One of the key technologies for autonomous driving is localization. Precise localization is inseparable from high-precision maps. Simultaneous Localization And Mapping (SLAM) is technologies for building maps and locating in unknown environments. One of the mainstream directions of current autonomous driving research is how to carry out efficient closed-loop processing over long periods of time, eliminating accumulated error in the odometer. Improving the accuracy of the map and ensuring the accuracy of localization is crucial for ensuring the safety of autonomous driving. Therefore, this paper proposes a featherweight and high-precision method for multi-sensor fusion of laser and GPS for loop detection. The closed-loop results are optimized together with IMU to improve the mapping accuracy.
This paper proposes a teacher-student network for industrial anomaly detection tasks using knowledge distillation techniques, where the feature representation capabilities of the teacher network model are utilized to guide the student model in efficiently identifying anomalous images. In detail, the VGG model is introduced as the backbone network for anomaly detection. Meanwhile, a residual attention mechanism is proposed to improve the feature output capability of the student network. The student network is able to learn good feature extraction abilities due to the robust feature representation of the teacher network. At the same time, the residual attention mechanism enhances the student network's ability to characterize the output of normal data features, amplifying the feature differences between the student and teacher networks for abnormal data. In summary, this paper addresses the problem of difficult detection of small targets in anomaly detection. Experimental validation has been performed on the industrial anomaly detection dataset MVTec-AD, which ultimately achieved 96.70% and 97.94% results for detecting and localizing, respectively. Compared with the other anomaly detection methods, the method proposed in this paper achieves a significant performance improvement.
In this paper, a novel adaptive terminal sliding-mode control (ATSMC) method is developed for road vehicles with uncertain dynamics. It is shown that the designed adaptive laws can recursively update the controller parameters based on the estimation values of the unknown parameter bound in the Lyapunov sense, and then the lateral position error can be driven to zero in a finite time. In comparison to the conventional sliding-mode-based vehicle lateral control method, the proposed ATSMC not only ensures fast error convergence, but also eliminates the requirement for known bounded information in controller design. Simulation results are presented to demonstrate the good performance of the proposed ATSMC scheme.
With the development of deep learning, object detection technology has been widely used in various areas. However, the large size of the model and the high computation cost limit the deployment on the devices of the robot. To address these problems, this study proposes a slight cross-stage object detection model YOLO-OR for the dish recycling robot. In detail, the cross-stage octave convolution combines with the residual structure to build a novel convolution block ORBlock, improving feature extraction capability through cross-stage convolution operations. Moreover, ORBlock significantly reduces the parameters via splitting the traditional convolution into two-dimension. In particular, the four outputs of the backbone’s last two ORBlocks are merged and transmitted to the YOLO neck for feature fusion. Based on the Dish-20 dataset, YOLO-OR achieves excellent performance with 99.56% mAP, 189.42 FPS, 4.64 G FLOPs and 3.29 MB parameters. Meanwhile, the performance of YOLO-OR outperforms other state-of-the-art YOLO models. These demonstrate the effectiveness of the proposal.
In recent years, there has been an increase in the number of patients with motor dysfunction due to neurological disorders such as stroke and spinal cord injury. Existing rehabilitation medical resources are unable to provide one-on-one rehabilitation treatment for patients. Lower limb rehabilitation robots (LLRRs) are used as assistants to rehabilitation physicians to help patients with rehabilitation training by simulating the walking gait of a normal person. Each patient’s individual degree of impairment varies from person to person, and the parameters of the lower extremity differ, which leads to different rehabilitation training programs used by the patient. Therefore, how to tailor the reference trajectory for patients according to their physical and gait parameters becomes the focus of research at this stage. In this paper, a gait generation model based on GLS network is designed to provide continuous personalized gait trajectories for patients. The GLS network is composed of an LSTM (Long Short-Term Memory) network based on the gap loss function and a sparrow search algorithm (SSA).The gap loss is presented based on the difference between the initial angle and the end joint angle of each cycle. The experimental results are analyzed by mean absolute deviation (MAD). The experimental results show that the gait trajectory generated by the GLS network reduces the experimental error by 16.9 % compared to the trajectory generated by the conventional algorithm (PSO). Therefore, our proposed GLS network-based generation of personalized continuous gait trajectories generates gait trajectories that are more similar to the actual gait trajectories.
