This article proposes a self-learning nonsingular terminal sliding mode control method for parallel robots with joint couplings considered. Based upon the nonlinear dynamical model, a nonsingular fast terminal sliding mode controller is developed by designing the switching control term matrix and the exponential reaching term matrix, which considers the coupling effects through the nondiagonal elements of these two gain matrices. The research findings indicate that when the gain matrix is of full rank, the number of gain parameters exceeds the number of conditional equations obtained through stability analysis. This results in the inability to obtain the value range of each gain parameter. Hence, the soft actor-critic algorithm is employed for tuning the gain parameters by using stability analysis constraints as action marginal conditions. Through experiments performed on the 3-degrees of freedom (DOF) parallel mechanism of a 6-DOF hybrid robot, the effectiveness of the proposed control method in enhancing joint tracking accuracy is verified. When the mechanism operates at high speed, the rms values of joint tracking errors are reduced by over 20% compared to analogous control methods that do not account for joint coupling.

The joint frictions will increase the pose error and trajectory deviation of a robot. Taking the TriMule-200 hybrid robot as an example, this paper proposes a dynamic modeling method considering both the active and passive joint frictions, along with a friction parameter identification method based on the Gaussian Quantum Particle Swarm Optimization (GQPSO) algorithm. The former integrates the Newton-Euler method with screw theory to formulate rigid-body dynamic equations for each component, leading to a linear equation system with driving forces/torques and joint constraint forces/moments as variables. A stepwise solution strategy is implemented to compute the driving forces/torques that considering joint frictions modeled by the Stribeck friction model. The latter adopts the global sensitivity analysis method based on variance and covariance decomposition (VCD-GSA) to identify highly sensitive friction parameters, which are then estimated using the GQPSO algorithm by minimizing the mean square error between the predicted and measured driving torques. The experimental results show that the predicted accuracy (R2) of the driving torques exceeds 96.63 % across both the identification and verification trajectories, thereby verifying the effectiveness of the proposed method.

This article proposes a self-learning nonsingular terminal sliding mode control method for parallel robots with joint couplings considered. Based upon the nonlinear dynamical model, a nonsingular fast terminal sliding mode controller is developed by designing the switching control term matrix and the exponential reaching term matrix, which considers the coupling effects through the nondiagonal elements of these two gain matrices. The research findings indicate that when the gain matrix is of full rank, the number of gain parameters exceeds the number of conditional equations obtained through stability analysis. This results in the inability to obtain the value range of each gain parameter. Hence, the soft actor-critic algorithm is employed for tuning the gain parameters by using stability analysis constraints as action marginal conditions. Through experiments performed on the 3-degrees of freedom (DOF) parallel mechanism of a 6-DOF hybrid robot, the effectiveness of the proposed control method in enhancing joint tracking accuracy is verified. When the mechanism operates at high speed, the rms values of joint tracking errors are reduced by over 20% compared to analogous control methods that do not account for joint coupling.

An analytical approach for elastodynamic modeling of a 2-limb 4-DOF (3T1R) high-speed parallel robot is proposed. Based on screw theory combined with structural dynamics, the kinetic energy and potential energy of the robot are derived, and the expressions for the first 6 natural frequencies of the mechanism are obtained. A full finite element analysis of the robot is carried out using finite element software, and the results show that the low-order natural frequencies are in good agreement with those from the analytical method. This indicates that the analytical elastodynamic model can estimate the low-order dynamic behaviors over the entire workspace in a very effective and accurate manner.

