Deep neural networks have achieved state-of-the-art performance in a wide range of recognition/classification tasks. However, when applying deep learning to real-world applications, there are still multiple challenges. A typical challenge is that unknown samples may be fed into the system during testing phase and traditional will wrongly recognize sample as one known classes. Open set recognition potential solution overcome this problem, where open classifier should ability reject well...

LiDAR-based SLAM may easily fail in adverse weathers (e.g., rain, snow, smoke, fog), while mmWave Radar remains unaffected. However, current researches are primarily focused on 2D <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(x,y)$</tex> or 3D ( xmlns:xlink="http://www.w3.org/1999/xlink">$x, y$</tex> , doppler) and LiDAR, limited work can be found for 4D y, z$</tex> doppler). As a new entrant to the market with unique characteristics,...

Large-scale visual place recognition (VPR) is inherently challenging because not all cues in the image are beneficial to task. In order highlight task-relevant feature embedding, existing attention mechanisms either based on artificial rules or trained a thorough data-driven manner. To fill gap between two types, we propose novel Semantic Reinforced Attention Learning Network (SRALNet), which inferred can benefit from both semantic priors and fine-tuning. The contribution lies two-folds. (1)...

The core of visual place recognition (VPR) lies in how to identify task-relevant cues and embed them into dis- criminative representations. Focusing on these two points, we propose a novel encoding strategy named Attentional Pyramid Pooling Salient Visual Residuals (APPSVR). It incorporates three types attention modules model the saliency local features individual, spatial cluster dimensions respectively. (1) To inhibit task-irrelevant features, semantic-reinforced weighting scheme is...

Abstract Background Understanding the metabolic activities of gut microbiome is vital for deciphering its impact on human health. While direct measurement these metabolites through metabolomics effective, it often expensive and time-consuming. In contrast, microbial composition data obtained sequencing more accessible, making a promising resource predicting metabolite profiles. However, current computational models frequently face challenges related to limited prediction accuracy,...

Place recognition is seen as a crucial factor to correct cumulative errors in Simultaneous Localization and Mapping (SLAM) applications. Most existing studies focus on visual place recognition, which inherently sensitive environmental changes such illumination, weather seasons. Considering these facts, more recent attention has been attracted use 3-D Light Detection Ranging (LiDAR) scans for demonstrates credibility by exerting accurate geometric information. Different from pure...

Simultaneous Localization and Mapping (SLAM) is moving towards a robust perception age. However, LiDAR-and visual- SLAM may easily fail in adverse conditions (rain, snow, smoke fog, etc.). In comparison, based on 4D Radar, thermal camera IMU can work robustly. But only few literature be found. A major reason the lack of related datasets, which seriously hinders research. Even though some datasets are proposed radar past four years, they mainly designed for object detection, rather than SLAM....

Deep neural networks have achieved state-of-the-art performance in a wide range of recognition/classification tasks. However, when applying deep learning to real-world applications, there are still multiple challenges. A typical challenge is that unknown samples may be fed into the system during testing phase and traditional will wrongly recognize sample as one known classes. Open set recognition potential solution overcome this problem, where open classifier should ability reject well...

Multi-LiDAR system is an important part of V2X (Vehicle to Everything) enhance the perception information for unmanned vehicles. To fuse from multiple 3D LiDARs, accurate extrinsic calibration between LiDARs essential. However, existing multi-LiDAR methods mainly focus on short baseline scenarios, where are closely mounted a single platform (e.g., vehicle). Besides, most typically use planar target calibration. Some require motion system. The above conditions severely limit application these...

Both 3D mmWave Radar (3D: <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$x, y, z$</tex> ) and thermal camera are robust in harsh environments. Fusing them is beneficial for V2X road side units unmanned vehicles to operate all-weather conditions. To fuse the two sensors, accurate extrinsic calibration indispensable. However, limited literatures can be found, since radar pushed into market just a short period. Most research focused on 2D (2D:...

Most of the existing mobile robot localization solutions are either heavily dependent on pre-installed infrastructures or having difficulty working in highly repetitive environments which do not have sufficient unique features. To address this problem, we propose a magnetic-assisted initialization approach that enhances performance infrastructure-free featureless environments. The proposed system adopts coarse-to-fine structure, mainly consists two parts: magnetic field-based matching and...

Adversarial examples can be imperceptible to human eyes but easily fool deep models. Such intrigue property has raised security issues for real-world industrial learning systems. To combat those malicious attacks, a novel defense strategy been proposed based on the conditional variational autoencoder (CVAE) and Bayesian network (BN). The main contribution lies in provided systematic dual-domain-based framework, which covers three modules named detection, diagnosis, recovery. Specifically,...

