值方法相结合，提出了一种有效的几何信息逆运动学方法。首先，将具有11个等效运动关节的CDRM重构为无关节偏移的7自由度机械臂；基于其几何特征，导出了重构无偏移机械手的解析运动学逆解，为数值求解提供了近似解空间中物理上可解释的迭代初值。数值计算结果表明，该方法继承了解析方法的优点，获得了精确的IK解，提高了计算效率和可行解的数量。此外，还解决了数值方法中由于初始值选择不合理而产生的发散问题。此外，通过直观地调整臂形参数和优化机械手构型，可以综合开发解空间，以避开周围障碍物，优化索张力分布。

{"title":"An Efficient Geometry-Informed Inverse Kinematics of a 7 DOF Cable-Driven Manipulator with Non-Sphere Shoulder and Wrist","authors":"Zhiwei Wu , Lei Yan , Tianhong Cheng , Wenfu Xu , Bin Liang","doi":"10.1016/j.robot.2026.105335","DOIUrl":"10.1016/j.robot.2026.105335","url":null,"abstract":"<div><div>The cable-driven redundant manipulator (CDRM) is characterized by its lightweight, low inertia, and inherent compliance, enabling a wide range of applications in fields such as home services and medical rehabilitation. However, due to its complicated cable drive and transmission mechanism, compared with traditional redundant manipulator, the additional coupling cable kinematics is introduced into the inverse kinematics. Further, as the number of coupled equivalent joints of CDRM increases, it becomes challenging to obtain an efficient as well as stable inverse kinematics solution. In this paper, we propose an efficient geometry-informed inverse kinematics method by combining the geometry-based analytical approach and gradient-based numerical approach. First, the CDRM with 11 equivalent kinematic joints is reconstructed into a 7-DOF manipulator without joint offset. Based on the geometric characteristics, the analytical inverse kinematics of the reconstructed offset-free manipulator is derived to provide physically explainable iterative initial values in approximate solution space for numerical approach. Several numerical calculation results demonstrate that our method inherits the advantages of analytical approach, achieving accurate IK solutions, and improving the computational efficiency and the number of feasible solutions. Additionally, it also addresses the divergence issue resulting from irrational selection of initial values in numerical approach. Furthermore, the solution space can be comprehensively exploited by intuitively adjusting the arm-shape parameters and optimizing the manipulator’s configuration, in order to avoid surrounding obstacles, and optimize cable-tension distribution.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105335"},"PeriodicalIF":5.2,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978332","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

Obstacle crossing in revolute and prismatic knee underactuated biped robots 旋转和移动膝关节欠驱动双足机器人的越障研究

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.2026.105340

Krishnendu Roy , R. Prasanth Kumar

Obstacle crossing is an important ability in biped and humanoid robots that are designed to traverse unstructured terrain. We consider the problem of determining the maximum (a) height, (b) width, (c) cross-sectional area, (d) thin vertical barrier height, and (e) square area of the obstacle that an underactuated biped robot with point-feet can cross while walking slowly. Two different biped robot configurations are compared for obstacle crossing: revolute knee and prismatic knee. The path needed to overcome the obstacle without touching it is determined with the help of binary occupancy grid in the sagittal plane and using genetic algorithm based maximization for each of the five cases, considering thin links as well as thick links for the biped robots. The determined collision free path for obstacle crossing is implemented as a trajectory and demonstrated in dynamic simulation in Mujoco simulation environment. In order to control the position of zero moment point (ZMP) and the ground projection of center of mass for stability, a reaction wheel in the torso is utilized. It is observed that increasing the thicknesses of the biped robot links in general has an effect of reducing the maximum size of the obstacle that can be crossed. Further, prismatic knee biped robot performs better than revolute knee biped robot in crossing large obstacles, especially with thick links. Experiments on a prismatic-knee biped robot further validate the results of GA and MuJoCo simulations.

越障是两足机器人和类人机器人在穿越非结构化地形时的一项重要能力。我们考虑的问题是确定(a)高度，(b)宽度，(c)横截面积，(d)薄垂直障碍物高度，(e)点足双足机器人在缓慢行走时可以穿过的障碍物的平方面积。比较了两种不同的双足机器人构型：旋转膝关节和移动膝关节。通过矢状面上的二元占用网格，结合两足机器人的细连杆和粗连杆，利用基于遗传算法的最大化方法，确定了在不接触障碍物的情况下克服障碍物所需的路径。确定的无碰撞过障路径以轨迹形式实现，并在Mujoco仿真环境中进行了动态仿真验证。为了控制零力矩点的位置和质心的地面投影以保持稳定性，在躯干上设置了反作用轮。我们观察到，一般来说，增加双足机器人连杆的厚度会减小可穿越障碍物的最大尺寸。此外，移动膝关节双足机器人在穿越大型障碍物，特别是粗连杆障碍物时，表现优于旋转膝关节双足机器人。在棱镜膝关节双足机器人上的实验进一步验证了遗传算法和MuJoCo仿真的结果。