Discontinuous corner smoothing in machining paths and excessive rapid traverse paths degrade both the machining quality and efficiency of hybrid robots. Therefore, two algorithms are proposed for dealing with the trajectory planning of a 6-axis hybrid machining robot within a single machining zone and across two machining zones, respectively. The constraints on the curvatures and position errors of smoothing path segments are considered in corner smoothing planning within a single machining zone, while the constraints on the curvatures and obstacle geometries are employed in planning obstacle-avoidance paths across two machining zones. Both of the above optimization problems are solved by adjusting the weights of spline curves through the Quantumbehaved particle swarm optimization (QPSO) algorithm combined with the greedy algorithm. Furthermore, the time allocation is optimized using the QPSO algorithm combined with the moving window planning method to enhance the machining efficiency of toolpaths. The comparative simulation results within a single zone indicate that, under identical position error constraints, the proposed method achieves a maximum curvature reduction of 27.29 % relative to the B-spline method, and reduces computational time by 48.35 % and 53.31 % compared to the PSO algorithm and the genetic algorithm, respectively. The simulation results across two machining zones demonstrate that the proposed method is capable of generating curvature-optimal obstacleavoidance toolpaths for obstacles with varying geometries. Additionally, the simulation results of butterfly and maple leaf contours show that the proposed method reduces machining time by 52.59 % and 40.92 % compared to the Bezier curve method and the Clothoid spline method, respectively. The experimental results show that the predicted curvatures are in close agreement with the measured ones along the butterfly and maple leaf contours, with a maximum error of 2.99 %. Furthermore, the measured tracking errors of the actuated joints are maintained within +0.028 mm and +1.3 x 10-4 rad for the parallel mechanism and serial wrist, respectively. These results fully demonstrate the effectiveness of the proposed method in enhancing machining efficiency and motion smoothness in the planning of curvature-optimal corner smoothing toolpaths.

To address the issues of excessive parameters and a long training process, this study takes the TriMule-200 robot as an example and proposes a model-based and data-driven dynamic parameter identification approach. The approach designs a two-phase strategy of "source training phase & target training phase": firstly, the rigid body dynamic model of the TriMule robot is established, and then 40 key dynamic parameters are selected by the Sobol global sensitivity analysis method. Subsequently, a neural network based on a Transformer encoder is constructed. In the source training phase, parameter sensitivity weights are introduced into the loss function, while in the target training phase, fine adjustment is made based on the error between the predicted torque of the model and the actual torque. Finally, a general identification software is developed based on this approach, which realizes the whole process modular integration from CAD model analysis, excitation trajectory design to fast parameter identification. The experiments on hybrid robot show that the predicted RMS error of the torque of the identified parameters is 3.16 & times;10(-2)N ext {N}\cdot ext {m}$ , which is 17% lower than the calculated value based on CAD parameters, and the correlation coefficient is increased to 90.12%. In addition, the identification software can quickly adapt to similar hybrid robots with different sizes or varying loads, and online update of parameters can be achieved only by target training, significantly improving the accuracy of identification and engineering applicability. Note to Practitioners-This work tackles the challenge of rapidly updating robot dynamic parameters when the end-effector payload or operating conditions change. The method implements a two-stage strategy of "source training and target training". Parameter identification is rapidly completed by conducting online target training based on the pre-completed offline source training. This approach significantly reduces reliance on prior expert knowledge for tuning. The framework is scalable and can be adapted to TriMule hybrid robots of different sizes and models by updating their CAD and dynamic models. Furthermore, the identified parameters provide valuable feedback that can be used to optimize robot design, manufacturing, and assembly processes.

Stiffness determines the task quality and machining accuracy of machining robots. Taking the TriMule hybrid robot as the research object, this paper proposes a stiffness modeling method considering servo characteristics and an active stiffness adjustment method based on deep learning. The former method constructs a transfer function with load disturbance torque as input and the angular displacement variation of actuated joints as output. On this basis, a semianalytical stiffness model of the robot is established with the servo elasticity of actuated joints. The latter method constructs a deep learning network that takes the current configuration and desired stiffness of the tool head as inputs and outputs the servo gain parameters. Training data for the network are obtained from the established electromechanical coupling semi-analytical stiffness model. Simulation and experimental results show that the prediction error for stiffness is within 5% under servo system operation and the active stiffness adjustment ensures that stiffness deviation between configurations is less than 5%, and the maximum reduction is about 10% compared with that before adjustment. The results demonstrate the effectiveness of the proposed method.

