Deformable linear object (DLO) manipulation is ubiquitous in complex manufacturing processes. However, reliable DLO manipulation in practical robotic systems remains challenging, due to high state dimensionality and complex deformation behaviors. Most existing methods assume fixed end-effector boundary conditions and simplified hand-object interactions, thereby limiting the accuracy of shape prediction and control performance. To address this limitation, a tactile rolling fingertip-informed DLO shape modeling and manipulation method is proposed for bimanual cable routing. The tactile rolling fingertip (TRF) is designed to provide extensive rolling tactile sensing and continuous estimation of hand-object relative poses. An enhanced DLO deformation model is formulated by explicitly incorporating fingertip feeding motion and dynamic hand-object pose parameters into the mass-spring representation. The improved model enables DLO shape prediction driven by coupled feeding-pose dynamics. The proposed framework is validated through simulations and real-world experiments. Experimental results demonstrate that the proposed method reduces the angular standard deviation of in-hand object pose estimation to 1.89° The root-mean-square deviation of the DLO deformation prediction is reduced by approximately 34 % compared to the baseline. These results indicate that the proposed approach improves the accuracy and robustness of closed-loop bimanual DLO manipulation for cable routing tasks.

With the extensive application of robotic manipulators in precision assembly, flexible grasping, human-robot collaboration, and complex contact manipulation tasks, safe and stable compliant force control has become a critical issue for enhancing robots’ environmental adaptability and interaction reliability. In complex task scenarios characterized by unknown stiffness, varying contact states, and significant force-position coupling, existing compliant control methods still face challenges such as the reliance of impedance parameters on manual tuning, insufficient parameter adaptability across diverse environments, difficulty in coordinating force regulation with position trajectory stability, and susceptibility to shocks and oscillations during the transition from free-space to contact-space. To address these challenges, this paper proposes CompliantLLM, an adaptive compliant force control method driven by multimodal large models for complex contact tasks. First, a Multi-modal Condition Awareness-driven Impedance Parameter Generation method is proposed. By fusing force, kinematic, visual, and historical state information for contact condition identification and parameter inference, it achieves the adaptive matching of compliant force control parameters in unknown environments. Second, a Safe Trajectory Compensation method under Force-Position Coupling constraints is proposed. By jointly utilizing contact force deviations, end-effector pose errors, and operational states to drive trajectory correction, it facilitates the synergistic improvement of force tracking accuracy and position trajectory stability. Finally, a Contact State Aware parameter Smoothing and Shock Inhibition Strategy is proposed. Through contact phase identification, continuous parameter transition, and safety monitoring constraints, this mechanism achieves reduced overshoot, oscillation suppression, and enhanced interaction safety during the transition from free-space to contact-space. Experimental results demonstrate that CompliantLLM effectively improves the force control accuracy, contact stability, trajectory tracking capability, and compliance of interaction for robotic manipulators across environments with varying stiffness.

During the grinding of aeroengine blade edges, complex time-varying nonlinear coupling and uncertain disturbances pose challenges to the adaptive regulation of constant force grinding, reducing process stability and precision. This paper proposed a multi-modal fusion-enhanced fuzzy adaptive variable impedance control with improved deep belief network (DBN) for robotic constant force blade grinding. Specifically, the three-dimensional model and point cloud model of the blade are integrated to extract accurate geometric information and generate reference grinding trajectories. Furtherly, the DBN training hyperparameters are optimized using linear success history-based adaptive differential evolution (LSHADE). This improves the DBN configuration and overcomes the limitations of conventional DBN based force compensation with fixed network structures and single modality inputs. On this basis, a fuzzy adaptive variable impedance control method based on the improved DBN is developed. Geometric, force/pose, and error modalities are fused to dynamically adjust the force compensation term. This design enables the controller to outperform conventional adaptive variable impedance methods under strongly time-varying conditions. It improves the interaction between the robot and the environment and realizes adaptive active compliant constant-force control in robotic grinding. Comparative experiments demonstrate the stability and reliability of the proposed method. Compared with mainstream methods, the proposed method reduces the grinding force error by 66.7% and 28.6%, respectively. The key error metrics MSE, RMSE, MAPE, and MAE are reduced by more than 71% and 20%, and the average surface roughness is reduced by approximately 15.6% and 5.8%, respectively

