A journal of IEEE and CAA , publishes high-quality papers in English on original theoretical/experimental research and development in all areas of automation
Advanced Search

Early Access

Display Method:
Enclosing Formation Control of Multi-Agent Systems under Denial-of-Service Attacks
Mahdi Baradarannia, Farzad Hashemzadeh, Tufan Kumbasar
, Available online  , doi: 10.1109/JAS.2026.126161
[Abstract](97) [FullText HTML] (0) [PDF 3511KB](18)
Abstract:
Multi-Agent System (MAS) operation relies on networked communication, making it vulnerable to cyber-attacks. While much research has focused on how MASs reach consensus when they are under cyber threats, this paper is the first to address the enclosing formation problem with the same geometry as targets for a peer-to-peer communicating MAS in the presence of Denial-of-Service (DoS) attacks. First, a distributed practical finite-time estimator is proposed that enables followers to estimate the geometric center of target agents despite communication disruptions due to DoS attacks. Using Lyapunov functional analysis, it has been shown that the estimation error converges to a bounded region within a finite time. Building on this estimator, a control strategy is developed that ensures the system remains input-to-state stable even in the presence of DoS attacks. To facilitate control gain tuning, a parameter is introduced, providing a systematic approach for determining the controller gains. The approach is supported by rigorous theoretical analysis. To validate its effectiveness, a multi-agent surveillance scenario is simulated in which the communications between followers and targets experience aperiodic disruptions due to DoS attacks. The system shows robust performance during both attack and normal intervals, demonstrating its effectiveness in maintaining an enclosing formation under adversarial conditions.
Optimality-Guaranteed Model-Free Output Feedback Control Using Reinforcement Learning
Li Zhang, Jialu Fan, Tianyou Chai, Frank L. Lewis
, Available online  , doi: 10.1109/JAS.2026.126134
[Abstract](124) [FullText HTML] (0) [PDF 1297KB](46)
Abstract:
This paper investigates a data-driven, model-free optimal output feedback (OPFB) control approach for linear time-invariant (LTI) continuous-time (CT) systems. Existing data-driven OPFB control methods are hindered by the challenge that the optimality of the control policy cannot be guaranteed within the OPFB framework. To overcome this limitation, the paper introduces two novel algorithms aimed at ensuring the optimal OPFB control solution. First, by redefining the traditional performance index, an offline learning algorithm based on an augmented Hamiltonian function (AHF) is introduced. This algorithm ensures the optimality of the OPFB control solution while eliminating the need for an initial control gain. Second, to extend the algorithm’s applicability to systems with unknown dynamics, a data-driven, model-free approach leveraging off-policy integral reinforcement learning (IRL) is developed. This method operates solely on system data, demonstrating robustness against probing noise and ensuring unbiased learning outcomes. The proposed algorithms are validated through numerical simulation experiments using an F-16 aircraft system, confirming their effectiveness in practical scenarios.
Towards the Bi-Imbalance: A Pruning-Based Contrastive Domain Adaptation Framework for Industrial Fault Diagnosis
Gecheng Chen, Zeyu Yang, Jianqiang Li, Xinkai Chen, Chengwen Luo
, Available online  , doi: 10.1109/JAS.2026.126002
[Abstract](216) [FullText HTML] (0) [PDF 3981KB](31)
Abstract:
Industrial fault diagnosis often encounters both distributional shift and severe class imbalance across domains, termed bi-imbalance. Existing unsupervised domain adaptation (UDA) methods largely overlook this, leading to poor generalization or negative transfer. We propose a Pruning-based Contrastive Domain Adaptation (PCDA) framework to address bi-imbalance without relying on pseudo labels or manual weighting. PCDA integrates two key components: 1) An imbalance-aware contrastive learning (Ia-CLR) module that leverages model pruning to detect and emphasize minority samples during self-supervised pretraining; 2) A boundary-aware adversarial domain adaptation (Ba-ADA) module that enhances feature discriminability via pruned supervised contrastive learning. The pruning-induced perturbation implicitly re-weights hard-to-learn samples from both domains. Extensive experiments on benchmark datasets demonstrate that PCDA significantly outperforms state-of-the-art UDA methods under bi-imbalanced conditions.