Direction-of-arrival (DOA) estimation for low-elevation targets plays a significant role in multiple-input-multiple-output (MIMO) radar applications. However, in realistic irregular terrain environments, conventional DOA estimation algorithms suffer from a substantial loss of accuracy due to severe multipath effects and the difficulty of accurately modeling such propagation phenomena. To address these challenges, a hierarchical optimization-based DOA estimation method is proposed for low-elevation targets. First, a multipath signal model adapted to irregular terrain is developed, where nonideal path reflections induced by terrain variation are modeled as stochastic perturbations. Then, high-order singular value decomposition (HOSVD) is applied to extract the tensor signal subspace from the element space, which is subsequently projected onto a lower dimensional beamspace for efficient processing. On this basis, a two-level hierarchical optimization framework is proposed. In the first stage, a sparse Bayesian learning (SBL) algorithm in the beamspace provides coarse estimates of the target elevation angle and multipath angle of arrival. These estimates are subsequently refined in the second stage through a joint optimization procedure that combines alternating iteration with the quasi-Newton Broyden-Fletcher-Goldfarb-Shanno (BFGS) method. Simulation and experimental results demonstrate that the proposed method exhibits the excellent estimation accuracy and robustness under irregular terrain conditions, while maintaining relatively low computational complexity.

The feedforward compensation based on friction model is an effective way to reduce the influence of friction on mechanical system. This paper presents an approach for friction compensation by considering the inertia change of the actuated joints of a 6-axis hybrid robot named TriMule. First, the tracking errors of each actuated joint of the parallel mechanism at high and low inertia configurations are displayed. Then, a compensation approach considering inertia change is developed by introducing the inertia term of the joint driving force into the traditional Stribeck model as thrust. Furthermore, the improved approach based on radial basis function interpolation without using the dynamic model is proposed, which has a simple calculation expression and acceptable compensation effect. Experiment results on a prototype machine show that compared to otherwise similar compensation method not considering joint inertia change, the tracking accuracy can be improved up to 25.73% at high inertia configurations.

To address the challenges of low efficiency and inconsistent quality in the finishing of concrete shield segments, this paper proposes an adaptive trajectory planning method for a robotic system utilizing point cloud data. The system integrates an automated guided vehicle (AGV), a six-degree-of-freedom serial manipulator, and a 3D vision system to create an intelligent finishing robot. A "rectangular" offline programming trajectory is employed, coupled with multi-coordinate system transformations for precise path mapping. A 3D camera captures the segment's surface point cloud, which is subsequently registered, fused, and analyzed by a neural network to identify surface irregularities and compute a look-ahead tilt angle for adaptive trajectory compensation. Experimental results demonstrate that our method significantly enhances finishing uniformity and surface quality, offering a viable technical solution for the automated finishing of complex curved components.

The Ice, Cloud, and land Elevation Satellite-2 has provided unprecedented global surface elevation measurements through photon-counting Lidar (Light detection and ranging), yet its low signal-to-noise ratio (SNR) poses significant challenges for denoising algorithms. Existing methods, relying on fixed parameters, struggle to adapt to dynamic noise distribution in rugged mountain regions where signal and noise change rapidly. This study proposes an adaptive Bayesian denoising algorithm integrating minimum spanning tree (MST) -based slope estimation and probabilistic parameter optimization. First, a simulation framework based on ATL03 data generates point clouds with ground truth labels under varying SNRs, achieving correlation coefficients > 0.9 between simulated and measured distributions. The algorithm then extracts surface profiles via MST and coarse filtering, fits slopes with >0.9 correlation to reference data, and derives the probability distribution function (PDF) of neighborhood photon counts. Bayesian estimation dynamically selects optimal clustering parameters (search radius and threshold), achieving F-scores > 0.9 even at extremely low SNR (1 photon/10 MHz noise). Validation against three benchmark algorithms (OPTICS, quadtree, DRAGANN) on simulated and ATL03 datasets demonstrates superior performance in mountainous terrain, with precision and recall improvements of 10–20% under high noise conditions. This work provides a robust framework for adaptive parameter selection in low-SNR photon-counting Lidar applications.