Harmonic drives are core components in robotic joints, yet their complex nonlinear friction severely constrains the positioning accuracy and motion smoothness of robots. To establish a high-precision and robust friction model, this paper proposes a systematic modeling method for joint harmonic drives based on multi-objective optimization. First, to comprehensively describe friction characteristics under complex working conditions, a decoupled friction model is established that considers the effects of velocity, temperature, and load. Second, to address the challenge that traditional parameter identification methods are prone to local optima and ignore physical constraints, this paper innovatively formulates parameter identification as a multi-objective optimization problem that balances model fitting accuracy with the physical meaningfulness of parameters. Furthermore, a unique two-step identification strategy is proposed: a genetic algorithm is used to globally optimize the physically feasible domain (search boundaries) of the parameters, followed by precise identification within the optimized boundaries. This ensures the physical rationality of the parameters and the generalization ability of the model. Finally, the proposed methods are validated on an experimental platform under various conditions. The results show that the method can effectively identify model parameters, and the established model achieves high accuracy (dynamic tracking error is less than 0.5A). The findings of this research provide a new theoretical and methodological foundation for friction modeling and compensation in robotic joints.

For welded components without CAD models or prior geometry, robots face significant challenges in translating and interpreting design information into welding-relevant knowledge. When facing unfamiliar workpieces, welding robots cannot autonomously understand seam, edge, and surface elements. Consequently, welding seam localization and trajectory planning still rely heavily on manual teaching and experience-based search. In this study, a weld local feature descriptor (W-LFD) is developed to capture the local patterns of typical weld joints using interpretable geometric statistics, including inter-cluster surface angle, local density coefficient of variation, and normalized dispersion ratio. Together with small-sample pre-training, a prior feature library is established by coupling the descriptor with its parameters. Moreover, weld area acquisition is formulated as an active exploration problem in point clouds space. A reinforcement learning-based point clouds exploration and active recognition (PAEAR) method is proposed with a translation-scaling multi-level action strategy. This design enables identification and segmentation of regional point clouds for different welding seam categories in part-level scenes. Experimental results show that, across 2 simulated and 2 real-world scenarios, PAEAR achieves an mIoU of 86.86% using only 4 labeled samples, improving upon mainstream semantic segmentation methods by up to 81.22%, while reducing the area recognition error rate to 5.4%. The proposed method can serve as a front-end perception module for welding systems, providing reliable regional point clouds inputs for downstream welding seam localization and welding trajectory planning, while reducing dependence on manual teaching and prior models.

Intelligent driving for humanoid robots has emerged as a novel interdisciplinary domain integrating robotics, perception, cognitive decision-making, and autonomous driving. Its primary goal is to enable humanoid robots—with human-like morphology and motion capabilities—to perform or assist in vehicle driving tasks under complex and dynamic environments. Unlike conventional autonomous driving systems constrained by task-specific hardware and rule-based models, humanoid intelligent driving features enhanced adaptability in human–machine interaction and strong generalization across diverse driving scenarios. This paper reviews the conceptual foundations, technological framework, and recent advances in intelligent driving for humanoid robots. Core technologies, including multimodal perception fusion, human-like driving behavior modeling, hierarchical task planning, cooperative control mechanisms, and simulation-to-reality verification platforms, are systematically analyzed. Remaining challenges such as motion coordination, control precision, real-time responsiveness, and safety assurance are highlighted. Finally, the paper outlines prospective applications in intelligent transportation, special operations, and human–robot collaboration, and discusses future trends emphasizing the integration of artificial intelligence with robot dynamics modeling and adaptive multi-task learning. The review aims to provide systematic insights and guidance for advancing humanoid robot intelligent driving research and applications.

