A5100356891- 3D Shape Modeling and Analysis
- Human Pose and Action Recognition
- Multimodal Machine Learning Applications
- Robot Manipulation and Learning
- Wireless Power Transfer Systems
- Reinforcement Learning in Robotics
- Robotics and Sensor-Based Localization
- Robotic Path Planning Algorithms
- Economic theories and models
- Data Visualization and Analytics
- Energy Harvesting in Wireless Networks
- Computer Graphics and Visualization Techniques
- Innovative Energy Harvesting Technologies
- Usability and User Interface Design
- Acoustic Wave Resonator Technologies
- Game Theory and Voting Systems
- Web Data Mining and Analysis
- 3D Surveying and Cultural Heritage
- Image Processing and 3D Reconstruction
- Optimization and Variational Analysis
- Educational Technology and Pedagogy
- Advanced Neural Network Applications
- Remote Sensing and LiDAR Applications
- RFID technology advancements
Institute for Advanced Study
2023
University of Oxford
2016-2021
National University of Defense Technology
2021
Sun Yat-sen University
2021
Hong Kong Polytechnic University
2021
Jiangsu University of Technology
2019
Science Oxford
2016
Imperial College London
2013
We study the problem of efficient semantic segmentation for large-scale 3D point clouds. By relying on expensive sampling techniques or computationally heavy pre/post-processing steps, most existing approaches are only able to be trained and operate over small-scale In this paper, we introduce RandLA-Net, an lightweight neural architecture directly infer per-point semantics The key our approach is use random instead more complex selection approaches. Although remarkably computation memory...
We study the problem of efficient semantic segmentation large-scale 3D point clouds. By relying on expensive sampling techniques or computationally heavy pre/post-processing steps, most existing approaches are only able to be trained and operate over small-scale In this paper, we introduce RandLA-Net, an lightweight neural architecture directly infer per-point semantics for The key our approach is use random instead more complex selection approaches. Although remarkably computation memory...
Due to the sparse rewards and high degree of environmental variation, reinforcement learning approaches, such as deep deterministic policy gradient (DDPG), are plagued by issues variance when applied in complex real-world environments. We present a new framework for overcoming these incorporating stochastic switch, allowing an agent choose between high- low-variance policies. The switch can be jointly trained with original DDPG same framework. In this article, we demonstrate power navigation...
Monitoring large civil structures during construction and over their entire lifetime is key to being able predict structural weaknesses provide predictive maintenance. Existing approaches health monitoring typically rely on wired or wireless sensors placed the exterior of structure. Although these approaches, easy power replace, useful information, they fall short measuring what happening deep within structure itself, e.g. foundations supporting beams. Wireless embedded inside are...
The ability to interact and understand the environment is a fundamental prerequisite for wide range of applications from robotics augmented reality. In particular, predicting how deformable objects will react applied forces in real time significant challenge. This further confounded by fact that shape information about encountered world often impaired occlusions, noise missing regions e.g. robot manipulating an object only be able observe partial view entire solid. this work we present...
Modelling the physical properties of everyday objects is a fundamental prerequisite for autonomous robots. We present novel generative adversarial network (DEFO-NET), able to predict body deformations under external forces from single RGB-D image. The based on an invertible conditional Generative Adversarial Network (IcGAN) and trained collection different interest generated by finite element model simulator. Defo-netinherits generalisation GANs. This means that reconstruct whole 3-D...
We study the problem of efficient semantic segmentation for large-scale 3D point clouds. By relying on expensive sampling techniques or computationally heavy pre/post-processing steps, most existing approaches are only able to be trained and operate over small-scale In this paper, we introduce RandLA-Net, an lightweight neural architecture directly infer per-point semantics The key our approach is use random instead more complex selection approaches. Although remarkably computation memory...
Due to the sparse rewards and high degree of environment variation, reinforcement learning approaches such as Deep Deterministic Policy Gradient (DDPG) are plagued by issues variance when applied in complex real world environments. We present a new framework for overcoming these incorporating stochastic switch, allowing an agent choose between low policies. The switch can be jointly trained with original DDPG same framework. In this paper, we demonstrate power navigation task, where robot...
Structural health monitoring of critical infrastructure is key to protect large structures from and potentially catastrophic failure. It clear that for long-term sustained operation, energy must be harvested some source. Although a amount work has considered how power sensors on the exterior with solar panels or vibration harvesting, very little attention was given problem deep within structure e.g. its foundations. In this paper, we investigate whether it possible wirelessly concrete using...
The purpose of this paper is to study the existence maximal elements with applications Nash equilibrium problems for generalized games in Hadamard manifolds. By employing a KKM lemma, we establish new element theorem As applications, some results equilibria are derived. unify, improve and extend known from literature.
With the continuous development of artificial intelligence, its impact on people's lives is becoming increasingly significant. Countries, universities, and students have increased their emphasis learning intelligence knowledge. This article designs proposes a teaching mode called mo-tutor, which achieves goal (AI)knowledge through video playback, segmented speech explanations, picture text sketching, online coding. effectively coordinates AI knowledge system reduces barrier for to learn AI.