As the main driving mechanism of a hybrid robot, the parallel mechanism is a nonlinear time-varying system. The load inertia of its actuated joints changes with the configuration of the robot. Analyzing and fitting the inertia variation is of great significance to the design and control of hybrid robots. By taking a hybrid robot named TriMule as an example, the variation of load inertia of each actuated joint in the whole workspace is first revealed based on the dynamic analyses of the robot. Then two methods based on the circular and elliptical membership are proposed to calculate fitted inertia over the whole workspace using inertia information at a few configurations. Finally, the fitting methods of the two membership functions are compared and discussed. The results show that the maximum value and global mean value of the fitting error of the elliptical membership method are 39.18% (51.23%) and 65.79% (81.25%) for actuated joint-1 (joint-2 and joint-3) lower than those of circular membership method, which promise a better global fitting accuracy. The proposed method can be used to estimate the joint load inertia or other control variables affected by inertia in a quick manner, allowing the algorithm to be easily integrated into the robot control system.

Focusing on the pick-and-place operation in the electronic, packaging, and other light industries, this paper proposed a new type 3T1R (T-Translation, R-Rotation) high-speed parallel robot with two limbs. Compared with other 4-DOF high-speed parallel robots with four limbs, its structure is simpler while the working space is larger, such that the potential for cheaper usage in the industry is achieved. The robot's kinematic analysis and trajectory planning, including key aspects, are also outlined here. Firstly, the robot's closed-loop constraint equation is established using the vector chain method. The displacement, velocity, and acceleration models are derived, and the Jacobian matrix is obtained. On this basis, the gate shape trajectory is planned using the 3-4-5 polynomial motion rule, such that the displacement, velocity, and acceleration of the driving joint are obtained using the kinematic models above. Ultimately, the accuracy of the kinematic models is confirmed by comparing the theoretical outcomes with the simulations carried out in SolidWorks. The study presented in this paper establishes a firm basis for the dimensional integration and the creation of a physical model of the robot.

Considering that joint inertial effects of the 3-DOF parallel mechanism within a hybrid robot vary with system configurations, a fuzzy-logic strategy is proposed for feedback and feedforward controller parameters tuning. This approach features the generation of a group of specific configurations representing different inertial levels via clustering analysis, and the creation of a membership function that matches the joint inertial distributions across the entire task workspace. Merging these two threads allows the controller parameters at any arbitrary configuration to be estimated by taking the parameters off-line tuned at the specific configurations as the inputs of the membership function. Both simulation and experimental results on a prototype machine show that only eight specific configurations are required, where the controller parameters of three actuated joints of the 3-DOF parallel mechanism need to be tuned for implementing the fuzzy control strategy. It also concludes that it is significant to use the proposed strategy when the robot moves in the region where the gradient of joint inertial effects varies sharply.

Industrial robots are widely used in industrial production because of their high work efficiency and high flexibility, but their low tracking accuracy makes them unable to be applied on a large scale in high-precision manufacturing. Considering the modeling error of robot, the problem of low tracking accuracy should be solved by combining model-based methods and data-based methods. Thus, a model-data compound approach is proposed for compensating robot tracking error. This method comprehensively considers two error compensation methods of model-driven and data-driven, which constructed the input matrix by using the model-based information and the data-based information. The former includes the desired position, velocity, acceleration, torque and friction torque; and the latter includes the actual torque. For the purpose of predicting the correlation between the joint error and the historical joint motion state, the historical information of the joint motion is also added in the input matrix for improving the error prediction accuracy. Then, a convolutional neural network prediction model is constructed for predicting and compensating the tracking error of each actuated joint. The experimental results show that the proposed method has good prediction accuracy and compensation effect. Compared to joint control without using the proposed method, the root mean square of joint tracking errors was reduced by up to more than 69%.

Drilling, one of the most used machining processes, has wide application in different industrial fields. Monitoring the system health and operation status of the drilling process is essential for maintaining production efficiency. In this study, a convolutional neural network (CNN), a deep-learning method, is applied to the defect diagnosis of drill bits. Four drill bits with different health conditions were used to drill holes in an aluminum block, and a vibration sensor collected the signals. Vibration spectrograms generated using short-time Fourier transform were applied to a 2D CNN algorithm, and they were then reconstructed into a 1D data set and applied to a 1D CNN algorithm. The input data size was reduced significantly compared to the raw vibration data after the data-reconstruction process. As a result, the 2D CNN process shows a diagnostic accuracy of 97.33%. On the other hand, the 1D CNN provides a diagnostic accuracy of 96.6%, but it only requires 2/3 of the computational time required by the 2D CNN.