Knowledge-Assistant Deep Reinforcement Learning for Multi-Agent Region Protection
Siqing Sun, Tianbo Li
, Available online  , doi: 10.1109/JAS.2025.125912
[Abstract](1573) [FullText HTML] (0) [PDF 1323KB](82)
Abstract:
Attribute-Augmented PPR Meets Self-Loops: Simple Yet Effective Defending Graph Neural Networks
Zeli Wang, Jie Li, Zhiqiu Ye, Longlong Lin, Mengdi Wang, Guoyin Wang
, Available online  , doi: 10.1109/JAS.2026.125990
[Abstract](437) [FullText HTML] (0) [PDF 2116KB](47)
Abstract:
Graph Neural Networks (GNNs) are effective in processing graph-structured data but are also known to be vulnerable to adversarial attacks. Limitations of existing defense methodologies include a suboptimal approach to edge and node weight assignment, as such methods generally focus on the graph’s topological structure or node attributes. This imbalance can lead to inaccurate assessments of the importance of neighboring nodes, especially those targeted by attacks. We posit that the model should diminish the influence of insignificant and contaminated neighboring nodes on graph representation learning, thereby mitigating the adverse effects of compromised neighbors. Hence, we propose a versatile GNN adversarial defense framework by combining a novel attribute-augmented Personalized PageRank (PPR) with an adaptive self-loop weight adjustment mechanism, coined GPROP, to defend against adversarial attacks. Specifically, the attribute-augmented PPR helps compute the node similarity more accurately based on global semantics, by considering multi-hop graph topology and node attributes. According to the similarity scores, GPROP assigns corresponding weights to edges and simultaneously prunes insignificant ones. When iteratively aggregating the neighbor information, the proposed self-loop mechanism dynamically adjusts the weight ratio of neighboring nodes and the nodes themselves based on the node’s degree. This strategy reduces the influence of neighboring nodes that may introduce unreliable or harmful information. Moreover, GPROP is model-agnostic and can be readily embedded into different GNN backbones, leading to a substantial improvement in resistance to adversarial perturbations. Experimental results across multiple benchmark datasets show that our proposal consistently outperforms existing approaches, providing enhanced resilience and accuracy.
Cartesian Space Control and Joint Tracking Control for a Robotic Arm System with Explicit-time Proportional Convergence
Wen Yan, Tao Zhao, Ben Niu, Zhiyi Shi, Edmond Q. Wu
, Available online  , doi: 10.1109/JAS.2026.125963
[Abstract](821) [FullText HTML] (6) [PDF 8510KB](92)
Abstract:
The Lyapunov synthesis method is a common controller design strategy in robotic arm motion control. However, it is difficult for this method to achieve fixed-time control without a nonlinear feedback design, whose nonlinearity may cause chattering in the robotic motion. To address this problem, a novel explicit-time control method is proposed using proportional feedback. Not only can the proposed method be applied to the Cartesian space control of the robotic arm system, but it can also be used for the joint-space tracking control. More specifically, under bounded initial condition, the origin of system is attracted to a predefined neighborhood of zero within an explicit fixed-time boundary. Based on that, a robust fixed-time tracking controller of robot is designed by using this linear time-invariant feedback. Besides, compared with other related methods, the proposed method has smoother and lower control input under the same initial condition. In particular, this method enables the robotic arm to achieve a tracking accuracy of 0.1 millimeters and 0.1 degrees within as short as 1.5 seconds, while the repeat positioning accuracy approaches the hardware limit, reaching 0.001 millimeters (±0.03 millimeters) and 0.001 degrees (±0.05 degrees). Theoretical analysis, simulation and experiment verify the main results. Code, data and video are also available, the corresponding links are printed in the relevant places.
Reinforcement Learning-Based Adaptive Optimal Control for a Snake Robot
Yang Xiu, Zhiyi Shi, Guanghong Liu, Rob Law, Dongfang Li, Aiguo Song, Edmond Q. Wu
, Available online  , doi: 10.1109/JAS.2025.125762
[Abstract](808) [FullText HTML] (0) [PDF 6395KB](41)
Abstract:
Due to the difficulty of accurately modeling snake robots, model-based control schemes are ineffective, and the constraints of motion velocity and energy consumption pose challenges to meandering gait. In this work, a two-layer reinforcement learning-based adaptive optimal control framework for snake robots is proposed to achieve trajectory tracking motion of optimal energy efficiency gait. A multi-objective problem for gait amplitude, frequency, and phase is established in the optimization layer, which balances minimizing energy consumption and maximizing velocity by weighted summation. Multiple matching results of gait parameters and performance are obtained through proximal policy optimization, allowing users to select the optimal combination. In the control layer, an actor-critic-identifier neural network-based reinforcement learning optimal controller is designed by considering the difficulty in solving dynamics unknowns and Bellman equation. It adaptively fits the cost function and control policy, reducing the dependence on an accurate model and avoiding computational complexity. Theoretical analysis demonstrates that the proposed method can guarantee stability of tracking errors for snake robots, with optimal cost. Comparative simulation experiment results show the effectiveness and superiority of this method.
