检索结果-内蒙古大学图书馆

TechRxiv 2021年

作者： Zhang, Tiantian Wang, Xueqian Liang, Bin Yuan, Bo The Intelligent Computing Lab Division of Informatics Shenzhen International Graduate School Tsinghua University Shenzhen518055 China The Center for Artificial Intelligence and Robotics Shenzhen International Graduate School Tsinghua University Shenzhen518055 China The Research Center for Navigation and Control Department of Automation Tsinghua University Beijing100084 China

The powerful learning ability of deep neural networks enables reinforcement learning (RL) agents to learn competent control policies directly from high-dimensional and continuous environments. In theory, to achieve stable performance, neural networks assume i.i.d. inputs, which unfortunately does no hold in the general RL paradigm where the training data is temporally correlated and non-stationary. This issue may lead to the phenomenon of "catastrophic interference" (a.k.a. "catastrophic forgetting") and the collapse in performance as later training is likely to overwrite and interfer with previously learned good policies. In this paper, we introduce the concept of "context" into the single-task RL and develop a novel scheme, termed as Context Division and Knowledge Distillation (CDaKD) driven RL, to divide all states experienced during training into a series of contexts. Its motivation is to mitigate the challenge of aforementioned catastrophic interference in deep RL, thereby improving the stability and plasticity of RL models. At the heart of CDaKD is a value function, parameterized by a neural network feature extractor shared across all contexts, and a set of output heads, each specializing on an individual context. In CDaKD, we exploit online clustering to achieve context division, and interference is further alleviated by a knowledge distillation regularization term on the output layers for learned contexts. In addition, to effectively obtain the context division in high-dimensional state spaces (e.g., image inputs), we perform clustering in the lower-dimensional representation space of a randomly initialized convolutional encoder, which is fixed throughout training. Our results show that, with various replay memory capacities, CDaKD can consistently improve the performance of existing RL algorithms on classic OpenAI Gym tasks and the more complex high-dimensional Atari tasks, incurring only moderate computational overhead. © 2021, CC BY.

关键词： Distillation

来源：评论

学校读者我要写书评

暂无评论

NccFlow: Unsupervised learning of optical flow with non-occlusion from geometry

arXiv

引用

arXiv 2021年

作者： Wang, Guangming Ren, Shuaiqi Wang, Hesheng The Key Laboratory of System Control and Information Processing of Ministry of Education Department of Automation Institute of Medical Robotics Shanghai Jiao Tong University Shanghai200240 China The Key Laboratory of Marine Intelligent Equipment and System of Ministry of Education Shanghai Engineering Research Center of Intelligent Control and Management Shanghai Jiao Tong University Shanghai200240 China

Optical flow estimation is a fundamental problem of computer vision and has many applications in the fields of robot learning and autonomous driving. This paper reveals novel geometric laws of optical flow based on the insight and detailed definition of non-occlusion. Then, two novel loss functions are proposed for the unsupervised learning of optical flow based on the geometric laws of non-occlusion. Specifically, after the occlusion part of the images are masked, the flowing process of pixels is carefully considered and geometric constraints are conducted based on the geometric laws of optical flow. First, neighboring pixels in the first frame will not intersect during the pixel displacement to the second frame. Secondly, when the cluster containing adjacent four pixels in the first frame moves to the second frame, no other pixels will flow into the quadrilateral formed by them. According to the two geometrical constraints, the optical flow non-intersection loss and the optical flow non-blocking loss in the non-occlusion regions are proposed. Two loss functions punish the irregular and inexact optical flows in the non-occlusion regions. The experiments on datasets demonstrated that the proposed unsupervised losses of optical flow based on the geometric laws in non-occlusion regions make the estimated optical flow more refined in detail, and improve the performance of unsupervised learning of optical flow. In addition, the experiments training on synthetic data and evaluating on real data show that the generalization ability of optical flow network is improved by our proposed unsupervised approach. Copyright © 2021, The Authors. All rights reserved.