Wireless power transmission (WPT) is a method of transmitting power wirelessly and has attracted much attention in recent years because of its advantages such as the absence of connections and the ability to ignore obstacles. The magnetic field resonant coupling method that we have adopted is one of the most robust among WPT methods in terms of misalignment between coils, and is capable of transmitting large amounts of power. However, WPT has a problem that transmission efficiency and transmission power decrease due to parameter fluctuations such as load changes. Therefore, we propose a control scheme for time-varying output load resistance. In this paper, we model a WPT system and show how to find solutions to the second-order differential equation for output voltage under time-varying loads. Furthermore, a control system that simultaneously controls power and transmission efficiency under time-varying load is presented, and finally, simulation results and experimental results using actual equipment are presented.
In this study, a device for writing letters is developed and designed using a thin dielectric elastomer actuator with three degrees of freedom. The 3-DOF thin dielectric elastomer actuator is capable of being driven in three directions (X, Y, and Z). Conventionally, the reserve elongation was adjusted with a weight, which is very inconvenient. A force-measured pre-stretch mechanism is developed to apply pre-stretch to the 3-DOF thin dielectric elastomer actuator. The pre-stretch mechanism with force measurement can easily apply pre-stretch and monitor it in real-time. The effectiveness of the device is verified through experiments on actual equipment, where a control system with feedback control is designed.
In this paper, a new parameter estimation method using system steady-state sinusoidal output (SSSO) measurements is proposed for a broad class of second-order linear systems. It is shown that when the input to the system is chosen as the linear combination of a group of sinusoids, the amplitudes of system SSSO components can be extracted using any fundamental period of system SSSO measurement data. Unknown system parameters can then be estimated with the SSSO amplitude information in the frequency domain. The advantages of the new parameter estimation method are three fold: (i) only a fundamental period of system SSSO measurement data is required; (ii) the estimation process has good performance against sensor noise; (iii) it provides a global optimization mechanism, in any fundamental period at steady-state, to accurately estimate unknown system parameters. Simulation examples are implemented to show the effectiveness of the proposed parameter estimation method.
This paper presents an evaluation of the Extreme Learning Machine (ELM), Kernel Extreme Learning Machine (KELM), and the Kernel Extreme Learning Machine optimized by the Whale Optimization Algorithm (WOA-KELM) in the context of fruit image classification. Fruits play a crucial role in everyday life, and the application of machine learning for fruit classification can significantly improve efficiency. The ELM, a single hidden layer feedforward neural network, is the foundation for this comparison. The KELM, an evolution of the ELM, incorporates a kernel function to map input data into a high-dimensional feature space, thereby enhancing performance. The development of WOA-KELM involves optimizing the regularization parameter in KELM and the width parameter of the radial basis function through the Whale Optimization Algorithm, further enhancing its capabilities. In classifying twelve types of fruit images, the ELM achieved a classification accuracy of 77%. In contrast, the WOA-KELM, as proposed in this paper, significantly outperforms with a remarkable classification accuracy of 92%.
Hand gesture recognition is a crucial aspect of human-machine interface (HMI) and assistive technology (AT) applications, allowing users to control devices and interfaces through intuitive and natural movements. This study proposed and compares two powerful machine learning (ML) methods, K-Nearest Neighbors (KNN) and fitcensemble (Ensemble of Learners for classification), for real-time hand gesture recognition (HGR) to control a powered wheelchair (PW). Using surface electromyography (sEMG) signals for HGR, this study first explores the KNN algorithm’s potential and demonstrates its exceptional offline accuracy of 99.74%. KNN leverages the similarity of nearby data points to classify hand gestures. Additionally, this study investigates the fitcensemble algorithm as an alternative approach for HGR. This ensemble learning method combines multiple decision trees for improving classification accuracy through diversity and ensemble aggregation. The result shows an impressive 97.68% offline accuracy for fitcensemble, making it promising for the real-time applications. In the real-time implementation and intuitive interface to control the PW by using HGR. This study demonstrates the seamless interaction between the HGR models and the wheelchair control, allowing users to maneuver the wheelchair in a user-friendly manner. Overall, this study contributes to the advancement of AT and HMI by showcasing the efficacy of KNN and fitcensemble for real-time HGR. The achieved accuracy and performance highlight the potential of these methods in real-world applications, promoting enhanced mobility and independence for individuals with limited physical abilities. The findings presented in this study provide valuable insights for researchers, developers, and practitioners working on gesture-based control systems and AT interfaces.