Adaptive Dynamic Trade-off Optimization between Manipulability and Sparsity for Redundant Manipulators
Zhaoyang Song, Wei Chen, Huichao Cao
, Available online  , doi: 10.1109/JAS.2026.125759
[Abstract](811) [FullText HTML] (0) [PDF 842KB](25)
Abstract:
Data-Driven Distributed Model Predictive Control for Large-Scale Systems with Actuator Faults
Yan Li, Hao Zhang, Huaicheng Yan, Yongxiao Tian, Yanfei Zhu
, Available online  , doi: 10.1109/JAS.2025.125858
[Abstract](950) [FullText HTML] (0) [PDF 431KB](52)
Abstract:
Majorization-Minimization-Based Neural Dynamics for Time-Variant Optimization Under Multi-Set Constraints
Ying Liufu, Yongji Guan
, Available online  , doi: 10.1109/JAS.2026.125768
[Abstract](988) [FullText HTML] (0) [PDF 678KB](22)
Abstract:
Distributed Optimal Consensus Control of Multi-Agent Systems Under Indifferent and Self-Sacrificing Alienation
Yue Zhang, Yan-Wu Wang, Xiao-Kang Liu
, Available online  , doi: 10.1109/JAS.2025.125861
[Abstract](1183) [FullText HTML] (0) [PDF 1254KB](102)
Abstract:
A New Parameter Estimation Methodology Using Steady State Yaw Rate Measurements for Lateral Vehicle Dynamics
Zhihong Man, Mingcong Deng, Zenghui Wang, Qing-long Han
, Available online  , doi: 10.1109/JAS.2025.125366
[Abstract](1275) [FullText HTML] (0) [PDF 5749KB](122)
Abstract:
In this paper, the lateral dynamics of road vehicles (LDRV) is further studied from the viewpoint of vehicle informatics. It is seen that LDRV is first decoupled and the vehicle slip angle is proved to be observable from the yaw rate measurements. A new methodology of parameter estimation using steady-state yaw rate measurements (PESYRM) is then developed to accurately estimate the parameters of LDRV. The important characteristics of PESYRM comprise four parts: ( i ) The steering angle input to LDRV is chosen as the linear combination of sinusoids; ( ii ) Only the steady state information of yaw rate in any fundamental period is required to accurately estimate the unknown parameters of LDRV; ( iii ) Unlike many existing parameter estimation methods, the time consuming computing of the inverse of high-dimensional data matrix is avoided by making full use of the orthogonal properties of trigonometric base functions; ( iv ) All of system information of LDRV is embedded in the measurements of the steady state yaw rate in any fundamental period. A simulation example is carried out to show the advantages and effectiveness of the new research findings for LDRV.
KT-RC: Kernel Time-Delayed Reservoir Computing for Time Series Prediction
Heshan Wang, Mengmeng Chen, Kunjie Yu, Jing Liang, Zhaomin Lv, Zhong Zhang
, Available online  , doi: 10.1109/JAS.2024.124986
[Abstract](1351) [FullText HTML] (0) [PDF 9327KB](78)
Abstract:
Reservoir computing (RC) is an efficient recurrent neural network (RNN) method. However, the performance and prediction results of traditional RCs are susceptible to several factors, such as their network structure, parameter setting, and selection of input features. In this study, we employ a kernel time-delayed RC (KT-RC) method for time series prediction. The KT-RC transforms input vectors linearly to obtain a high-dimensional set of time-delayed linear eigenvectors, which are then transformed by various kernel functions to represent the nonlinear characteristics of the input signal. Finally, the Bayesian optimization algorithm adjusts the few remaining weights and kernel parameters to minimize the manual adjustment process. The advantages of KT-RC can be summarized as follows: 1) KT-RC solves the problems of uncertainty in weight matrices and difficulty in large-scale parameter selection in the input and hidden layers of RCs. 2) The KT module can avoid massive reservoir hyperparameters and effectively reduce the hidden layer size of the traditional RC. 3) The proposed KT-RC shows good performance, strong stability, and robustness in several synthetic and real-world datasets for one-step-ahead and multistep-ahead time series prediction. The simulation results confirm that KT-RC not only outperforms some gate-structured RNNs, kernel vector regression models, and recently proposed prediction models but also requires fewer parameters to be initialized and can reduce the hidden layer size of the traditional RCs. The source code is available at https://github.com/whs7713578/RC.
Efficient Centralized Traffic Grid Signal Control Based on Meta-Reinforcement Learning
Jia Wu, Yican Lou
, Available online  , doi: 10.1109/JAS.2023.123270
[Abstract](1913) [FullText HTML] (0) [PDF 703KB](88)
Abstract:
Supplementary File of “Push-Sum Based Algorithm for Constrained Convex Optimization Problem and Its Potential Application in Smart Grid”
Qian Xu, Zao Fu, Bo Zou, Hongzhe Liu, Lei Wang
, Available online  
[Abstract](1408) [FullText HTML] (0) [PDF 131KB](1)
Abstract:
Supplementary Material for “Collision and Deadlock Avoidance in Multi-Robot Systems Based on Glued Nodes”
Zichao Xing, Xinyu Chen, Xingkai Wang, Weimin Wu, Ruifen Hu
, Available online  
[Abstract](1925) [FullText HTML] (0) [PDF 8129KB](6)
Abstract: