Publications

Abstract Disk model is usually used to represent graphene nanoplatelets (GNPs) which are considered as frame‐like structure with edges and corners, and it has lack of quantitative accuracy. In order to minimize error caused by morphology, distribution, and interaction between GNPs and matrix, square and folded plate models were constructed to predict the percolation volume fraction ( ϕ c ) of GNPs‐based nanocomposites by calculating connection possibility. Meanwhile, disk model is used for comparison. The results revealed that the ϕ c of square and folded plate models is smaller than that of disk model with consistent parameters, and it is concluded that the ϕ c of GNPs‐based nanocomposites predicted by disk model should be higher than that of experimental. The correctness of mixed model of square and folded plate is also verified by experimental. Due to the agglomeration of GNPs under the actual situation, the result of simulation is slightly smaller than that of experiment.

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.

An algorithm of localization for hybrid sensor networks is proposed in this paper. It is designed to be used in the unknown indoor environments. Received signal strength indicator is used to evaluate the distance between mobile node and static nodes. Mobile node takes a random walk and emits signal packets including its positional information. The static nodes can improve its positional precision via running Bayesian filters. The proposed approach does not require any navigation equipments like GPS or IMU in the nodespsila hardware. The real experimental results show that the localization of static nodes is approximately accurate and the distribution of static nodespsila location become further precise after using Bayesian filters.

With the improvement of three-dimensional (3D) reconstruction technology, virtual objects with realistic shapes have opened up the possibility for the blind or visually impaired (BVI) to improve the cognitive ability of objects by perceiving a large number of objects. In order to help BVI learn and recognize objects more conveniently, we constructed a virtual object shape recognition system (VOSRS) based on 3D reconstruction. We first proposed an improved 3D Gaussian Splatting (3DGS) method for precise 3D reconstruction of objects. Then, we introduced a mesh extraction method for 3D Gaussian to obtain a smooth 3D mesh model suitable for touch perception. Finally, we performed haptic rendering on the 3D mesh model and built an interactive system. We introduced the object 3D reconstruction process and mesh extraction steps in detail, and investigated the effect of object shape cognitive learning using the constructed system through user experiments. We conducted several experiments on the NeRF Synthetic dataset and the DTU dataset. The experimental results show that the improved 3DGS method can effectively reduce spikes and artifacts, speed up the iteration process, and improve the reconstruction quality of objects. The proposed mesh extraction method generates a smooth mesh model without holes, making it more suitable for the needs of shape perception through continuous touch. The constructed VOSRS can enable BVIs to learn and recognize the shape characteristics of virtual objects through force feedback during continuous touch. Our code can be available at https://github.com/CdpLab/VOSRS.git.

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.

With the development of virtual reality, the practical requirements of the wearable haptic interface have been greatly emphasized. While passive haptic devices are commonly used in virtual reality, they lack generality and are difficult to precisely generate continuous force feedback to users. In this work, we present SmartSpring, a new solution for passive haptics, which is inexpensive, lightweight and capable of providing controllable force feedback in virtual reality. We propose a hybrid spring-linkage structure as the proxy and flexibly control the mechanism for adjustable system stiffness. By analyzing the structure and force model, we enable a smart transform of the structure for producing continuous force signals. We quantitatively examine the real-world performance of SmartSpring to verify our model. By asymmetrically moving or actively pressing the end-effector, we show that our design can further support rendering torque and stiffness. Finally, we demonstrate the SmartSpring in a series of scenarios with user studies and a just noticeable difference analysis. Experimental results show the potential of the developed haptic display in virtual reality.