This paper studies the multi-sensor fusion positioning methods of Ultra-wideband (UWB), Global Navigation Satellite System (GNSS) and Inertial measurement unit (IMU), and constructs a seamless positioning system based on UWB/IMU/GNSS multi-sensor fusion. In this paper, the indoor location information provided by UWB is fused with IMU location data through extended Kalman filter, which solves the problems of signal loss, track interruption and jump in positioning information when unmanned systems enter indoor environment, and also improves the positioning accuracy of the overall system. In the experiment of the indoor and outdoor scenes of the unmanned vehicle system, the trajectory diagram of each sensor combination positioning mode is counted, and the error of positioning information is analyzed and studied. The results show that the overall positioning error of this method is small, it can provide stable and reliable whole-time positioning information, and achieve the high-precision seamless positioning required by unmanned systems.
This paper proposes a novel finite-time generalized proportional integral observer (FTGPIO) based a sliding mode control (SMC) scheme for the tracking control problem of high order uncertain systems subject to fast time-varying disturbances. For this purpose, the construction of the controller consists of two consecutive steps. First, the novel FTGPIO is designed to observe unmeasurable plant dynamics states and disturbance with its higher time derivatives in finite time rather than infinite time as in the standard GPIO. In the FTGPO estimator, the finite time convergence rate of estimations is well achieved, whereas the convergence rate of estimations by classical GPIO is asymptotic and slow. Secondly, on the basis of the finite and fast estimations, the SMC scheme is constructed in order to compensate estimation errors in both states and disturbance. Finally, the tracking efficiency is verified by carrying out many simulations on the application example of flexible joint robot (FJR). Besides, comparing the proposed method with the standard GPIO based SMC method is performed.
Generally, motion vector information primarily stems from moving objects, while static objects contribute minimally to the estimation task. Traditional approaches for motion vector estimation typically rely on scene flow methods that depend on deep models to extract features from individual points at a high cost. These methods then acquire flow information through complex matching mechanisms or feature decoding. Such approaches are computationally expensive and exhibit substantial latency. Moreover, they neglect the importance of motion objects in motion vector estimation and the interference from static objects. Therefore, this paper introduces a novel method that first performs point cloud motion segmentation and subsequently estimates motion vectors. This approach focuses on leveraging point cloud information annotated with motion objects to estimate three-dimensional scene flow more effectively. By employing motion segmentation, we can obtain annotations for moving objects, enabling greater emphasis on the estimation of motion vectors for more challenging cases. In our experiments conducted on the KITTI dataset, the proposed method demonstrates superior performance compared to existing scene flow estimation methods. Specifically, without considering motion segmentation errors, the error in the motion direction is only 0.0363 m/s, showcasing better performance. Additionally, our method achieves an error of 0.076 m for three-dimensional endpoint error (EPE3D), showcasing distinct advantages over current scene flow networks.
Chest X-ray images are critical in detecting diseases such as Covid-19, pneumonia, and lung opacity. This paper presents an approach to enhance the accuracy of chest X-ray image classification by implementing the Whale Optimization Algorithm (WOA) to the Kernel Extreme Learning Machine (KELM). The initial step involves applying the Singular Value Decomposition (SVD) method on chest X-ray images for efficient feature extraction. Following this, Principal Component Analysis (PCA) and Lasso regularization techniques are applied for dimensionality reduction and feature selection respectively, which significantly lowers the likelihood of overfitting. This paper significantly improves the classification accuracy by optimizing the regularization parameter in KELM and the width parameter in the Radial Basis Function using WOA. Experimental results underscore the superiority of the WOA-KELM, recording an accuracy rate of 94.9008% in a three-class (Covid-19, pneumonia, normal) X-ray image classification task, a drastic improvement over the conventional ELM (88%) and unoptimized KELM (68.7819%).
