A5047341502- Robotics and Sensor-Based Localization
- Robotic Path Planning Algorithms
- Advanced Image and Video Retrieval Techniques
- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Indoor and Outdoor Localization Technologies
- Underwater Vehicles and Communication Systems
- Multimodal Machine Learning Applications
- Human Pose and Action Recognition
- Modular Robots and Swarm Intelligence
- Energy Efficient Wireless Sensor Networks
- Advanced Vision and Imaging
- Robotics and Automated Systems
- Cooperative Communication and Network Coding
- Infrared Target Detection Methodologies
- Target Tracking and Data Fusion in Sensor Networks
- Robot Manipulation and Learning
- Image Processing Techniques and Applications
- Embedded Systems Design Techniques
- Advanced Decision-Making Techniques
- Technology Assessment and Management
- COVID-19 diagnosis using AI
- Distributed Control Multi-Agent Systems
- Advanced Memory and Neural Computing
- Software Engineering Research
Shenzhen Academy of Robotics
2022
Intel (United States)
2021
Fudan University
2013-2015
Datang Telecom Group (China)
2011
Service robots should be able to operate autonomously in dynamic and daily changing environments over an extended period of time. While Simultaneous Localization And Mapping (SLAM) is one the most fundamental problems for robotic autonomy, existing SLAM works are evaluated with data sequences that recorded a short In real-world deployment, there can out-of-sight scene changes caused by both natural factors human activities. For example, home scenarios, objects may movable, replaceable or...
A robust and efficient Simultaneous Localization Mapping (SLAM) system is essential for robot autonomy. For visual SLAM algorithms, though the theoretical framework has been well established most aspects, feature extraction association still empirically designed in cases, can be vulnerable complex environments. This paper shows that with deep convolutional neural networks (CNNs) seamlessly incorporated into a modern framework. The proposed utilizes state-of-the-art CNN to detect keypoints...
Many signal processing problems in wireless sensor networks can be solved by graph filtering techniques. Finite impulse response (FIR) filters (GFs) have received more attention the literature because they enable distributed computation sensors. However, FIR GFs are limited their ability to represent global information of network. This letter proposes a family with infinite (IIR) and provides algorithms for realization networks. IIR bring flexibility GF designers, as designed realized even...
The recent breakthroughs in computer vision have benefited from the availability of large representative datasets (e.g. ImageNet and COCO) for training. Yet, robotic poses unique challenges applying visual algorithms developed these standard due to their implicit assumption over non-varying distributions a fixed set tasks. Fully retraining models each time new task becomes available is infeasible computational, storage sometimes privacy issues, while naïve incremental strategies been shown...
Recent breakthroughs in computer vision areas, ranging from detection, segmentation, to classification, rely on the availability of large-scale representative training datasets. Yet, robotic poses new challenges towards applying visual algorithms developed these datasets because latter implicitly assume a fixed set categories and time-invariant distribution tasks. In practice, assistive robots should be able operate dynamic environments with everyday changes. The variations four commonly...
Recent advances have enabled a single neural network to serve as an implicit scene representation, establishing the mapping function between spatial coordinates and properties. In this paper, we make further step towards continual learning of representation directly from sequential observations, namely Continual Neural Mapping. The proposed problem setting bridges gap batch-trained representations commonly used streaming data in robotics vision communities. We introduce experience replay...
The environment of most real-world scenarios such as malls and supermarkets changes at all times. A pre-built map that does not account for these becomes out-of-date easily. Therefore, it is necessary to have an up-to-date model the facilitate long-term operation a robot. To this end, paper presents general lifelong simultaneous localization mapping (SLAM) framework. Our framework uses multiple session representation, exploits efficient updating strategy includes building, pose graph...
Progress in the last decade has brought about significant improvements accuracy and speed of SLAM systems, broadening their mapping capabilities. Despite these advancements, long-term operation remains a major challenge, primarily due to wide spectrum perturbations robotic systems may encounter.Increasing robustness algorithms is an ongoing effort, however it usually addresses specific perturbation. Generalisation across large variety challenging scenarios not well-studied nor understood....
The Heterogeneous Extensible Robot Open (HERO) platform is designed for autonomous robotic research. While bringing in the flexible computational capacities by CPU and FPGA, it addresses challenges of heterogeneous computing embracing OpenCL programming. We propose approaches three fundamental tasks: simultaneous localization mapping (SLAM), motion planning convolutional neural network (CNN) inference. With FPGA acceleration, SLAM tasks are performed 2–4 times faster on HERO against...
