【作者】 王维；

【导师】 贺汉根；

【作者基本信息】 国防科学技术大学 ， 控制科学与工程， 2011， 博士

【摘要】 在基于虚拟现实的人机工程和维修工程等研究领域中,虚拟人运动控制技术是研究重点之一,它包含两个关键性的问题:运动规划问题和运动合成问题。本文以武器装备预先研究项目“基于虚拟现实的维修性XXX系统”、国家“863”课题“高品质重型商用车集成开发先进技术”,以及国家探月工程中的“月面巡视探测器导航与控制”为研究背景,从理论和应用两个层面围绕上述关键性问题开展研究,主要包括:基于快速扩展随机树(Rapidly-exploring Random Tree, RRT)的路径规划算法,RRT算法概率完备性的证明,虚拟人手臂及上躯干的操控规划方法,以及基于运动基元的虚拟人运动合成方法。本文的研究成果将提高虚拟人运动仿真的效率和逼真性,在人机工程、维修工程等领域具有重要理论意义和应用价值。本文的主要成果和创新点如下:1.提出了自适应多树RRT算法,通过改进“桥测试采样”算法准确标识配置空间中的狭窄通道,再利用增强学习中的随机组合优化算法构建自适应树选择策略,有效克服了路径规划中的“狭窄通道”问题,比传统RRT算法的规划效率有显著提高。本文路径规划算法解决了“基于虚拟现实的维修性XXX系统”项目中的虚拟人行走规划问题。典型实例表明本文算法可推广运用于月球车路径规划问题。2.在RRT算法的完备性理论研究中,首先证明了经典RRT算法和自适应多树RRT算法的概率完备性。然后,通过对随机优化组合算法εt-greedy性质的研究,证明了自适应多树RRT扩展过程的最优性。3.提出了基于RRT的虚拟人操控规划求解方法。建立了统一形式的可行配置空间数学模型,分别给出了基于RRT的手臂操控规划算法和上躯干操控规划算法。在上躯干操控规划中,利用RRT实现快速逆向运动学求解,通过“采样-投影”算法解决操控对象的运动约束问题。本文设计并实现了跨平台的虚拟人运动仿真软件EDMAN(Engineering Digital MAN),解决了“基于虚拟现实的维修性XXX系统”项目中的虚拟人维修作业仿真问题,研究结果表明操控规划算法的应用显著提高了虚拟维修仿真的效率。4.提出了基于HMM(Hidden Markov Model)运动基元的虚拟人运动合成框架。在运动基元建模过程中,首先采用加权PCA(Principal Component Analysis)降维算法和基于高斯混合模型(Gaussian Mixture Model, GMM)的聚类算法来分割运动片段,然后通过分层HMM聚类算法,实现运动基元的分类与建模。在运动检索与合成过程中,利用基于HMM相似性度量的动态规划算法,实现运动片段的检索和时序排列。本文的运动合成方法解决了“高品质重型商用车集成开发先进技术”项目中的虚拟人上下车运动仿真问题,为该项目的完成做出了贡献。更多还原

【Abstract】 Motion control for virtual human is one of the most important aspects in such areas as VR (Virtual Reality) -based ergonomics and maintenance engineering. It has two key problems: motion planning and motion synthesis. Both the theoretic and the practical research on the above two problems are carried out in this thesis, funded by Weapon and Equipment Advanced Research project“VR-based Maintainability XXX System”, National“863”project“Advanced Integrated Developing Technology for High-quality Heavy Business Truck”, and National Lunar Exploring project“Navigation and Control for Lunar Rover Vehicle”. The main content include: the RRT (Rapidly-exploring Random Tree) based path planning algorithms, proof of probabilistic completeness of the RRT algorithms, the manipulation planning methods for the arm work and the upper-body work, as well as the motion synthesis methods based on motion primitives. The outcomes of the research will improve the efficiency and the realism of the virtual human motion simulation, which is of great importance to both the theoretic research and the application in the field of ergonomics and maintenance engineering.The main contributions and innovations are as following:1. An adaptive multi-RRTs algorithm has been proposed to effectively address the narrow passage problems, in which a Bridge-Test algorithm has been improved to accurately identify narrow passages in configuration space, and an adaptive tree selection strategy has been established by random combinatory optimization algorithm from reinforcement learning. The performance of the proposed algorithm is dramatically improved compared with that of the traditional RRT algorithms. The RRT-based path planning algorithm has been applied to address the path planning problem for the walk task of virtual human in the project“VR-based Maintainability XXX System”. Typical applications show that the proposed algorithm can be generalized to address the path planning problem of lunar rover vehicle.2. In the research of probabilistic completeness theory of RRT algorithm, traditional RRT algorithms and the adaptive multiple-RRT algorithm have been proved to be probabilistic complete firstly. Next, the expansion process of the adaptive multiple-RRT algorithm has been proved to be optimal based on the analysis of the property of the random combinatory optimization strategyεt-greedy.3. A uniform RRT based manipulation planning framework for virtual human has been proposed. The math model of a uniform feasible configuration space is described, based on which the RRT-based arm and upper-body manipulation planning algorithms have been proposed. Regarding the problem of upper-body manipulation planning, the RRT algorithm is applied to rapidly solve the inverse kinematics problem. Thereafter, a sample-project algorithm has been proposed to solve the problem of motion constraints of manipulated object. EDMAN (Engineering Digital MAN), a cross-platform virtual human motion simulation software has been designed and implemented to address the problem of the virtual human maintenance task simulation in the project“VR-based Maintainability XXX System”. The experimental results show that the application of the proposed algorithms has dramatically improved the efficiency of the virtual maintenance simulation.4. A virtual human motion synthesis framework has been proposed based on HMM motion primitives. In the process of the motion primitive modeling, firstly, a weighted PCA algorithm, as well as a clustering algorithm based on Gaussian Mixture Model (GMM), is adopted to segment the motion capture data into motion clips. Secondly, a hierarchical HMM clustering algorithm has been proposed to implement the classification and the modeling of the motion primitives. In the process of motion retrieval and synthesis, a dynamic programming algorithm based on HMM similarity measurement has been proposed to retrieve the motion clips and sort them according to the time sequence. The above motion synthesis method has been applied to address the problem of the truck ingress/egress motion simulation, which makes contributions to the accomplishment of the project“Advanced Integrated Developing Technology for High-quality Heavy Business Truck”.更多还原