This research addresses the increasing demand for precise displacement control with larger ranges in piezoelectric actuators for positioning stages. To fulfill this requirement, a compliant piezoelectrically driven micro-positioning stage with a high amplification ratio is proposed. The structural design incorporates a twin-bridge type mechanism to amplify actuator displacement and enhance modal shape frequencies. Displacement transmissions and amplification mechanisms are paired with leaf flexure-based parallelogram mechanisms to linearize the output movement. Additionally, four major bridge mechanisms are arranged in parallel to improve the piezoelectric actuator’s output displacement. Computational optimization of the positioning stage’s architecture is performed to maximize output displacement and increase the modal shape frequencies. The impact of geometrical design parameters on the stage’s displacement is investigated through Ansys analysis. Experimental validation confirms that the positioning stage achieves a motion precision of 0.5 µm and an amplification ratio of 28.78, demonstrating its effectiveness for precise displacement applications.
This study discussed the nonlinear forced vibration control (NFVC) design for a vertical wing plate system with piezoelectric actuators. The nonlinearity mainly considered the actuator’s input-output hysteresis nonlinearity in this paper. Also, the Prandtl-Ishlinskii (P-I) model is used to describe nonlinearity from the actuator. In addition, the nonlinear system’s stability is ensured by operator theory. And, the NFVC performance is guaranteed by designed operators and estimation structure, respectively. Finally, the simulation case study showed the effectiveness of the constructed system.
Gait-based human identification (GHID) is a promising biometric technology that recognizes individuals based on their unique walking patterns. This paper explores the contribution of each single body position and the data fusion of multiple positions to GHID. We develop two models, namely temporal network (TNET) and spatial network (SNET), for extract temporal and spatial features in the GHID task. The experimental results on our collected multiple-body wearable data demonstrate that the Right Hip position achieves the highest accuracy of 96.45% in the single-position setting using TNET. Subsequently, the fusion of data from Right Hip with other body positions is explored. The combination of Right Hip and Right Elbow performs the best, achieving the two-position fusion setting with an accuracy of 99.11% based on TNET.
In this paper, we propose a new spacing policy that integrates the conventional CS (Constant Spacing) and CTH (Constant Time Headway) into one for a vehicle platoon’s cooperative adaptive cruise control (CACC). To achieve excellent vehicle distance control performance and string stability simultaneously, the bidirectional vehicle-to-vehicle communication topology is adopted such that each vehicle receives the positions, velocities, and accelerations from both the adjacent predecessor and successor vehicles, by onboard sensors and/or wireless communication. The vehicle dynamics is a second-order nonlinear dynamic motion equation and a local robust controller is designed for each vehicle. The string stability of the platoon and the robust stability of the overall cooperative control system are analyzed theoretically. And finally, simulation studies are performed to support the theoretical insights.
In recent years, the aging of skilled workers and the shortage of young workers in agriculture have limited harvest quality control, especially in fruit visual inspection. Therefore, with the technological development of computer versions, deep learning is attended for visual inspection. Deep learning-based anomaly detection models have attracted attention and are applied for visual inspection in the industry. However, few studies have performed experiments on crops in agriculture. In this paper, we propose defect detection method for apples as a representative example of fruits. We use PatchCore, which is one of the state-of-the-art anomaly detection methods, while changing feature extractor, which is trained by ImageNet. In experiment 1, we train the model with normal images, inference with the test images, and evaluate the model. According to the experiment1 results, WideResNet50 is the most effective feature extractor. Furthermore, it can be seen that the tip of the branch of the apple is erroneously detected as a defective portion. From experiment 1 result, in experiment 2, input data are preprocessed using YOLOv7 and OpenCV to remove the branch tip portion of an apple, and we evaluated in the same way as in experiment 1. According to the experiment2 results, we achieved 0.902 for AUROC and 0.889 for AUPR, proving the effectiveness of our proposal.f
The average age of agricultural workers is getting older caused of the declining birthrate and aging population. Though the labor shortage is becoming more serious, much labor is needed to check defects such as scratches and rot on apples before shipping. This research proposes a method to automatically inspect the appearance of deliverables by a deep learning-based object detection method. In detail, we use YOLO, which is used widely as an object detection model, deployed the attention module, changed the resolution of feature maps, and optimized the times of convolution. Our research aims to make higher accuracy and fewer parameter models. The model’s parameter is also improved from 5.05G to 0.73 G. Our proposal methods improve the detection score from 86.04% to 87.59%. The experimental results show the proposal helps automate the visual inspection of apples and alleviate labor shortages in the agricultural industry.