{"title":"Obstacle crossing in revolute and prismatic knee underactuated biped robots","authors":"Krishnendu Roy , R. Prasanth Kumar","doi":"10.1016/j.robot.2026.105340","DOIUrl":"10.1016/j.robot.2026.105340","url":null,"abstract":"<div><div>Obstacle crossing is an important ability in biped and humanoid robots that are designed to traverse unstructured terrain. We consider the problem of determining the maximum (a) height, (b) width, (c) cross-sectional area, (d) thin vertical barrier height, and (e) square area of the obstacle that an underactuated biped robot with point-feet can cross while walking slowly. Two different biped robot configurations are compared for obstacle crossing: revolute knee and prismatic knee. The path needed to overcome the obstacle without touching it is determined with the help of binary occupancy grid in the sagittal plane and using genetic algorithm based maximization for each of the five cases, considering thin links as well as thick links for the biped robots. The determined collision free path for obstacle crossing is implemented as a trajectory and demonstrated in dynamic simulation in Mujoco simulation environment. In order to control the position of zero moment point (ZMP) and the ground projection of center of mass for stability, a reaction wheel in the torso is utilized. It is observed that increasing the thicknesses of the biped robot links in general has an effect of reducing the maximum size of the obstacle that can be crossed. Further, prismatic knee biped robot performs better than revolute knee biped robot in crossing large obstacles, especially with thick links. Experiments on a prismatic-knee biped robot further validate the results of GA and MuJoCo simulations.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105340"},"PeriodicalIF":5.2,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145939678","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

Dynamic merging and splitting for large-scale swarm navigation of UAVs and UGVs in unknown area 未知区域无人机与ugv大规模群导航的动态合并与分裂

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.2026.105337

Jiang Zhao , Shukun Chen , Pei Chi , Yingxun Wang

This paper addresses the challenge of achieving efficient collaborative navigation and obstacle avoidance for aerial-ground unmanned swarm in unknown area by proposing a dynamic merge and split method. Firstly, a UAV-dominated collaborative decision-making framework is proposed, which includes a UAV allocation method driven by environmental information and a UGV planning and allocation method guided by UAVs. This framework enables collaborative and dynamic decision-making for the merge and split of the aerial-ground unmanned swarm. Secondly, an integrated dual-model and self-organizing motion control scheme for the aerial-ground unmanned swarm is designed. For the UAV swarm, a tracking–navigation dual-model motion control is developed to enhance robustness and sensing efficiency in unknown area. For the ground vehicles, a dynamic-boundary-based self-organizing motion control is proposed to leverage their high autonomy. Collectively, these control strategies yield highly efficient and coordinated motion control for the aerial-ground unmanned swarm. Finally, numerical simulation experiments are conducted and performance metrics are designed to compare the proposed method with existing representative approaches. The results indicate that the advantages of the proposed method gradually become more pronounced as the swarm scale increases.

本文提出了一种动态合并与分裂方法，解决了未知区域中地空无人机群高效协同导航与避障的难题。首先，提出了以无人机为主导的协同决策框架，包括环境信息驱动下的无人机分配方法和无人机引导下的UGV规划分配方法。该框架实现了空中-地面无人机群合并和分裂的协作和动态决策。其次，设计了一种集成的双模型自组织无人机群运动控制方案。针对无人机群，提出了一种跟踪-导航双模型运动控制方法，提高了未知区域的鲁棒性和感知效率。针对地面车辆的高自主性，提出了一种基于动态边界的自组织运动控制方法。总的来说，这些控制策略为空中-地面无人蜂群提供了高效和协调的运动控制。最后，进行了数值模拟实验，并设计了性能指标，将所提出的方法与已有的代表性方法进行了比较。结果表明，随着群体规模的增大，该方法的优势逐渐显现。

{"title":"Dynamic merging and splitting for large-scale swarm navigation of UAVs and UGVs in unknown area","authors":"Jiang Zhao , Shukun Chen , Pei Chi , Yingxun Wang","doi":"10.1016/j.robot.2026.105337","DOIUrl":"10.1016/j.robot.2026.105337","url":null,"abstract":"<div><div>This paper addresses the challenge of achieving efficient collaborative navigation and obstacle avoidance for aerial-ground unmanned swarm in unknown area by proposing a dynamic merge and split method. Firstly, a UAV-dominated collaborative decision-making framework is proposed, which includes a UAV allocation method driven by environmental information and a UGV planning and allocation method guided by UAVs. This framework enables collaborative and dynamic decision-making for the merge and split of the aerial-ground unmanned swarm. Secondly, an integrated dual-model and self-organizing motion control scheme for the aerial-ground unmanned swarm is designed. For the UAV swarm, a tracking–navigation dual-model motion control is developed to enhance robustness and sensing efficiency in unknown area. For the ground vehicles, a dynamic-boundary-based self-organizing motion control is proposed to leverage their high autonomy. Collectively, these control strategies yield highly efficient and coordinated motion control for the aerial-ground unmanned swarm. Finally, numerical simulation experiments are conducted and performance metrics are designed to compare the proposed method with existing representative approaches. The results indicate that the advantages of the proposed method gradually become more pronounced as the swarm scale increases.</div></div>","PeriodicalId":49592,"journal":{"name":"Robotics and Autonomous Systems","volume":"198 ","pages":"Article 105337"},"PeriodicalIF":5.2,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037688","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