关键词： Computer vision

来源：评论

学校读者我要写书评

暂无评论

A Survey of Robot Manipulation in Contact

arXiv

引用

arXiv 2021年

作者： Suomalainen, Markku Karayiannidis, Yiannis Kyrki, Ville Center of Ubiquitous Computing Faculty of Information Technology and Electrical Engineering University of Oulu Finland Systems and Control Department of Electrical Engineering Chalmers University of Technology Gothenburg Sweden Intelligent Robotics research group Department of Electrical Engineering and Automation School of Electrical Engineering Aalto University Helsinki Finland

In this survey, we present the current status on robots performing manipulation tasks that require varying contact with the environment, such that the robot must either implicitly or explicitly control the contact force with the environment to complete the task. Robots can perform more and more manipulation tasks that are still done by humans, and there is a growing number of publications on the topics of 1) performing tasks that always require contact and 2) mitigating uncertainty by leveraging the environment in tasks that, under perfect information, could be performed without contact. The recent trends have seen robots perform tasks earlier left for humans, such as massage, and in the classical tasks, such as peg-in-hole, there is a more efficient generalization to other similar tasks, better error tolerance, and faster planning or learning of the tasks. Thus, in this survey we cover the current stage of robots performing such tasks, starting from surveying all the different in-contact tasks robots can perform, observing how these tasks are controlled and represented, and finally presenting the learning and planning of the skills required to complete these tasks. © 2021, CC BY.

关键词： Surveys

来源：评论

学校读者我要写书评

暂无评论

TLD tracking algorithm for metro construction site security

TLD tracking algorithm for metro construction site security

引用

Chinese automation Congress (CAC)

作者： Jiang Zhu Yang Chen Jiaji Cai Xiujuan Zheng Huaiyu Wu Engineering Research Center for Metallurgical Automation and Measurement Technology of Ministry of Education Wuhan China Institute of Robotics and Intelligent Systems Wuhan University of Science and Technology Wuhan China

ISBN: (数字)9781728176871

ISBN: (纸本)9781728176888

In view of security issues, this paper studies how to track the target at the metro construction site accurately and effectively. It is proposed to replace the tracking module in the TLD(Tracking-Leaning-Detection) algorithm with the GOTURN algorithm, to enhance the tracking robustness of the original algorithm in response to the target's illumination changes, rotation, occlusion, etc. To solve the problem of inaccurate sample division in the TLD algorithm learning module, a sample constraint mechanism is introduced to divide positive and negative samples more objectively and accurately. The sample deletion mechanism of weighted model matching is introduced to delete samples that are weak in characterizing the current target, which improves the tracking accuracy and efficiency of the algorithm. The experimental results show that the improved algorithm can well reduce the interference that is caused by the target pedestrian's illumination change, turning, occlusion, etc. It can accurately track the target pedestrian for a long time, and has good robustness and real-time performance.

关键词： Target tracking Classification algorithms Feature extraction Tracking Robustness Training Real-time systems

来源：评论

学校读者我要写书评

暂无评论

Spatial-Temporal Evolution Prediction of Gas Distribution based on PSO-Elman Neural Network

Spatial-Temporal Evolution Prediction of Gas Distribution ba...

引用

第三十九届中国控制会议

作者： Chaonan Shen Lei Cheng Qin Liu Yang Chen Huaiyu Wu Engineering Research Center for Metallurgical Automation and Measurement Technology of Ministry of Education Wuhan University of Science and Technology Institute of Robotics and Intelligent Systems Wuhan University of Science and Technology

The rapid development of the industrial economy has increased the use of hazardous chemicals in chemical ***-time monitoring of dangerous gas leakage has become an important approach to ensure the safety of life and *** a gas leak occurs, it is difficult to estimate the distribution of dangerous gases in space due to the limited detection range of monitoring points, and it is impossible to understand how the gas will evolve in space in the future. In this study, we propose a prediction algorithm for the temporal and spatial evolution of gas concentration distribution. At first, this study uses Elman neural network(ENN) to predict the trend of concentration distribution, and it is optimized by particle swarm optimization(PSO). Then,Inverse distance weighting(IDW) was used to spatially interpolate the gas concentration data predicted by the PSO-Elman to obtain the predicted concentration values of undistributed monitoring points in the experimental area. Simulation validations are conducted on the basis of public data set from Mobile robotics and Olfaction lab in ?rebro University, and the rusults show that the model established by the PSO-Elman can accurately predicted the trend of gas concentration distribution for a long *** proposed method can be applied to important fields such as environmental monitoring, chemical leak detection, inspection of landfills and so on.

