检索结果-内蒙古大学图书馆

arXiv 2019年

作者： Shao, Kun Tang, Zhentao Zhu, Yuanheng Li, Nannan Zhao, Dongbin State Key Laboratory of Management and Control for Complex Systems Institute of Automation Chinese Academy of Sciences Beijing100190 China University of Chinese Academy of Sciences Beijing China

—Deep reinforcement learning (DRL) has made great achievements since proposed. Generally, DRL agents receive high-dimensional inputs at each step, and make actions according to deep-neural-network-based policies. This learning mechanism updates the policy to maximize the return with an end-to-end method. In this paper, we survey the progress of DRL methods, including value-based, policy gradient, and model-based algorithms, and compare their main techniques and properties. Besides, DRL plays an important role in game artificial intelligence (AI). We also take a review of the achievements of DRL in various video games, including classical Arcade games, first-person perspective games and multi-agent real-time strategy games, from 2D to 3D, and from single-agent to multi-agent. A large number of video game AIs with DRL have achieved super-human performance, while there are still some challenges in this domain. Therefore, we also discuss some key points when applying DRL methods to this field, including exploration-exploitation, sample efficiency, generalization and transfer, multi-agent learning, imperfect information, and delayed spare rewards, as well as some research directions. Copyright © 2019, The Authors. All rights reserved.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Convolutional Neural Networks on Apache Storm

Convolutional Neural Networks on Apache Storm

引用

Chinese Automation Congress (CAC)

作者： Wenyu Zhang Yanfeng Lu Yi Li Hong Qiao The State Key Laboratory of Management and Control for Complex Systems Institute of Automation Chinese Academy of Sciences Beijing China School of Information Engineering Nanchang University Nanchang China

the performance of deep learning largely depends on the size of data. One data source is real-time streaming data, produced from mobile devices, sensors or social media, etc. Streaming data is high-speed and large-scale, which needs real-time processing. However, current mainstream frameworks are mainly designed for off-line data. To suit this, we first propose a deep learning framework based on Apache Storm, which is a distributed stream processing frame, fast and fault-tolerant. Our framework implements the distributed training of CNNs. which is different from MMLSpark or TensorFlowOnSpark that is a pure Java implementation. The design of message passing and synchronization is also suitable to other MapReduce-family distributed computing platforms. To validate our work, MNIST and Cifar -10 datasets are used for evaluation and comparison with similar architectures. The results show our framework, in resource-limited environment, realizes about 10 times speedup.

关键词： Fasteners Storms Topology Parallel processing Training Neural networks Machine learning

来源：评论

学校读者我要写书评

暂无评论

A Bio-inspired Soft Robotic Arm： Kinematic Modeling and Hydrodynamic Experiments

引用

Journal of Bionic Engineering 2018年第2期15卷 204-219页

作者： Zheyuan Gong Jiahui Cheng Xingyu Chen Wenguang Sun Xi Fang Kainan Hu Zhexin Xie Tianmiao Wang Li Wen School of Mechanical Engineering and Automation Beihang University Beijing 100191 China State Key Laboratory of Management and Control for Complex Systems Institute of Automation Chinese Academy of Sciences Beijing 100190 China . Beijing Advanced Innovation Center for Biomedical Engineering Beihang University Beijing 100191 China

Soft robotics has several promising properties for aquatic applications, such as safe interaction with environments, lightweight, low cost, etc. In this paper, we proposed the kinematic modeling and hydrodynamics experiments of a soft robotic arm with 3D locomotion capacity. We developed a mathematical model that incorporates the angle correction, as well as the open-loop model-based motion control. The model could precisely predict the three-dimensional （3D） movement, and the location error is less than 5.7 mm in different attitudes. Furthermore, we performed the hydrodynamic investigations and simultaneously measured the hydrodynamic forces and the wake flows at different amplitudes （50 mm, 100 mm, 150 mm, 200 mm） and frequencies （0.3 Hz, 0.4 Hz, 0.5 Hz） of the soft arm. Surprisingly, we found that the magnitudes of the hydrodynamic force （〈1 N） and the torques （〈0.08 N-m） of dynamically moving soft arm were tiny, which leads to negligible inertial effect for the underwater vehicle than those of the traditional rigid underwater manipulator. Finally, we demonstrated underwater picking and placing tasks of the soft manipulator by using a computer program that controls the tip attitude and velocity. This study may inspire future underwater manipulators that have properties of low-inertial, low power cost and can safely interact with the aauatic environments.

