检索结果-南通市图书馆

Ensemble successor representations for task generalization in offline-to-online reinforcement learning

Science China(Information Sciences) 2024年第7期67卷 240-255页

作者： Changhong WANG Xudong YU Chenjia BAI Qiaosheng ZHANG Zhen WANG space Control and inertial technology research Center Harbin Institute of Technology Shanghai Artificial Intelligence Laboratory Shenzhen research Institute of Northwestern Polytechnical University School of Cybersecurity Northwestern Polytechnical University

In reinforcement learning(RL), training a policy from scratch with online experiences can be inefficient because of the difficulties in exploration. Recently, offline RL provides a promising solution by giving an initialized offline policy, which can be refined through online interactions. However, existing approaches primarily perform offline and online learning in the same task, without considering the task generalization problem in offline-to-online adaptation. In real-world applications, it is common that we only have an offline dataset from a specific task while aiming for fast online-adaptation for several tasks. To address this problem, our work builds upon the investigation of successor representations for task generalization in online RL and extends the framework to incorporate offline-to-online learning. We demonstrate that the conventional paradigm using successor features cannot effectively utilize offline data and improve the performance for the new task by online fine-tuning. To mitigate this, we introduce a novel methodology that leverages offline data to acquire an ensemble of successor representations and subsequently constructs ensemble Q functions. This approach enables robust representation learning from datasets with different coverage and facilitates fast adaption of Q functions towards new tasks during the online fine-tuning phase.Extensive empirical evaluations provide compelling evidence showcasing the superior performance of our method in generalizing to diverse or even unseen tasks.

关键词： offline reinforcement learning online fine-tuning task generalization successor representations ensembles

来源：

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Standing wave drift of hemispherical resonator with quality factor non-uniformity under a ring electrode excitation

引用

Chinese Journal of Aeronautics 2022年第1期35卷 160-172页

作者： Yan HUO Zeyu WANG Shunqing REN Zhennan WEI Guoxing YI Changhong WANG space cotrol and inertial technology research centre Harbin Institute of TechnologyHarbin 150080China HIT(Anshan)Institute of Industrial technology Anshan 114000China

Hemispherical Resonator Gyroscope(HRG)is a classical high precision Coriolis Vibration Gyroscope(CVG),which performs attitude estimation of carrier by detecting the precession of standing wave of resonator,thus,the drift of standing wave of resonator has a great influence on the output accuracy of gyroscope,where the quality factor non-uniformity of resonator is one of main error sources.Ring electrode is a classical excitation structure of HRG because the standing wave can precess freely under its excitation,which makes the gyroscope have more accurate scale factor,larger measurement range and better dynamic characteristics.In this paper,the equations of motion of an ideal resonator excited by a ring electrode are derived by the elastic thin shell theory and Lagrange mechanical principle,then the corresponding equivalent mechanical model is established.According to the“average method”,it can be seen that the ideal resonator excited by the ring electrode works in integral mode,and any position in the circumferential direction of resonator can be a working point,which means that the quality factor non-uniformity has a great effect on the drift of standing wave.Therefore,the equations of motion of resonator with quality factor non-uniformity under the ring electrode excitation are deduced by the equivalent mechanical model,and the drift model of standing wave is established by the“average method”,it can be found that both the amplitude of quality factor non-uniformity and angle between the“inherent damping axis”and antinode axis of standing wave can affect the drift rate of standing wave.Moreover,the drift model indicates that if the input angular rate does not reach the threshold,the precession angular rate of standing wave will appear“self-locking”phenomenon,that is,the gyroscope will lose the integral effect.

关键词： Hemispherical resonator Theory of elastic thin shell Motion equations Average method Quality factor non-uniformity “Self-locking”phenomenon

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Ion and electron motions in the outer electron diffusion region of collisionless magnetic reconnection

引用

Earth and Planetary Physics 2024年第3期8卷 472-478页

作者： Cong Chang QuanMing Lu San Lu Kai Huang RongSheng Wang Deep space Exploration Laboratory School of Earth and Space SciencesUniversity of Science and Technology of ChinaHefei 230026China Chinese Academy of Sciences(CAS)Center for Excellence in Comparative Planetology CAS Key Laboratory of Geospace EnvironmentHefei 230026China Collaborative Innovation Center of Astronautical Science and technology Harbin 150001China