Mobile robot systems are increasingly being integrated into various scenarios.However,connecting these systems to networks exposes them to the risk of cyber-attacks,potentially leading to functional failures and system destabilization.This paper focuses on the stability of mobile robot systems under denial-of-service(DoS)attacks and proposes a robust control strategy based on H∞ control.In the strategy,the impact of DoS attacks on the system is first modeled as random packet loss with a Bernoulli distribution.When the mobile robot system is attacked,robust control is achieved by using feedback compensation based on a state observer.Sufficient conditions for exponential mean-square stability and H∞ control of the closed-loop system are derived using Lyapunov stability theory.n order to acquire the observer and controller gain matrices,which allow for robust system control and anti-interference effects,these criteria are used to solve linear matrix inequality constraints.Finally,through simulation experiments,it is demonstrated that the proposed strategy effectively mitigates the impact of DoS attacks on the system,ensuring the stability of the system.

Riveting is widely used as a primary connection method in the assembly of large aircraft structural components. During the drilling and riveting process, robotic systems typically rely on preassembled-hole positioning to improve hole-making accuracy. However, this process often requires compensation for deviations between the nominal 3D model and the actual workpiece, resulting in increased positioning time and reduced accuracy in subsequent operations. To address these challenges, this paper proposes a hybrid visual-ranging servo control method based on deep reinforcement learning (DRL-HVRS) for robotic motion planning in preassembled hole alignment tasks. The DRL-HVRS controller takes the observed image features and laser-ranging features as inputs and employs the TD3 algorithm to dynamically optimize the key parameters of the HVRS, thereby generating an effective motion trajectory for the robot. Finally, a robotic simulation environment was developed in CoppeliaSim for training and evaluation. Compared with existing PBVS and IBVS methods, this work is the first to integrate TD3 deep reinforcement learning with a hybrid visual-ranging servo model, enabling adaptive optimization of weighting and velocity factors, thereby improving the positioning success rate, convergence speed, and generalization performance. Experimental results on rivet hole insertion tasks demonstrate the effectiveness and generalizability of the proposed approach. In 50 random trials on a 15-mm-thick workpiece, the proposed method achieved a positioning success rate of 82.0

Existing obstacle avoidance algorithms mostly treat dynamic obstacles as short-term static ones without considering their motion states. This limitation results in a great loss of computational resources and obstacle avoidance effectiveness for the robots. A time-aligned prediction (TAP) method is proposed in this paper. This method achieves the prediction of obstacle movements by incorporating time-aligned prediction into dynamic window approach (DWA) method. Firstly, the TAP method predicts the stable motion of obstacles within a near-term time window via extended Kalman filter. Time stamps are added to the future motion trajectories of robots and obstacles to achieve time-aligned prediction. Then, the collision detection module is proposed to use the same time stamps for calculating the distance between the robot and obstacles. Finally, the TAP method integrates time-aligned prediction into the optimal trajectory objective function to refine the evaluation metrics of robot motion trajectories. In 500 random obstacle avoidance experiments, compared with DWA, the TAP method reduced the average time to avoid ten dynamic obstacles and reach the destination by 12.23

In addressing the coordinated control problem of linear MASs (multi-agent systems) in a dynamic task scenario, this paper proposes a distributed coordination method based on task consensus. Firstly, a mapping between tasks and system states is established to design a task representation method for MASs under distributed directed topologies. An event-triggered approach is designed for task switching in a dynamic task scenario to alleviate communication burden. Secondly, a definition of task consensus is introduced to address the different expected states of system agents under multiple tasks. A coordination control protocol based on task consensus is proposed, utilizing the states and task information of neighboring agents. The stability of this coordination control protocol is theoretically proven, establishing sufficient conditions for the system to achieve task consensus. Lastly, numerical simulations are conducted on a second-order linear MAS with a flocking prey task. Experimental results demonstrate that the proposed task-consensus-based coordination control method achieves stable convergence of task errors in MASs, presenting a novel approach for the coordinated control of MASs with dynamic tasks.

