A5101826599- Autonomous Vehicle Technology and Safety
- Video Surveillance and Tracking Methods
- Traffic control and management
- Robotics and Sensor-Based Localization
- Hydraulic Fracturing and Reservoir Analysis
- Advanced Manufacturing and Logistics Optimization
- Anomaly Detection Techniques and Applications
- Advanced Neural Network Applications
- Optimization and Packing Problems
- Internet Traffic Analysis and Secure E-voting
- Advanced Image and Video Retrieval Techniques
- Transportation Planning and Optimization
- Reservoir Engineering and Simulation Methods
- Remote-Sensing Image Classification
- 3D Shape Modeling and Analysis
- Access Control and Trust
- Advanced Vision and Imaging
- Robotic Path Planning Algorithms
- Hydrocarbon exploration and reservoir analysis
- Traffic Prediction and Management Techniques
- Human Pose and Action Recognition
- Automated Road and Building Extraction
- Superconducting Materials and Applications
- Vehicular Ad Hoc Networks (VANETs)
- Remote Sensing and Land Use
Chongqing University of Posts and Telecommunications
2023-2024
University of Science and Technology of China
2024
High Magnetic Field Laboratory
2024
Chinese Academy of Sciences
2023-2024
Qingdao Institute of Marine Geology
2024
Shanghai University of Electric Power
2023
National University of Defense Technology
2018-2023
Nanjing University
2023
Shanghai Institute of Technical Physics
2023
Tsinghua University
2022
Behavior prediction in dynamic, multi-agent systems is an important problem the context of self-driving cars, due to complex representations and interactions road components, including moving agents (e.g. pedestrians vehicles) information lanes, traffic lights). This paper introduces VectorNet, a hierarchical graph neural network that first exploits spatial locality individual components represented by vectors then models high-order among all components. In contrast most recent approaches,...
Predicting the future behavior of moving agents is essential for real world applications. It challenging as intent agent and corresponding unknown intrinsically multimodal. Our key insight that prediction within a moderate time horizon, modes can be effectively captured by set target states. This leads to our target-driven trajectory (TNT) framework. TNT has three stages which are trained end-to-end. first predicts an agent's potential states $T$ steps into future, encoding its interactions...
Duckietown is an open, inexpensive and flexible platform for autonomy education research. The comprises small autonomous vehicles ("Duckiebots") built from off-the-shelf components, cities ("Duckietowns") complete with roads, signage, traffic lights, obstacles, citizens (duckies) in need of transportation. offers a wide range functionalities at low cost. Duckiebots sense the world only one monocular camera perform all processing onboard Raspberry Pi 2, yet are able to: follow lanes while...
Constructing HD semantic maps is a central component of autonomous driving. However, traditional pipelines require vast amount human efforts and resources in annotating maintaining the semantics map, which limits its scalability. In this paper, we introduce problem map learning, dynamically constructs local based on onboard sensor observations. Meanwhile, learning method, dubbed HDMapNet. HDMapNet encodes image features from surrounding cameras and/or point clouds LiDAR, predicts vectorized...
The research community has increasing interest in autonomous driving research, despite the resource intensity of obtaining representative real world data. Existing self-driving datasets are limited scale and variation environments they capture, even though generalization within between operating regions is crucial to overall viability technology. In an effort help align community's contributions with real-world problems, we introduce a new large scale, high quality, diverse dataset. Our...
We solve a challenging yet practically useful variant of 3D Bin Packing Problem (3D-BPP). In our problem, the agent has limited information about items to be packed into single bin, and an item must immediately after its arrival without buffering or readjusting. The item's placement also subjects constraints order dependence physical stability. formulate this online 3D-BPP as constrained Markov decision process (CMDP). To we propose effective easy-to-implement deep reinforcement learning...
