3D列印的普及讓民眾可利用簡易的建模軟體建立3D模型，但是如何建立可動模型對一般使用者而言卻是一道難題，因為必須考驗建模者對於模型設計與結構的熟悉程度，以及受到目前3D印表機的可印程度等因素限制。目前市面上的3D列印成型技術，以FDM最為普遍，優點是便宜、無毒等，而缺點則為精細度低及需印出支撐材，與其他3D列印技術相比限制較多，且失敗率相對較高。本論文的目標主要是將3D靜態(不可動)的肢體動物模型，以FDM印表機為輸出目標，產生出一體化關節可動模型。根據使用者輸入的模型與骨架，本系統會自動找出關節位置，利用關節點與模型頂點的距離計算各部位關節的半徑大小，並利用外積與旋轉矩陣將可動關節與肢體方向對齊，並調整模型至可動關節可嵌入的大小。使用者可藉由本系統所提供的直覺式操作介面，進行參數化調整，以印出不需組裝之可動模型，我們邀請了10位受試者，透過系統操作教學及任務，讓受試者學習如何使用此系統，並透過問卷的方式探討系統的優缺點。問卷的評分方式是採5分量表，從實驗結果顯示，系統整體有用性平均分數為4.5分，表示本系統能有效的幫助使用者建立可動關節模型；而易用性的平均分數是3.9分，代表本研究在介面設計上雖非重點，但仍有改進的空間；易學性方面的平均分數為4.5 分，表示本系統的操作對使用者是容易學習的。整體而言，實驗結果顯示，本研究所建立的可動關節模型系統已達到輔助使用者建立可動模型的目標，並證實了本研究的發展價值。 / The popularity of 3D printing has allowed people to design 3D models through common 3D modeling software. However, it is still difficult for a regular user to build a model with movable joints because most users are not familiar with mechanical design and it is a great challenge to design such a model that is printable with current 3D printing technology. FDM is the most popular type with the advantages of being cheap. However, its disadvantage is low precision, which make its failure rate higher than others. In this thesis, we aim to design a system that can take a static articulated model and convert it into a non-assemble model with movable joints that is printable on a FDM 3D printer. Our system can automatically find the positions of the joints according to the input mesh and skeleton and compute the radius of the maximal enclosing circle for each joint model. After aligning the joint model with the limb, the system can automatically adjust the size of the whole model such that the joint model can be embedded in the body model. A user can also tune system parameters through an intuitive interface to determine the orientation and limits of each joint. In order to evaluate our system, we invited ten persons to test user our system by completing assigned tasks and filling a usability survey. The survey is a questionnaire consisting of typical five-point Likert-scale items. The survey reveals that the usefulness of our system is 4.5, which means that our system can effectively help the users construct movable joint models. The score of ease-of-use is 3.9, which means that our user interface still have room for improvement although it is the current focus of our system. The score of ease-of-learning is 4.5, which means that our system is easy for the users to learn. In short, from the experimental results, we believe that our system has achieved the goal of providing a 3D modeling system that can assist users in building non-assemble moveable joint models that are printable on FDM 3D printers.

大部的類神經網路均為解決特定問題而設計，並非真正去模擬人腦的功能 ，在本論文中介紹一個模擬人類學習方式的類神經網路，稱為推理類神經 網路（The Reasoning Neural Network），其主要兩個組成為強記（ cram -ming）及推理（reasoning）部份，透過彈性的組合這兩個部份可 使類神經網路具有類似人類的學習程序。在本論文中介紹其中一個學習程 序並用四個實驗來評估推理類神經網路的績效，從實結果得知，推理類神 經網路能以合理的隱藏節點數（hidden nodes）達到學習的目標，並建立 一個網路內部表示方式（internal representation），及具有好的推理 能力（g eneralization ability）。 / Most of artification Neural Networks are designed to resolve spe -cific problems, rather than to model the brain. The Reasoning N -eural Network (RNN) that imitates the way of human learning is presented here. Two key components of RNN are the cramming and t -he reasoning. These components coulds be arranged flexibly to a -chieve the human-like learning procedure. One edition of the RNN used in experiments is introduces, and four different proble -ms are used to evaluate the RNN's performance. From simulation results, the RNN accomplishes the goal of learning with a reason -able number of hidden nodes, and evolves a good internal repres -entation and a generalization ability.

在製作動畫上，模擬人體的運動一直是困難的課題；但在如線上遊戲等急速成長的虛擬環境應用中，人物運動的動畫常是不可或缺的一環。過去在此方面的相關研究雖然為數不少，但大多數的系統皆只適用於某特定的地形或事先給定的腳步落點；能根據地形特徵而自動產生對應之行走運動者並不常見。本論文提議的系統，便是一個能即時模擬人體走路動作的運動計劃器。我們以反向關節運動的方式，分析人體在不平路面上行走時的運動特徵，並以貝茲曲線表示懸浮腿的運動軌跡。透過貝茲曲線控制點的調整，可以讓下半身的肢體避免碰觸到凸起的路面。其次，此系統也包含了腳步計劃的機制，讓虛擬人物能以行進效率為準則，計劃未來數步內保證可行的步伐。再者，我們根據實際測量的資料與觀察，找出行進過程中每個階段在時間分配上的差異，並利用製作動畫的原理，加入緩入與緩出的概念，以調整行走步伐的節奏，使動畫更具真實感。最後，我們將此模擬系統套用於「線上模擬」與「即時操控」兩種不同模式的應用系統，以驗證此系統之即時性與實用性。 / Simulating human motion has been an important and challenging topic in computer graphics for many years, especially after the booming of virtual environment applications such as on-line games. Although there has been much research on this topic, most previous systems are only capable of generating a realistic locomotion for a set of given footsteps on a flat ground in an off-line manner. The system we propose in this thesis is a lower-body motion simulator for humanoid capable of planning efficient footsteps and automatically generating collision-free locomotion in real time. First, we observe and analyze the motion characteristics of human walking and use Bézier curves to represent the trajectory of a floating leg during a stride. We use an inverse kinematics approach to compute the corresponding joint angles for a given leg trajectory. By adjusting the control points of the curve, we can change its shape to avoid collisions with the ground. Second, the system also includes a footstep planner that can generate successful and efficient gaits over an uneven terrain with an empirical energy consumption model. Third, according to observation and measured data, we use the “ease-in” and “ease-out” techniques and appropriate timing for each phase of a walking cycle to generate more realistic motions. Finally, we have applied this motion simulator to a virtual environment system with two types of operation modes: on-line simulation and real-time navigation which are verified the efficiency and practicability of such a system.