We present a collaborative visual simultaneous localization and mapping (SLAM) framework for service robots. With an edge server maintaining map database performing global optimization, each robot can register to existing map, update the or build new maps, all with unified interface low computation memory cost. design elegant communication pipeline enable real-time information sharing between novel landmark organization retrieval method on server, acquire landmarks predicted be in its view,...
Unmanned Aerial Vehicles (UAVs) can significantly improve the autonomy of mining industry, and self-localization is key to autonomous flights underground UAVs. A localization method visual-inertial sensor data fusion proposed in this paper. The aims accuracy robustness UAVs dynamic environments. First, an algorithm for point detection rejection presented, which combines a semantic segmentation neural network, optical flow method, epipolar constraint method. Second, used enhance performance...
Visual semantic segmentation, which is represented by the segmentation network, has been widely used in many fields, such as intelligent robots, security, and autonomous driving. However, these Convolutional Neural Network (CNN)-based networks have high requirements for computing resources programmability hardware platforms. For embedded platforms terminal devices particular, Graphics Processing Unit (GPU)-based cannot meet terms of size power consumption. In contrast, Field Programmable...
The Simultaneous Localization and Mapping (SLAM) algorithm is a hotspot in robot application research with the ability to help mobile robots solve most fundamental problems of “localization” “mapping”. visual semantic SLAM fused information enables understand surrounding environment better, thus dealing complexity variability real scenarios. DS-SLAM (Semantic towards Dynamic Environment), one representative works SLAM, enhances robustness dynamic scene through information. However,...
Service robots should be able to operate autonomously in dynamic and daily changing environments over an extended period of time. While Simultaneous Localization And Mapping (SLAM) is one the most fundamental problems for robotic autonomy, existing SLAM works are evaluated with data sequences that recorded a short In real-world deployment, there can out-of-sight scene changes caused by both natural factors human activities. For example, home scenarios, objects may movable, replaceable or...
Feature extraction plays an important role in visual localization. Unreliable features on dynamic objects or repetitive regions will interfere with feature matching and challenge indoor localization greatly. To address the problem, we propose a novel network, RaP-Net, to simultaneously predict region-wise invariability point-wise reliability, then extract by considering both of them. We also introduce new dataset, named OpenLORIS-Location, train proposed network. The dataset contains 1553...
The recent breakthroughs in computer vision have benefited from the availability of large representative datasets (e.g. ImageNet and COCO) for training. Yet, robotic poses unique challenges applying visual algorithms developed these standard due to their implicit assumption over non-varying distributions a fixed set tasks. Fully retraining models each time new task becomes available is infeasible computational, storage sometimes privacy issues, while na\"{i}ve incremental strategies been...
This paper presents a hierarchical segment-based optimization method for Simultaneous Localization and Mapping (SLAM) system. First we propose reliable trajectory segmentation that can be used to increase efficiency in the back-end optimization. Then buffer mechanism first time improve robustness of segmentation. During optimization, use global information optimize frames with large error, interpolation instead update well-estimated hierarchically allocate amount computation according error...
There are two difficulties to utilize state-of-the-art object recognition/detection/segmentation methods robotic applications. First, most of the deep learning models heavily depend on large amounts labeled training data, which expensive obtain for each individual application. Second, categories must be pre-defined in dataset, thus not practical scenarios with varying categories. To alleviate reliance big this paper proposes a customized recognition and segmentation method. It aims recognize...
A robust and efficient Simultaneous Localization Mapping (SLAM) system is essential for robot autonomy. For visual SLAM algorithms, though the theoretical framework has been well established most aspects, feature extraction association still empirically designed in cases, can be vulnerable complex environments. This paper shows that with deep convolutional neural networks (CNNs) seamlessly incorporated into a modern framework. The proposed utilizes state-of-the-art CNN to detect keypoints...
The most rapid development often stems from a exact schedule plan, the problems of each project should be found answers realizable. In actual management, estimating and establishing are difficulties for long time, these not only we do have historical data also realizable easy methods. Therefore it is necessary us to study methods estimation how practice in planning. Therefore, article focus on capability analysis, research quantitative analysis model software product estimate quality...