Two-dimensional particle-in-cell simulations are performed to study the coupling between ion and electron motions in collisionless magnetic reconnection.The electron diffusion region(EDR),where the electron motions are demagnetized,is found to have a two-layer structure:an inner EDR near the reconnection site and an outer EDR that is elongated to nearly 10 ion inertial lengths in the outflow direction.In the inner EDR,the speed of the electron outflow increases when the electrons move away from the X line.In the outer EDR,the speed of the electron outflow first increases and then decreases until the electrons reach the boundary of the outer EDR.In the boundary of the outer EDR,the magnetic field piles up and forms a depolarization front.From the perspective of the fluid,a force analysis on the formation of electron and ion outflows has also been investigated.Around the X line,the electrons are accelerated by the reconnection electric field in the out-of-plane direction.When the electrons move away from the X line,we find that the Lorentz force converts the direction of the accelerated electrons to the x direction,forming an electron outflow.Both electric field forces and electron gradient forces tend to drag the electron outflow.Ion acceleration along the x direction is caused by the Lorentz force,whereas the pressure gradient force tends to decelerate the ion outflow.Although these two terms are important,their effects on ions are almost offset.The Hall electric field force does positive work on ions and is not negligible.The ions are continuously accelerated,and the ion and electron outflow velocities are almost the same near the depolarization front.

关键词： collisionless magnetic reconnection electron diffusion region force analysis particle-in-cell simulation

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Posture Control of Legged Locomotion Based on Virtual Pivot Point Concept

引用

Journal of Bionic Engineering 2023年第6期20卷 2683-2702页

作者： Hao Sun Junjie Yang Yinghao Jia Changhong Wang space Control and inertial technology research Center Harbin Institute of TechnologyHarbin150001China

This paper presents a novel control approach for achieving robust posture control in legged locomotion,specifically for SLIP-like bipedal running and quadrupedal bounding with trunk stabilization.The approach is based on the virtual pendulum concept observed in human and animal locomotion experiments,which redirects ground reaction forces to a virtual support point called the Virtual Pivot Point(VPP)during the stance phase.Using the hybrid averaging theorem,we prove the upright posture stability of bipedal running with a fixed VPP position and propose a VPP angle feedback controller for online VPP adjustment to improve performance and convergence speed.Additionally,we present the first application of the VPP concept to quadrupedal posture control and design a VPP position feedback control law to enhance robustness in quadrupedal bounding.We evaluate the effectiveness of the proposed VPP-based controllers through various simulations,demonstrating their effectiveness in posture control of both bipedal running and quadrupedal bounding.The performance of the VPP-based control approach is further validated through experimental validation on a quadruped robot,SCIT Dog,for stable bounding motion generation at different forward speeds.

关键词： Legged robots Bionic control Virtual pivot point Posture stability Dynamic locomotion

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Spring-Loaded Inverted Pendulum Hopping via Hybrid Averaging and Control Lyapunov Function

引用

Journal of Bionic Engineering 2023年第1期20卷 291-307页

作者： Hao Sun Junjie Yang Yinghao Jia Qinghua Li Changhong Wang space Control and inertial technology research Center Harbin Institute of TechnologyHarbin150001China

The Spring-Loaded Inverted Pendulum(SLIP)has been regarded as a canonical model for hopping and running dynamics of legged robots.This paper presents a novel control of the actuated-SLIP hopping on unknown terrains.We propose that in the neighborhood of the desired stable hybrid limit cycle,the local dynamical behavior of a hybrid system can be expressed by a set of phase coordinates and transverse coordinates.Under some acceptable assumptions,the hybrid averaging theorem is applied on the SLIP non-integrable dynamics to simplify the controller design.Using the inherent symmetry of SLIP dynamics,a control Lyapunov function-based hybrid averaging controller is developed to ensure the exponential stability of the desired gait orbit.This results in a set of linear constraints on the control signal,which can be readily solved by a quadratic programming optimization.Furthermore,a novel method is introduced to improve the robustness against unknown disturbances through the online constraint adjustment.The proposed controller is evaluated in various simulations,demonstrating the SLIP hopping on diverse terrains,including flat,sin-wave,and unregular terrains.The performance of the approach is also validated on a quadruped robot SCIT Dog for generating dynamic gaits such as pronking.