关键词： bio-inspired soft robotic arm aquatic environment kinematics digital particle image velocimetry

来源：评论

学校读者我要写书评

暂无评论

Attention Grasping Network: A Real-time Approach to Generating Grasp Synthesis

Attention Grasping Network: A Real-time Approach to Generati...

引用

IEEE International Conference on Robotics and Biomimetics

作者： Qipeng Gu Jianhua Su Xueting Bi the State Key Laboratory of Management and Control for Complex Systems Institute of Automation Chinese Academy of Sciences Beijing China Key Laboratory of Electromagnetic Radiation and Sensing Technology Institute of Electronics Chinese Academy of Sciences Beijing China

This paper presents a real-time, pixelwise method to generate grasp synthesis based on fully convolutional netural networks (FCN). Our proposed Attention Grasping Network (AGN) applies a novel attention mechanism to robotic grasp detection, which automatically learns to focus on salient features of the input image. The model with attention mechnisms can compensate for the loss of detail information in standard FCN, which increases the sensitivity of the model and accuracy of prediction. In addition,in order to ensure a real-time grasp and save computing resources, the light-weight AGN model predicts the position and angle of grasping point. Our method only takes 22ms to execute the grasp detection pipeline on a GPU-equipped computer, and achieves 97.8% accuracy on Cornell Grasping Dataset.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Compare contact model-based control and contact model-free learning: A survey of robotic peg-in-hole assembly strategies

arXiv

引用

arXiv 2019年

作者： Xu, Jing Hou, Zhimin Liu, Zhi Qiao, Hong State Key Laboratory of Tribology Beijing Key Laboratory of Precision/Ultra-Precision Manufacturing Equipment Control Department of Mechanical Engineering Tsinghua University Beijing100084 China State Key Laboratory of Management and Control for Complex Systems Institute of Automation Chinese Academy of Sciences Beijing100190 China

In this paper, we present an overview of robotic peg-in-hole assembly and analyze two main strategies: Contact model-based and contact model-free strategies. More specifically, we first introduce the contact model control approaches, including contact state recognition and compliant control two steps. Additionally, we focus on a comprehensive analysis of the whole robotic assembly system. Second, without the contact state recognition process, we decompose the contact modelfree learning algorithms into two main subfields: Learning from demonstrations and learning from environments (mainly based on reinforcement learning). For each subfield, we survey the landmark studies and ongoing research to compare the different categories. We hope to strengthen the relation between these two research communities by revealing the underlying links. Ultimately, the remaining challenges and open questions in the field of robotic peg-in-hole assembly community are discussed. The promising directions and potential future work are also considered. Copyright © 2019, The Authors. All rights reserved.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Parallel intelligence： toward lifelong and eternal developmental AI and learning in cyber-physical-social spaces

引用

Frontiers of Computer Science 2018年第3期12卷 401-405页

作者： Fei-Yue WANG Jun Jason ZHANG Xiao WANG The State Key Laboratory for Management and Control of Complex Systems Institute of Automation Chinese Academy of Sciences Beijing 100190 China School of Electrical Engineering Wuhan University Wuhan 430072 China Qingdao Academy of Intelligent Industries Qingdao 266109 China

Dear editor, Fundamentally, there are only two main approaches so far in artificial intelligence （AI）： reasoning-oriented formal logic approach and function-oriented computational intelligence approach, so called Neats vs. Scuruffies, which is a reflection of the historical fight between two schools of thought for formalism and empiricism respectively in the field of AI that is continuing even today.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A Grasping CNN with Image Segmentation for Mobile Manipulating Robot

A Grasping CNN with Image Segmentation for Mobile Manipulati...

引用

IEEE International Conference on Robotics and Biomimetics

作者： Yingying Yu Zhiqiang Cao Shuang Liang Zhicheng Liu Junzhi Yu Xuechao Chen State Key Laboratory of Management and Control for Complex Systems Chinese Academy of Sciences Beijing China Beijing Advanced Innovation Center for Intelligent Robots and Systems Intelligent Robotics Institute BIT Beijing China

This paper presents a grasping convolutional neural network with image segmentation for mobile manipulating robot. The proposed method is cascaded by a feature pyramid network FPN and a grasping network DrGNet. The FPN network combined with point cloud clustering is used to obtain the mask of the target object. Then, the grayscale map and the depth map corresponding to the target object are combined and sent to the DrGNet network for providing multi-scale images. On this basis, depthwise separable convolution is used for encoding. The results of encoders are refined according to the light-weight RefineNet as well as sSE, which can achieve a better grasp detection. The proposed method is verified by the experiments on mobile manipulating robot.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Object Reconstruction with Deep Learning: A Survey