关键词： Gas distribution mapping Temporal and spatial evolution PSO-Elman IDW

来源：评论

学校读者我要写书评

暂无评论

Information collection system of construction progress based on SLAM and edge extraction

Information collection system of construction progress based...

引用

Chinese automation Congress (CAC)

作者： Fan Wu Huaiyu Wu Xiujuan Zheng Engineering Research Center for Metallurgical Automation and Measurement Technology of Ministry of Education Wuhan China Institute of Robotics and Intelligent Systems Wuhan University of Science and Technology Wuhan China

ISBN: (数字)9781728176871

ISBN: (纸本)9781728176888

In view of the shortage of construction progress supervision and management, a construction progress information acquisition system based on simultaneous localization and mapping(SLAM) and edge extraction is proposed. The system first uses Hokuyo UXM-30LX-EW lidar and SLAM technology to obtain a global map; Then pass the global map as input to composite feature network (richer convolutional features), and use the edge detection method to optimize the global map. This part is to extract the outline of the target environment and reduce the amount of subsequent engineering calculations; Finally, through the image similarity comparison algorithm, the similarity between the current construction site scene and the design drawing is compared to obtain the construction progress information. This paper designs an experimental scenario and verifies the feasibility of the proposed system using experiments. The results show that the construction progress information collection operation of this system is simple, and it is both feasible and robust.

关键词： Simultaneous localization and mapping Image edge detection Feature extraction Real-time systems Data mining Robots Monitoring

来源：评论

学校读者我要写书评

暂无评论

Dynamic Balance and Trajectory Tracking Control of Quadruped Robots Based on Virtual Model Control

Dynamic Balance and Trajectory Tracking Control of Quadruped...

引用

第三十九届中国控制会议

作者： Guangrong Chen Bowen Hou Sheng Guo Junzheng Wang Robotics Research Center Beijing Jiaotong University Beijing Engineering and Technology Research Center of Rail Transit Line Safety and Disaster Prevention State Key Laboratory of Intelligent Control and Decision of Complex Systems School of Automation Beijing Institute of Technology

As for a motion control framework of robots, virtual model control（VMC） can use virtual components to create virtual forces/torques. Actually, the virtual forces/torques are generated by joint actuators when the virtual components interact between robots and environments. In this paper, a virtual model control is proposed to do the dynamic balance control of quadruped robots in trot gait. In each leg, virtual model control includes swing phase control of robots and stance phase counterparts. In whole body, based on the forces/torques distribution method between two stance legs, virtual model control is mainly about the attitude control containing roll, pitch and yaw. Then, an intuitive approach of velocity control is employed for the locomotion of quadruped robots. Based on the velocity planning and control, a trajectory tracking control approach is investigated by considering four factors: terrain complexity index, curvature radius of given trajectory, distance to terminal, and maximum velocity of quadruped robots. Finally, the effectiveness of proposed controllers is validated by co-simulations.

关键词： Virtual model control Dynamic balance control Trajectory tracking control Quadruped robots

来源：评论

学校读者我要写书评

暂无评论

A Guide for Human Walking Model and Control——Insights from Mechanical Property Analysis of Human Walking

A Guide for Human Walking Model and Control——Insights from...

引用

第32届中国控制与决策会议

This paper aims to establish the simplest human walking model and provide a guide for controlling biped ***,a human motion capture system was utilized to sample the motion data of human walking and a representative subject was ***,one of the simplest human walking model was set *** mechanical property analysis of human walking was done based on the established walking model via ***,the feasibility and effectiveness of proposed walking model were validated by numerical ***,the intrinsical relationships among human walking,biped walking model and control were discussed to provide a guide and insight for controlling biped robots.

