A5068172501- Robotic Path Planning Algorithms
- Multimodal Machine Learning Applications
- Reinforcement Learning in Robotics
- Domain Adaptation and Few-Shot Learning
- Robotics and Sensor-Based Localization
- Modular Robots and Swarm Intelligence
- Human-Automation Interaction and Safety
- Risk and Safety Analysis
- Ergonomics and Human Factors
- Autonomous Vehicle Technology and Safety
- Advanced Clustering Algorithms Research
- Digital Marketing and Social Media
- Logic, Reasoning, and Knowledge
- Knowledge Management and Sharing
- Formal Methods in Verification
- Quality Function Deployment in Product Design
- Anomaly Detection Techniques and Applications
- Parallel Computing and Optimization Techniques
- Control and Dynamics of Mobile Robots
- Evacuation and Crowd Dynamics
- Advanced Computational Techniques and Applications
- Advanced Neural Network Applications
- Advanced Research in Systems and Signal Processing
- 3D Surveying and Cultural Heritage
- Image Retrieval and Classification Techniques
University of Science and Technology of China
2021-2024
Chongqing Normal University
2024
China Southern Power Grid (China)
2024
Institute of Microelectronics
2023
Chinese Academy of Sciences
2023
University of Chinese Academy of Sciences
2007-2023
Guangdong University Of Finances and Economics
2017
Guangdong University of Finance
2017
Tensor program tuning is a non-convex objective optimization problem, to which search-based approaches have proven be effective. At the core of lies design cost model. Though deep learning-based models perform significantly better than other methods, they still fall short and suffer from following problems. First, their feature extraction heavily relies on expert-level domain knowledge in hardware architectures. Even so, extracted features are often unsatisfactory require separate...
Simultaneous localization and mapping (SLAM) based on laser sensors has been widely adopted by mobile robots autonomous vehicles. These SLAM systems are required to support accurate with limited computational resources. In particular, point cloud registration, i.e., the process of matching aligning multiple LiDAR scans collected at locations in a global coordinate framework, deemed as bottleneck step SLAM. this paper, we propose feature filtering algorithm, PFilter, that can filter out...
Embodied PointGoal navigation is a fundamental task for embodied agents. Recent works have shown that the performance of agent degrades significantly in presence visual corruption, including Spatter, Speckle Noise, and Defocus Blur, showing weak robustness agent. To improve agents to various corruptions, we propose framework called Regularized Denoising Masked AutoEncoders Navigation (RDMAE-Nav). In nutshell, RDMAE-Nav mainly consists two modules: module policy module. module,...
User behavior has been paid lots of attention in the Information System (IS) research. Previous research on IS user focused in-role behaviors such as adoption, initial usage and continuous usage, while little to extra-role organizational citizenship (OCB). However, with application information technology (IT) organizations, boundary employees' task becomes ambiguous, OCB plays an important role achieving performance. Therefore, it is empirically analyze effects performance context. Based...
Abstract For a long time, the low-voltage distribution network has problems of untimely management and complex frequently changing lines, which makes problem missing grid topology information increasingly serious. This study proposes an automatic detection model based on lasso algorithm t-distributed random neighbor embedding algorithm. The identifies household-variable relationship through algorithm, then station area model. experimental results indicated that constant least squares ridge...
The integration of large language models (LLMs) with robotics has significantly advanced robots' abilities in perception, cognition, and task planning. use natural interfaces offers a unified approach for expressing the capability differences heterogeneous robots, facilitating communication between them, enabling seamless allocation collaboration. Currently, utilization LLMs to achieve decentralized multi-heterogeneous robot collaborative tasks remains an under-explored area research. In...
Abstract Recent advances in measuring high-dimensional modalities, including protein levels and DNA accessibility, at the single-cell level have prompted need for frameworks capable of handling multi-modal data while simultaneously addressing multiple tasks. Despite these advancements, much work domain remains limited, often focusing on either a single-modal or single-task perspective. A few recent studies ventured into multimodal, multi-task learning, but we identified ① Optimization...
It is challenging to develop an online path planning algorithm for Ackermann-steering vehicles find collision-free and kinematically-feasible paths, that efficient dense environments, adaptable various suitable environments with narrow passages. In this paper, we propose a kinematically constrained RRT-based integrating trajectory parameter space (TP-space) three novel improvements meet the above requirements. specific, introduce new way choose candidate nodes expand tree algorithm, which...
This paper addresses the challenge of generating safety-critical scenarios with multiple adversarial vehicles for testing autonomous vehicles. Such must be plausible and collision-avoidable while resulting in a collision vehicle-under-test. However, tremendous number low ratio makes previous methods squander primary resources on implausible scenarios, degenerating their efficiency. We propose two-stage framework called ASP-based Avoidable Collision Scenario Testbench (A²CoST) to overcome...
It is important for deep reinforcement learning (DRL) algorithms to transfer their learned policies new environments that have different visual inputs. In this paper, we introduce Prompt based Proximal Policy Optimization (P <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> O), a three-stage DRL algorithm transfers representations from target source environment by applying prompting. The process of P O consists three stages: pre-training,...
Robot navigation using deep reinforcement learning (DRL) has shown great potential in improving the performance of mobile robots. Nevertheless, most existing DRL-based methods primarily focus on training a policy that directly commands robot with low-level controls, like linear and angular velocities, which leads to unstable speeds unsmooth trajectories during long-term execution. An alternative method is train DRL outputs path directly. However, two roadblocks arise for paths: (1) The...