High Definition (HD) maps are with precise definitions of road lanes rich semantics the traffic rules. They critical for several key stages in an autonomous driving system, including motion forecasting and planning. However, there only a small amount real-world topologies geometries, which significantly limits our ability to test out self-driving stack generalize onto new unseen scenarios. To address this issue, we introduce challenging task generate HD maps. In work, explore autoregressive...
Abstract We propose a novel approach to robot‐operated active understanding of unknown indoor scenes, based on online RGBD reconstruction with semantic segmentation. In our method, the exploratory robot scanning is both driven by and targeting at recognition segmentation objects from scene. Our algorithm built top volumetric depth fusion framework performs real‐time voxel‐based labeling over reconstructed volume. The guided an estimated discrete viewing score field (VSF) parameterized 3D...
This paper aims to evaluate the combined impact of communication topology (CT) and vehicular spatial distribution (SD) on mixed traffic human-driven vehicles (HDVs) connected autonomous (CAVs). A generalized car-following (CF) model is presented capture dynamics, integrating effects CT, human reaction time, information delay optimal velocity changes with memory. Then, linear stable condition derived using perturbation method. Finally, extensive simulations are conducted under different CTs...
Behavior prediction in dynamic, multi-agent systems is an important problem the context of self-driving cars, due to complex representations and interactions road components, including moving agents (e.g. pedestrians vehicles) information lanes, traffic lights). This paper introduces VectorNet, a hierarchical graph neural network that first exploits spatial locality individual components represented by vectors then models high-order among all components. In contrast most recent approaches,...
In autonomous driving, goal-based multi-trajectory prediction methods are proved to be effective recently, where they first score goal candidates, then select a final set of goals, and finally complete trajectories based on the selected goals. However, these usually involve predictions sparse predefined anchors. this work, we propose an anchor-free model, named DenseTNT, which performs dense probability estimation for trajectory prediction. Our model achieves state-of-the-art performance,...
We study the problem of learning online packing skills for irregular 3D shapes , which is arguably most challenging setting bin problems. The goal to consecutively move a sequence objects with arbitrary into designated container only partial observations object sequence. take physical realizability account, involving physics dynamics and constraints placement. policy should understand geometry be packed make effective decisions accommodate it in physically realizable way. propose...
Traditional policies often focus on access control requirement and there have been several proposals to define policy algebras handle their compositions. Recently, obligations are increasingly being expressed as part of security policies. However, the compositions interactions between these two not yet studied adequately. In this paper, we propose an algebra capturing both authorization obligation The consists constants six basic operations. It provides language independent mechanisms manage...
In this paper, we describe a framework for refinement scheme located in centralized policy server that consists of three components: knowledge database, rule set, and repository. The process includes two successive steps: transformation composition. Our takes policies written our logic-based abstract language as input generates low level rules directly implementable by individual enforcement points. We provide concrete examples coalition scenario forms mobile ad hoc network (MANET)....
We solve a challenging yet practically useful variant of 3D Bin Packing Problem (3D-BPP). In our problem, the agent has limited information about items to be packed into bin, and an item must immediately after its arrival without buffering or readjusting. The item's placement also subjects constraints collision avoidance physical stability. formulate this online 3D-BPP as constrained Markov decision process. To we propose effective easy-to-implement deep reinforcement learning (DRL) method...
Filling holes in binary images is often required during medical image processing and analysis. However, traditional hole filling algorithms for expose some disadvantages such as possible edge degradations relatively low efficiency. To overcome limits, a algorithm cell based on largest connected region extraction proposed this paper. Since there are less pixels foreground areas usual images, the can be simply filled by extracting & correspondent negative together with following OR...
Topographic shadow and water body are seriously mixed in the classification of remotely sensed images by traditional methods, such as ISODATA, maximum likelihood simple index (NDWI) based classification, especially roughly terrain areas. There some theoretical methods for retrieval correction topographic shadow, but they not operable practice because parameters involved. In this paper, a novel multi-spectral operational method (i.e., (( 4) ( 5)) 3)) d band −× − ) is proposed from Landsat...