关键词： Legged robots-Spring-Loaded Inverted Pendulum Control Lyapunov function Dynamic hopping

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Multi-Source Adaptive Selection and Fusion for Pedestrian Dead Reckoning

引用

IEEE/CAA Journal of Automatica Sinica 2022年第12期9卷 2174-2185页

作者： Yuanxun Zheng Qinghua Li Changhong Wang Xiaoguang Wang Lifeng Hu NORINCO GROUP Air Ammunition research Institute Harbin 150030China the research Center of space Control and inertial technology Harbin Institute of TechnologyHarbin 150001China research Center of space Control and inertial technology Harbin Institute of TechnologyHarbin 150001China

Accurate multi-source fusion is based on the reliability, quantity, and fusion mode of the sources. The problem of selecting the optimal set for participating in the fusion process is nondeterministic-polynomial-time-hard and is neither sub-modular nor super-modular. Furthermore, in the case of the Kalman filter(KF) fusion algorithm, accurate statistical characteristics of noise are difficult to obtain, and this leads to an unsatisfactory fusion result. To settle the referred cases, a distributed and adaptive weighted fusion algorithm based on KF has been proposed in this paper. In this method, on the basis of the pseudo prior probability of the estimated state of each source, the reliability of the sources is evaluated and the optimal set is selected on a certain threshold. Experiments were performed on multi-source pedestrian dead reckoning for verifying the proposed algorithm. The results obtained from these experiments indicate that the optimal set can be selected accurately with minimal computation, and the fusion error is reduced by 16.6% as compared to the corresponding value resulting from the algorithm without improvements.The proposed adaptive source reliability and fusion weight evaluation is effective against the varied-noise multi-source fusion system, and the fusion error caused by inaccurate statistical characteristics of the noise is reduced by the adaptive weight evaluation.The proposed algorithm exhibits good robustness, adaptability,and value on applications.

关键词： Adaptive reliability evaluation adaptive weight evaluation Kalman filter(KF) multi-source fusion optimal set selection

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Machine learning algorithm partially reconfigured on FPGA for an image edge detection system

引用

Journal of Electronic Science and technology 2024年第2期22卷 48-68页

作者： Gracieth Cavalcanti Batista Johnny Oberg Osamu Saotome Haroldo F.de Campos Velho Elcio Hideiti Shiguemori Ingemar Soderquist Division of Electronic and Embedded Systems KTH Royal Institute of TechnologyStockholm 16440Sweden Electronic Engineering Division Aeronautics Institute of TechnologySao Josédos Campos SP 12228-900Brazil Laboratory of Applied Computing and Mathematics National Institute for Space ResearchSao Josédos Campos SP 12227-900Brazil Department of C4ISR Institute of Advanced StudiesSao Josédos Campos SP 12228-001Brazil Saab AB Linkoping 58188Sweden

Unmanned aerial vehicles(UAVs)have been widely used in military,medical,wireless communications,aerial surveillance,etc.One key topic involving UAVs is pose estimation in autonomous navigation.A standard procedure for this process is to combine inertial navigation system sensor information with the global navigation satellite system(GNSS)signal.However,some factors can interfere with the GNSS signal,such as ionospheric scintillation,jamming,or spoofing.One alternative method to avoid using the GNSS signal is to apply an image processing approach by matching UAV images with georeferenced images.But a high effort is required for image edge extraction.Here a support vector regression(SVR)model is proposed to reduce this computational load and processing time.The dynamic partial reconfiguration(DPR)of part of the SVR datapath is implemented to accelerate the process,reduce the area,and analyze its granularity by increasing the grain size of the reconfigurable region.Results show that the implementation in hardware is 68 times faster than that in software.This architecture with DPR also facilitates the low power consumption of 4 mW,leading to a reduction of 57%than that without DPR.This is also the lowest power consumption in current machine learning hardware implementations.Besides,the circuitry area is 41 times smaller.SVR with Gaussian kernel shows a success rate of 99.18%and minimum square error of 0.0146 for testing with the planning trajectory.This system is useful for adaptive applications where the user/designer can modify/reconfigure the hardware layout during its application,thus contributing to lower power consumption,smaller hardware area,and shorter execution time.

关键词： Dynamic partial reconfiguration(DPR) Field programmable gate array(FPGA)implementation Image edge detection Support vector regression(SVR) Unmanned aerial vehicle(UAV) pose estimation

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Motion equations of hemispherical resonator and analysis of frequency split caused by slight mass non-uniformity

引用

Chinese Journal of Aeronautics 2020年第10期33卷 2660-2669页

作者： Yan HUO Shunqing REN Guoxing YI Changhong WANG space Control and inertial technology research centre Harbin Institute of TechnologyHarbin 150080China

