Publications

Research on Behavioral Decision Reward Mechanism of Unmanned Swarm System in Confrontational Environment

A study of the appropriateness, geometry, size, and fillers boundary restraint on representative volume element in electrical simulation for carbon-based nanocomposites

Overview of Reinforcement Learning for Person Re-Identification

For intelligent surveillance, the issue of person re-identification has attracted extensive research interest due to its great academic value and broad application prospect. This issue aims to build the correspondence between human appearances across deployed non-overlapping cameras. However, the captured human images are inevitably influenced and deteriorated by various factors, such as illumination variations, pose changes, viewpoint alterations, partial occlusions, and surrounding clutters, which largely barricade the progress of re-identification performance. While, owing to the development and prosperity of reinforcement learning, researchers have made a big breakthrough in this issue. Motivated by the research breakthrough, this paper provides an overview of reinforcement learning for person re-identification. The contributions of this paper can mainly be summarized as follows: (a) introducing the background and challenges of person re-identification, and describing the framework and characteristics of general reinforcement learning methodology; (b) surveying the latest and most representative researches on reinforcement learning methods for this issue, and analyzing the mechanisms and essences of these methods; (c) clarifying the advantages of reinforcement learning, discussing its potential problems, proposing the conceptual schemes around it, and thus illuminating its future trend for this issue.

Dynamic Compensation for Two-Axis Robot Wrist Force Sensors

To improve the dynamic characteristic of two-axis force sensors, a dynamic compensation method is proposed. The two-axis force sensor system is assumed to be a first-order system. The operation frequency of the system is expanded by a digital filter with backward difference network. To filter high-frequency noises, a low-pass filter is added after the dynamic compensation network. To avoid overcompensation, parameters of the proposed dynamic compensation method are defined by trial and error. Step response methods are utilized in dynamic calibration experiments. Compared to experiment data without compensation, the response time of the dynamic compensated data is reduced by 30%~40%. Experiments results demonstrate the effectiveness of our method.

Automatic Pipeline Defect Detection Based on A Sewer Robot and Instance Segmentation

Robotic Object Perception Based on Multispectral Few-Shot Coupled Learning

In order to enable intelligent robots to recognize unknown objects as accurately as human beings, object perception research is of great significance in service and industrial robot application scenarios. However, object perception using spectral measurements under few-shot learning usually leads to a poor result because of inadequate training samples. To overcome this problem, this work proposes a novel few-shot learning with coupled dictionary learning (FSL-CDL) framework. First, a hybrid feature fusion method is developed to extract the multiple dimension-reduced features of original spectral measurements to build the hybrid features. Then, based on the hybrid features, a multitask coupled learning method is developed to effectively recognize unknown objects under few-shot learning. In this method, two coupling patterns, i.e., interspectroscopy coupling and intraspectroscopy coupling, effectively bridge the gap between two spectral measurements. Finally, the proposed FSL-CDL is compared with other advanced algorithms on the SMM50 dataset, and reaches 97.5% and 98.4% recognition accuracy under one-shot and five-shot learning, respectively, which are better than other algorithms. Besides, FSL-CDL can be extended to other perception tasks which contains multiple heterogeneous measurements.

Adaptive Finite-Time Synchronization Control for Teleoperation System With Varying Time Delays

In practical application, the synchronization tracking of teleoperation system requires the fast speed and strong robustness. It is the ideal control effect that the synchronization errors between master and slave robots can converge to zero in finite time. In this paper, we propose the new nonsingular terminal sliding mode and the adaptive finite-time control method for position tracking in teleoperation system. First, a novel nonsingular terminal sliding mode is designed to provide higher tracking precision and robustness. Second, the radial basis function neural networks are applied to solve dynamic uncertainties, and the adaptive laws are proposed to estimate the uncertain parameters and upper bounds of estimation. Then, the corresponding finite-time controllers of master and slave robots are designed. Third, based on the Lyapunov stability theory, synchronization performances of the closed-loop system are proved to be stable state and finite time. Finally, simulations are achieved, and some comparisons with two nonsingular terminal sliding mode control schemes and two PD methods are shown. The simulation results verify the effectiveness of the proposed control laws.

Stable haptic display with energy-compensating control

Sampled-data system nature is the main factor for a haptic system to exhibit non-passive behaviors or instabilities through "energy leaks", particularly for stiff objects rendering. An energy-compensating method is presented aiming to improve the haptic system's performance based on the concept of doing work. Using ideal continuous-time haptic system as a reference, we define an energy-compensating controller (ECC controller) which compensates the energy leaks caused by zero-order-hold and asynchronous switching during entering and leaving contacts. The ECC controller not only removes the unwanted over-work from the virtual environment during leaving contacts to eliminate the system's potential non-passive behaviors, but also enlarges the work that should be done by the operator according to the reference system's value when new contacts arise. The proposed method was tested and verified with the implementation of a "virtual stiff-wall" via Delta 6-DOF haptic device.