关键词： Human Walking Model Biped Robots Leg Stiffness Legged Locomotion Model and Control

来源：评论

学校读者我要写书评

暂无评论

Holographic MIMO Communications with Arbitrary Surface Placements: Near-Field LoS Channel Model and Capacity Limit

arXiv

引用

arXiv 2023年

作者： Gong, Tierui Wei, Li Huang, Chongwen Yang, Zhijia He, Jiguang Debbah, Mérouane Yuen, Chau School of Electrical and Electronics Engineering Nanyang Technological University Singapore639798 Singapore College of Information Science and Electronic Engineering Zhejiang University Hangzhou310027 China Hangzhou310027 China State Key Laboratory of Robotics Shenyang Institute of Automation Chinese Academy of Sciences Shenyang110016 China Key Laboratory of Networked Control Systems Chinese Academy of Sciences Shenyang110016 China Institutes for Robotics and Intelligent Manufacturing Chinese Academy of Sciences Shenyang110169 China Technology Innovation Institute Masdar City Abu Dhabi9639 United Arab Emirates KU 6G Research Center Khalifa University of Science and Technology P O Box 127788 Abu Dhabi United Arab Emirates CentraleSupelec University Paris-Saclay Gif-sur-Yvette91192 France

Envisioned as one of the most promising technologies, holographic multiple-input multiple-output (H-MIMO) recently attracts notable research interests for its great potential in expanding wireless possibilities and achieving fundamental wireless limits. Empowered by the nearly continuous, large and energy-efficient surfaces with powerful electromagnetic (EM) wave control capabilities, H-MIMO opens up the opportunity for signal processing in a more fundamental EM-domain, paving the way for realizing holographic imaging level communications in supporting the extremely high spectral efficiency and energy efficiency in future networks. In this article, we propose a generalized EM-domain near-field channel modeling and study its capacity limit of point-to-point H-MIMO systems that equips arbitrarily placed surfaces in a line-of-sight (LoS) environment. Two effective and computational-efficient channel models are established from their integral counterpart, where one is with a sophisticated formula but showcases more accurate, and another is concise with a slight precision sacrifice. Furthermore, we unveil the capacity limit using our channel model, and derive a tight upper bound based upon an elaborately built analytical framework. Our result reveals that the capacity limit grows logarithmically with the product of transmit element area, receive element area, and the combined effects of 1/d2mn, 1/d4mn, and 1/d6mn over all transmit and receive antenna elements, where dmn indicates the distance between each transmit element n and receive element m. Particularly, 1/d6mn dominates in the near-field region whereas 1/d2mn dominates in the far-field region. Numerical evaluations validate the effectiveness of our channel models, and showcase the slight disparity between the upper bound and the exact capacity, which is beneficial for predicting practical system performance. Copyright © 2023, The Authors. All rights reserved.

关键词： MIMO systems

来源：评论

学校读者我要写书评

暂无评论

Extracting Semantic Knowledge from GANs with Unsupervised Learning

arXiv

引用

arXiv 2022年

作者： Xu, Jianjin Zhang, Zhaoxiang Hu, Xiaolin The School of Mathematics and Computer Science Panzhihua University The Department of Computer Science and Technology Tsinghua University China The Insitute of Automation Chinese Academy of Sciences The Centre for Artificial Intelligence and Robotics Hong Kong Innovation and Science Insitute Chinese Academy of Sciences Hong Kong The Department of Computer Science and Technology Institute for Artificial Intelligence State Key Laboratory of Intelligent Technology and Systems THU-Bosch JCML Center THBI IDG/McGovern Institute for Brain Research BNRist Tsinghua University Beijing China Beijing China

Recently, unsupervised learning has made impressive progress on various tasks. Despite the dominance of discriminative models, increasing attention is drawn to representations learned by generative models and in particular, Generative Adversarial Networks (GANs). Previous works on the interpretation of GANs reveal that GANs encode semantics in feature maps in a linearly separable form. In this work, we further find that GAN’s features can be well clustered with the linear separability assumption. We propose a novel clustering algorithm, named KLiSH, which leverages the linear separability to cluster GAN’s features. KLiSH succeeds in extracting fine-grained semantics of GANs trained on datasets of various objects, e.g., car, portrait, animals, and so on. With KLiSH, we can sample images from GANs along with their segmentation masks and synthesize paired image-segmentation datasets. Using the synthesized datasets, we enable two downstream applications. First, we train semantic segmentation networks on these datasets and test them on real images, realizing unsupervised semantic segmentation. Second, we train image-to-image translation networks on the synthesized datasets, enabling semantic-conditional image synthesis without human annotations. © 2022, CC BY-SA.

关键词： Unsupervised learning

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：