With the wide deployment of digital image capturing equipment, the need of denoising to produce a crystal clear image from noisy capture environment has become indispensable. In this article, a novel type-2 fuzzy-based filter is proposed for denoising images corrupted by impulse noise, especially for the high density of salt-and-pepper noise. It operates two stages, namely, type-2 fuzzy identifier and matrix completion denoiser. In the proposed method, the type-2 fuzzy identifier is first employed to identify and trim the entries contaminated by impulse noise in the data matrix from fuzzy system. Then, the trimmed data matrix is utilized to retrieve the noiseless data matrix with the matrix completion technology. Herein, a novel matrix completion technique is developed without $a$ $priori$ rank information compared to its counterparts. Simulation results are presented, which vividly show the denoised images obtained by the proposed method can achieve crystal clear image with strong structural integrity, and are showing good performance in terms of peak signal-to-noise ratio.

Emotional Internet of Things (EmIoT), which provides Internet of Things (IoT) devices cognitive and socialization capabilities, has been regarded as a future direction to improve users’ experiences. With the development of intelligent techniques, the requirement of EmIoT is not only sensing the users’ emotional states but also providing emotional feedbacks. Human–computer interaction has been studied to achieve speech interaction with IoT devices. The recent advances in neural text-to-speech (TTS) have made “human parity” synthesized speech possible for IoT-enabled human–computer interaction. Furthermore, emotion control can be achieved by using the emotional codes in a unified model, referred to as emotional TTS (or ETTS for short). Such ETTS models have achieved promising emotional expressiveness using large-scale emotion-annotated English data set; however, they are not practical in IoT environments with other mainstream languages, especially for Chinese. In fact, the limited available large-scale emotion-annotated data set is challenging the development of Chinese ETTS. To address that we propose a multistage deep transfer learning scheme to design a high-quality Chinese ETTS system under a small-scale training corpus to achieve EmIoT in Mandarin environments. In this scheme, the pretrained knowledge from the former stages corresponding to a large-scale neutral English and a medium-scale emotional English corpora is transferred to a Mandarin ETTS model. Thereby, the trained model can achieve high-quality emotional speech with limited available emotional corpus, which is able to serve various EmIoT-oriented applications. The experiments have been conducted to demonstrate the effectiveness and superiority of the proposed model as compared to other counterparts in terms of naturalness and emotional expressiveness. We refer readers to visit our demo Webpage1 enjoy the synthesized speech samples.

Shortening the packet length has been a consensus in wireless network design for supporting the ultra-low latency Internet of Things (IoT) applications. Yet, with short-packet transmission, the rate loss would occur, which further depends on the blocklength, making the network optimization notoriously difficult, especially for random access networks. This paper focuses on the representative random access network, i.e., Aloha, with short packet transmission, namely, short-packet Aloha. Specifically, we aim to optimize the sum rate and access delay of short-packet Aloha. By deriving the probability of successful transmissions of packets, both the network sum rate and the probability generating function of access delay are obtained as explicit functions of key system parameters. The maximum sum rate and the minimum mean access delay are further derived by jointly tuning the packet transmission probability and the blocklength of packets. The effect of system parameters on the optimal sum rate and access delay performance is investigated. It is shown that the maximum sum rate is insensitive to the retry limit $M$ , while deteriorates as the information bits per packet $k$ decreases. In contrast, the optimal delay performance can be improved with a small $M$ or $k$ . The reliability performance is also evaluated and shown to be enhanced with a large retry limit $M$ . The analysis sheds important light on the access design of practical short-packet Aloha networks. By taking LTE-M as an example, it is found that to improve access delay performance, the information bits per packet $k$ should not exceed an upperbound, which polynomially decreases as the network size increases.