Object Reconstruction with Deep Learning: A Survey

引用

IEEE International Conference on Cyber Technology in Automation, control, and Intelligent systems

作者： Zishu Gao En Li Guodong Yang Zhe Wang Yunong Tian Zize Liang Rui Guo Shengchuan Li The State Key Laboratory of Management and Control for Complex Systems Institute of Automation Chinese Academy of Sciences and The University of Chinese Academy of Sciences Beijing China The State Key Laboratory of Management and Control for Complex Systems Institute of Automation Chinese Academy of Sciences Beijing China State Grid Shandong Electric Power Company Jinan China State Grid Liaoning Electric Power Company Limited Shenyang China

ISBN: (数字)9781728107707

ISBN: (纸本)9781728107714

Object reconstruction is one of the most crucial branches of computer vision. With the development of deep learning, many tasks have achieved remarkable improvements in computer vision. 3D reconstruction with deep learning also has attracted much attention in recent years. Deep learning methods based on CNN-based and GAN-based architectures have been adopted for 3D object prediction. In addition, researchers utilize different inputs such as RGB and depth images to achieve prediction based on different problem. In this paper, we provide a detailed overview of recent advances in 3D object reconstruction. The reviewed approaches are categorized into three groups depending on the input modality: RGB-based, depth-based and other-input-based. Particularly, we introduce the various methods and indirectly classify the shape representation. As a survey, we discuss the strong and weak points of exciting approaches.

关键词： Three-dimensional displays Image reconstruction Shape Machine learning Generators Two dimensional displays Automation

来源：评论

学校读者我要写书评

暂无评论

3D Deep Learning for 3D Printing of Tooth Model

3D Deep Learning for 3D Printing of Tooth Model

引用

IEEE International Conference on Service Operations and Logistics, and Informatics, SOLI

作者： Yiwen Liu Xiuqin Shang Zhen Shen Bin Hu Zhihai Wang Gang Xiong Faculty of Information The State Key Laboratory for Management and Control of Complex Systems Institute of Automation Chinese Academy of Sciences Beijing University of Technology Beijing China The State Key Laboratory for Management and Control of Complex Systems Institute of Automation Chinese Academy of Sciences Beijing China Beijing Engineering Research Center of Intelligent Systems and Technology Chinese Academy of Sciences Beijing China Faculty of Information Beijing University of Technology Beijing China

Recently, with the development of both 3D sensors and 3D virtual network that bring the needs of interaction with the real world, many 3D applications burst out. However, it is difficult to understanding these three-dimensional scenes with a fixed program. Then, a data-driven method is required to process these 3D data, which brings a strong demand of 3D Deep Learning in 3D data. Towards this goal, with an end-to-end deep learning, the experiment is based on PointNet++, a well proposed method for feature extraction. The experiment optimizes the network structure and parameters to improve the classification results. Finally, the network is applied to tooth model for classification and identification so that the dental model can be found from different perspectives.

关键词： Three-dimensional displays Deep learning Transforms Cameras Object segmentation Image segmentation Sensors

来源：评论

学校读者我要写书评

暂无评论

3D Face Reconstruction System From A Single Photo Based On Regression Neural Network

引用

IFAC-PapersOnLine 2020年第5期53卷 71-76页

作者： Yatai Ji Keqiang Li Huaiyu Wu Gang Xiong Zhen Shen Xiuqin Shang Bangwen Xi The State Key Laboratory for Management and Control of Complex Systems Institute of Automation Chinese Academy of Sciences Beijing 100190 China Department of Automation Tsinghua University Beijing 100084 China University of Chinese Academy of Sciences Beijing 100049 China The Guangdong Engineering Research Center of 3D Printing and Intelligent Manufacturing Cloud Computing Center Chinese Academy of Sciences Donggguan 523808 China The Beijing Engineering Research Center of Intelligent Systems and Technology Institute of Automation Chinese Academy of Sciences Beijing 100190 China

3D human face modeling is one of the most popular research directions in the field of computer stereo vision. On the one hand, the complex physiological structure of human face as well as various expressions and attitude changes bring great challenges to three-dimensional modeling, which makes the modeling method of human face have high research value and reference significance. On the other hand, 3D face reconstruction has a broad application prospect, and the application demand attracts more researchers to invest in it. Based on the 3DMM and 3DDFA method, this paper utilizes regression neural network to realize end-to-end face reconstruction. After that, we build a visualization program with the training model to realize face modeling based on a single face image of any pose or expression. It’s a 3D face reconstruction system which has functions such as face alignment, face rotation and so on.

关键词： face reconstruct 3DMM 3DDFA regression network

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：