Admittance control is an important method for providing collaborative robots with precise manipulation and flexible contact behavior in industrial settings that often involve physical interaction. However, too rigid or high-frequency interactions by non-specialists will jeopardise the stability of the system. To address this issue, this research presents a novel admittance control framework for collaborative robots to detect oscillatory states and maintain stability by adjusting the controller parameters. In particular, a recursive haptic stability observer is designed to provide a quantitative assessment of the system stability, while a variable admittance controller based on model predictive control is constructed for optimal tuning of stability and flexibility to meet the requirements of a variable task. The effectiveness of the present algorithm is verified in experiments simulating two industrial tasks conducted on the AUBO I5 collaborative robot with 23 volunteers. In addition, the algorithm is tested for application in real collaborative tasks.

Robotic grinding has become an effective method for improving the surface quality of aeroengine blades. To address the asymmetric relationship between the surface roughness and the material removal rate during grinding of these types of blades, a multiobjective optimization method for the robotic grinding process parameters of aeroengine blades was proposed based on an integration of the Fuzzy Analytic Network Process (FANP) and a Gray Relational Analysis (GRA). First, the FANP assigned different Intuitionistic Trapezoidal Fuzzy Numbers (ITrFNs) according to the importance of the grinding parameters, which quantified the uncertainties in their relationships. Second, GRA was employed to address the mapping relationship between multiple grinding objectives and the process parameters, converting the multiobjective optimization problem into a single-objective problem. Then, through a mean analysis, the combination of process parameters that had the greatest impact on the multiobjective quality indicators was identified. Finally, an experimental validation revealed that the surface roughness was reduced by 7.76%, the material removal rate was increased by 22.72%, and the gray relational degree was increased by 7.18%, achieving the goal of simultaneously reducing the surface roughness and increasing the material removal rate.

The ability to rapidly and accurately estimate external forces applied to collaborative robots is crucial for ensuring safe physical human-robot interaction and minimizing injury risk from potential collisions. In this research, a novel finite-time observation method for sensorless collaborative robots to estimate time-varying external forces is put forward. First, a generalized finite-time observer is constructed by integrating generalized momentum dynamics and adding fractional power terms into the classical higher-order finite-time method to achieve the estimation of time-varying external forces using joint control torque. A performance analysis approach is introduced to assess this higher-order finite-time observer with mixed fractional power terms, demonstrating optimization of dynamic response speed and steady-state robustness. Furthermore, a Lyapunov-based recursive verification is provided to confirm the finite-time convergence and convergence bounds of this generalized observer. Finally, experiments are conducted on the AUBO I5 robot in two practical scenarios. The results demonstrate that the proposed observer significantly improves the detection speed of collision force by more than 5%, while simultaneously reducing the estimation error by over 4% in comparison to those of the existing finite-time observers. Moreover, the estimation error of time-varying interaction force is reduced by over 23% in comparison to existing ones. Furthermore, the collision detection and safe response experiments of a real collaborative robot body and end-effector have been successfully accomplished.

A collaborative robot is a type of robot designed to work alongside humans and accomplish tasks in collaboration with them. Executing grasping tasks within unstructured dynamic environments by collaborative robot constitutes a challenge that often requires a combination of visual perception and grasping strategies. This paper proposes a strategy based on iterative comparison method, which is intended to track and grasp dynamic targets with an arm-hand system. Combined with our previous work: real-time object recognition positioning system, the strategy proposed in this paper can track and grasp moving objects. We demonstrate how our arm-hand system--the AUBO i3 robotic arm equipped with the Inspire RH56DFX-2R Hand--could track and grasp different color yogurt bottles. The experiment showing that the average grasping success rate on different objects is 95.6 %, validated the effectiveness of the proposed strategy.