The mass non-uniformity of hemispherical resonator is one of reasons for frequency split,and frequency split can cause gyroscope to drift.Therefore,it is of great significance to analyze the relationship between mass non-uniformity and frequency split,which can provide a theoretical basis for mass balance of imperfect resonator.The starting point of error mechanism analysis for gyroscope is the motion equations of resonator.Firstly,based on the Kirchhoff-Love hypothesis in the elastic thin shell theory,the geometric deformation equations of resonator are deduced.Secondly,the deformation energy equation of resonator is derived according to the vibration mode and relationship between the stress and strain of hemispherical thin shell.Thirdly,the kinetic energy equation of resonator is deduced by the Coriolis theorem.Finally,the motion equations of resonator are established by the Lagrange mechanics principle.The theoretical values of precession factor and natural frequency are calculated by the motion equations,which are substantially consistent with the ones by the finite element method and practical measurement,the errors are within a reasonable range.Simultaneously,the varying trend of natural frequency with respect to the geometrical and physical parameters of resonator by the motion equations is consistent with that by the finite element analysis.The above conclusions prove the correctness and rationality of motion equations.Similarly,the motion equations of resonator with mass non-uniformity are established by the same modeling method in case of ignoring the input angular rate and damping,and the state equations with respect to the velocity and displacement of vibration system are derived,then twonatural frequencies are solved by the characteristic equation.It is concluded that one of reasons for frequency split is the 4 th harmonic of mass non-uniformity,and thus much attention should be paid to minimizing the 4 th harmonic of mass non-uniformity in the course of mass balancing for imperfect resonator.

关键词： Hemispherical resonator Lagrange mechanics principle Theory of elastic thin shell Motion equations Mass non-uniformity Frequency split

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Comparison of strapdown inertial navigation algorithm based on rotation vector and dual quaternion

引用

Chinese Journal of Aeronautics 2013年第2期26卷 442-448页

作者： Wang Zhenhuan Chen Xijun Zeng Qingshuang space Control and inertial technology research Center Harbin Institute of Technology

For the navigation algorithm of the strapdown inertial navigation system, by comparing to the equations of the dual quaternion and quaternion, the superiority of the attitude algorithm based on dual quaternion over the ones based on rotation vector in accuracy is analyzed in the case of the rotation of navigation frame. By comparing the update algorithm of the gravitational velocity in dual quaternion solution with the compensation algorithm of the harmful acceleration in traditional velocity solution, the accuracy advantage of the gravitational velocity based on dual quaternion is addressed. In view of the idea of the attitude and velocity algorithm based on dual quaternion, an improved navigation algorithm is proposed, which is as much as the rotation vector algorithm in computational complexity. According to this method, the attitude quaternion does not require compensating as the navigation frame rotates. In order to verify the correctness of the theoretical analysis, simulations are carried out utilizing the software, and the simulation results show that the accuracy of the improved algorithm is approximately equal to the dual quaternion algorithm.

关键词： Dual quaternion inertial navigation system Navigation algorithm Quaternion Rotation vector

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Study on damage mechanism and damage distribution of the rear plate under impact of debris cloud

引用

Defence technology（防务技术） 2024年第5期35卷 151-167页

作者： Chenyang Wu Xiaowei Chen Qiguang He State Key Laboratory of Explosion Science and technology Beijing Institute of TechnologyHaidian DistrictBeijing 100081China School of Mechatronical Engineering Beijing Institute of TechnologyHaidian DistrictBeijing 100081China Institute of Fluid Physics China Academy of Engineering PhysicsMianyang 621900SichuanChina

The debris cloud generated by the hypervelocity impact(HVI)of orbiting space debris directly threatens the spacecraft.A full understanding of the damage mechanism of rear plate is useful for the optimal design of protective structures.In this study,the hypervelocity yaw impact of a cylindrical aluminum projectile on a double-layer aluminum plate is simulated by the FE-SPH adaptive method,and the damage process of the rear plate under the impact of the debris cloud is analyzed based on the debris cloud structure.The damage process can be divided into the main impact stage of the debris cloud and the structural response of the rear plate.The main impact stage lasts a short time and is the basis of the rear plate damage.In the stage of structure response,the continuous deformation and inertial motion of the rear plate dominate the perforation of the rear plate.We further analyze the damage mechanism and damage distribution characteristics of the rear plate in detail.Moreover,the connection between velocity space and position space of the debris cloud is established,which promotes the general analysis of the damage law of debris cloud.Based on the relationship,the features of typical damage areas are identified by the localized fine analysis.Both the cumulative effect and structural response cause the perforation of rear plate;in the non-perforated area,cratering by the impact of hazardous fragments is the main damage mode of the rear plate.

关键词： Damage of rear plate Debris cloud Secondary impact Hypervelocity yaw impact FE-SPH adaptive method

来源：

维普期刊数据库博看期刊

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

欢迎您,

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

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

请选择保存的检索档案：

请选择收藏分类：

通借通还

欢迎您,

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

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

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：