Symmetric environment is one of the most intractable and challenging scenarios for mobile robots to accomplish global localization tasks, due highly similar geometrical structures insufficient distinctive features. Existing solutions in such either depend on pre-deployed infrastructures which are expensive, inflexible, hard maintain; or rely single sensor-based methods whose initialization module incapable provide enough unique information. Thus, this paper proposes a novel Multi-Sensor...

Visual Place Recognition (VPR) has become an indispensable capacity for mobile robots to operate in large-scale environments. Existing methods this field mostly focus on exploring high-performance encoding strategies, while few attempts are devoted lightweight models that balance per-formance and computational cost. In work, we propose a Lightweight Self-attentional Distillation Network (LSDNet) aiming obtain advantages of both performance efficiency. (1) From perspective, attentional...

Navigating delivery robot along the sidewalk safely and robustly in a campus environment is extremely challenging due to narrow motion space, appearance changes unstable GPS localization signal under canopies of trees, etc. To that end, we have completed systematic implementation for navigation, where robust vision based navigation algorithm has been proposed. And it consists three main modules: segmentation, costmap generation planning. More Specifically, first module find drivable area...

Accurate global localization and tracking are essential ingredients for autonomous mobile robots (AMRs) operating in enclosed or partially repetitive environments (e.g., office corridors, industrial warehouses, transportation centers). In such degenerated challenging environments, navigation satellite system signals unreliable severely degraded. Existing infrastructure-based beacons/QR codes-based) solutions suffer from inflexibility high maintenance costs, while traditional geometric...

Traditional Gaussian process regression suffers from the cubic complexity and excessive computation burdens for industrial big data. To get rid of such defect, this work proposes a scalable soft sensor called bagging stochastic variational GP (SVGPR). We first formulate within sparse approximation framework. Then, inference (SVI) mechanism is induced, which can significantly break formidable obstacle to nonlinear data modeling. In addition, imposed automatic relevance determination strategy...

For applications in robotics and autonomous vehicles, fusing 3D LiDAR, RGB camera, thermal camera can improve the perception day night environments. To fuse sensors, accurate extrinsic calibration is important. However, most existing methods use offline calibration, which tedious, as it requires of special targets human intervention. Meanwhile, among online methods, little attention has been paid to LiDAR calibration. Thus, this paper, an method proposed solve problem, called RGBDTCalibNet....

The gut microbiome has been regarded as one of the fundamental determinants regulating human health, and multi-omics data profiling increasingly utilized to bolster deep understanding this complex system. However, stemming from cost or other constraints, integration often suffers incomplete views, which poses a great challenge for comprehensive analysis. In work, novel model named Incomplete Multi-Omics Variational Neural Networks (IMOVNN) is proposed integration, disease prediction...

Place recognition plays a vital role in eliminating accumulated drift from visual odometry SLAM system. Bag- of-Words (BoW) -based approach is the most popular solution due to its efficiency and robustness. We propose use Line- Junction-Line (LJL) build BoW for place urban environments. LJL simple structure of two lines with their intersection. Different point features which are detected based on pixel intensity patterns, it represents physical existence, more robust challenging scenarios....

Robust visual place recognition (VPR) against significant appearance changes is crucial for the life-long operation of mobile robots. Focusing on this task, we propose a Co-Attentive Hierarchical Image Representations (CAHIR) framework VPR, which unifies attention-sharing global and local descriptor generation into one encoding pipeline. The hierarchical descriptors are applied to coarse-to-fine VPR system with retrieval geometric verification. To explore high-quality matches between...

Visual Place Recognition (VPR) is essential for autonomous robots and unmanned vehicles, as an accurate identification of visited places can trigger a loop closure to optimize the built map. The most prevalent methods tackle VPR single-frame retrieval task, which uses CNN-based encoder describe compare each individual frame. These methods, however, overlook temporal information between frames. Other improve this by searching database with consecutive frames, greatly reduce false positives....

Performing collaborative semantic mapping is a critical challenge for multi-robot systems to maintain comprehensive contextual understanding of the surroundings. In this paper, novel hierarchical map fusion framework proposed, where problem addressed in low-level single robot and high level global fusion. process, Bayesian rule used label occupancy probability updating, information added geometric grid. High covers decentralized sharing updating. Collaborative conducted two scenarios, that...