A sparse diverse-branch large kernel convolutional neural network for human activity recognition using wearables

Broadband wide-width acoustic topological waveguides with intermittent coupling

Haptic Interface Device Using Cable Tension Based on Ultrasonic Phased Array

This paper presents a new haptics interface device based on cable-driven parallel manipulators and an ultrasonic phased array. Unlike previous studies, our method combines force and tactile feedback at the same time. It consists of two main parts: force feedback generated by cable tension and tactile feedback provided by an ultrasonic phased array. While these parts have no direct interference, combining them can offer a synergistic effect on haptic perception. Through a series of experiments, the tactile rendering algorithm for this device is established. The output pressure of ultrasonic phased array is dependent on input command value, modulation frequency, modulation waveform, and the position of the focal point in the workspace. The results of psychophysical experiments are evaluated to determine the absolute threshold of perceivable ultrasonic tactile feedback when the cable-driven force acts on fingers. Finally, we carry out a test to confirm the accuracy advantage of our system in the virtual environment. The results of our study indicate that this device has a wide range of applications in the field of aerial haptic display.

Cross-subject emotion recognition using hierarchical feature optimization and support vector machine with multi-kernel collaboration

Objective. Due to individual differences, it is greatly challenging to realize the multiple types of emotion identification across subjects.Approach. In this research, a hierarchical feature optimization method is proposed in order to represent emotional states effectively based on peripheral physiological signals. Firstly, sparse learning combined with binary search is employed to achieve feature selection of single signals. Then an improved fast correlation-based filter is proposed to implement fusion optimization of multi-channel signal features. Aiming at overcoming the limitations of the support vector machine (SVM), which uses a single kernel function to make decisions, the multi-kernel function collaboration strategy is proposed to improve the classification performance of SVM.Main results. The effectiveness of the proposed method is verified on the DEAP dataset. Experimental results show that the proposed method presents a competitive performance for four cross-subject types of emotion identification with an accuracy of 84% (group 1) and 85.07% (group 2). Significance. The proposed model with hierarchical feature optimization and SVM with multi-kernel function collaboration demonstrates superior emotion recognition accuracy compared to state-of-the-art techniques. In addition, the analysis based on DEAP dataset composition characteristics presents a novel perspective to explore the emotion recognition issue more objectively and comprehensively.

Haptic Display of Image Based on Multi-Feature Extraction

Image feature extraction is one of the key technologies of image haptic display. In this paper, multi-feature extraction method of the object in image is proposed to improve image-based haptic perception. The multi-feature extraction includes contour shape extraction, pattern extraction and detail texture extraction. Firstly, we use an intrinsic decomposition method to decompose an image into shading image and reflectance image. The reflectance image describes nonillumination affected color patterns spread on the surface. Then, the shading image is utilized in contour shape and detail texture extraction. Contour shape extraction is based on partial differential equation (PDE), to reconstruct three-dimensional (3D) surface model in virtual environments. Detailed texture extraction is based on fractional differential method simultaneously. Finally, the various features extracted above are haptic rendered by different methods. The experimental results show the effectiveness and potentiality of the proposed method for improving the ability of haptic perception and recognition of human in virtual environments.

ProtoHAR: Prototype Guided Personalized Federated Learning for Human Activity Recognition

Federated Learning (FL) has recently attracted great interest in sensor-based human activity recognition (HAR) tasks. However, in real-world environment, sensor data on devices is non-independently and identically distributed (Non-IID), e.g., activity data recorded by most devices is sparse, and sensor data distribution for each client may be inconsistent. As a result, the traditional FL methods in the heterogeneous environment may incur a drifted global model that causes slow convergence and a heavy communication burden. Although some FL methods are gradually being applied to HAR, they are designed for overly ideal scenarios and do not address such Non-IID problem in the real-world setting. It is still a question whether they can be applied to cross-device FL. To tackle this challenge, we propose ProtoHAR, a prototype-guided FL framework for HAR, which aims to decouple the representation and classifier in the heterogeneous FL setting efficiently. It leverages the global prototype to correct the activity feature representation to make the prototype knowledge flow among clients without leaking privacy while solving a better classifier to avoid excessive drift of the local model in personalized training. Extensive experiments are conducted on four publicly available datasets: USC-HAD, UNIMIB-SHAR, PAMAP2, and HARBOX, which are collected in both controlled environments and real-world scenarios. The results show that compared with the state-of-the-art FL algorithms, ProtoHAR achieves the best performance and faster convergence speed in HAR datasets.

An Improved Level Set Method for Reachability Problems in Differential Games

This study focuses on reachability problems in differential games. An improved level set method for computing reachable tubes is proposed in this paper. The reachable tube is described as a sublevel set of a value function, which is the viscosity solution of a Hamilton-Jacobi equation with running cost. We generalize the concept of reachable tubes and propose a new class of reachable tubes, which are referred to as cost-limited one. In particular, a performance index can be specified for the system, and a cost-limited reachable tube is a set of initial states of the system's trajectories that can reach the target set before the performance index increases to a given admissible cost. Such a reachable tube can be obtained by specifying the corresponding running cost function for the Hamilton-Jacobi equation. Different non-zero sublevel sets of the viscosity solution of the Hamilton-Jacobi equation at a certain time point can be used to characterize the cost-limited reachable tubes with different admissible costs (or the reachable tubes with different time horizons), thus reducing the storage space consumption. Several examples are provided to illustrate the validity and accuracy of the proposed method.

Digital image correlation calculation method for RGB-D camera multi-view matching using variable template

Adaptive tracking control for a class of continuous-time uncertain nonlinear systems using the approximate solution of HJB equation