When robots replicate human actions in peg-in-hole assembly tasks, such as USB Type-A insertion and removal, the complexity of the process and frequent obstructions from the inner walls make it difficult for robots to handle collisions or avoid jamming. These difficulties contribute to a low success rate in assembly. This paper proposes a vision-guided reinforcement learning pre-assembly combined with tactile feedback-based pose estimation adjustment method for peg-in-hole assembly, achieving significant improvement in success rates for complex assembly tasks. First, during the pretraining process of reinforcement learning, high-reward sample data is collected, and a behaviour cloning (BC) algorithm is constructed based on sample data structure. The network is pretrained as a policy regression layer. Under sparse reward conditions, outputs of the twin delayed deep deterministic policy gradient (TD3) network and the BC network are combined to improve training stability and accelerate convergence, enhancing the efficiency of vision-based assembly. Then, to address the instability caused by collisions with the inner and outer walls of the hole when vision-based assembly remains incomplete, an in-hand pose estimation algorithm based on the Gelsight visuotactile sensor is integrated. This algorithm facilitates real-time adjustments to the position of the robot’s end-effector, improving the likelihood of successful peg-in-hole assembly. Finally, to validate the effectiveness of the proposed method, experiments were conducted using the V-REP simulation platform and the real Franka robot platform. In the experiments, success rates of 90-93% and 80-85%, respectively, were achieved.

Traditional active disturbance rejection control (ADRC) speed control systems for permanent magnet synchronous motors (PMSMs) often suffer from a large number of adjustable parameters and complex tuning. To address this issue, an improved super-twisting sliding mode algorithm is designed and combined with linear active disturbance rejection control (LADRC). This results in the development of a super-twisting sliding mode ADRC speed controller for PMSM speed control systems. The tracking differentiator and linear state error feedback control rate in ADRC are optimized to enhance the system's rapid performance and simplify parameter tuning. Second, in order to realize the position sensorless control, the back-electromotive force in the motor equation was incorporated into the unknown disturbance and estimated by a linear extended state observer (LESO), and finally, the speed and rotor position information were extracted from the estimated back-electromotive force. Simulation results validate the effectiveness of this method.

One of such important courses in the curriculum of the electromechanical programs worldwide is Embedded systems that connect the theoretical background with the practical aspects of engineering. In response to the limitations of traditional teaching methods related to embedded systems courses, this study proposes a mobile robot platform-based integration model for science and education teaching reform. In this study, we introduce a mobile robot experimental development course based on this model, realizing an integrated teaching model that integrates theoretical teaching and practical operation, modular teaching design, project-based learning, undertaking research competitions, and completing actual engineering projects. Consequently, students' practical operation ability, innovation consciousness and overall application ability have been greatly promoted. This work not only stimulates students interest to learning, but also plays a positive role in enhancing their education outcomes, and have a wide and solid foundation for their future study and work.

The magnetic actuated flexible-joint robotic surgery (MAFRS) camera system enhances laparoscopic surgeries by extending operational periods, achieved through the elimination of onboard motors. However, current methods face challenges in providing precise tilt motion control due to the variability in abdominal environments and the complexity of magnetic field interactions. To overcome these challenges, we propose a virtual muscle-model-modified reinforcement learning (RL) approach. This approach employs the deep deterministic policy gradient algorithm, optimized for continuous action spaces, thereby improving system robustness and response to nonlinear dynamics. The virtual muscle concept, drawing inspiration from human musculature, is integrated to mitigate camera chattering within the RL framework. Our system demonstrates exceptional control precision across various abdominal wall thicknesses, maintaining accuracy within 0.2(degrees)-a value approaching the resolution limit of our sensors. This level of precision signifies a significant advancement in laparoscopic robotic technology.

Robotic joint control in systems with harmonic reducers encounters challenges in dynamic torque scenarios that outstrip the capabilities of traditional proportional-integral (PI) controllers. Although effective in numerous situations, PI controllers may fall short in ensuring torque robustness and position accuracy under dynamic torque disturbances. This paper introduces a torque-robust dual-stage predictive control frame-work (TRDS-PCF), leveraging model predictive control (MPC) to address these challenges. The TRDS-PCF refines conventional control methods by incorporating a dual-layer predictive strategy alongside a harmonic reducer compensation module, significantly enhancing control accuracy and responsiveness. Validation through simulation studies demonstrates the TRDS-PCF's superior performance, evidencing substantial reductions in adjustment times without overshoot for no-load scenarios and sustained robustness under variable torque conditions. This development highlights the TRDS-PCF's potential to improve the performance and reliability of robotic systems substantially.