published_or_final_version / Electrical and Electronic Engineering / Master / Master of Philosophy

published_or_final_version / abstract / Industrial and Manufacturing Systems Engineering / Doctoral / Doctor of Philosophy

在世界各地，自然或人為災難隨時可能發生。災難回應作為災難處理的重要環節顯得尤為重要，隨著科學技術的進步和提高，人們希望通過使用各種科學手段來提高災難的回應效率。機器人技術作為21世紀高科技結合的產物被廣泛應用於這一領域。一般情況下，設計者會採用功能集成的思想對機器人進行設計，他們的主要設計思想是根據自己對環境的理解和認知得到機器人的設計需求，然後針對設計需求，通過功能集成和疊加的方式來完成對機器人的設計，採用這種方式機器人一旦設計完畢，其功能也隨之確立並不可更改，這種設計思想是基於環境狀況的，即一旦災難現場的環境不符合預先的設定，機器人的執行能力將大幅下降，同時功能疊加的設計方式會產生功能與功能之間相互約束，影響其專業性。 / 本文介紹了一種基於分散式設計思想的全新設計理念，並且根據這一理念設計了一套基於任務需求的救援機器人系統。機器人系統不會根據設計者對災難現場的預先理解和認知而被一體化設計，相反根據"如何到達"和"如何操作"把機器人系統拆分成移動單元和操作單元兩個環節，針對每個環節分別設計了符合現場需求的通用移動模組和任務執行模組，救援人員可以根據災難現場的即時任務需求而迅速搭建出有針對性的機器人系統任務解決方案，和傳統的機器人系統相比，具適應性廣、靈活性高、針對性強等特點。 / 在本論文中，對三種通用的移動平臺和兩種通用的模組化關節以及一個快速連接器分別進行了結構設計、理論分析及樣機設計，並採用基本的通用模組，根據即時的任務需求構建出有針對性的多個機器人系統。實驗表明該機器人系統可以提供對災難環境有針對性的系統解決方案，具有一定容錯性、經濟性及災難環境的適應性。文章的創新點如下，首次針對于救援機器人提出分散式的設計思想，並以該思想為基礎設計了基於通用模組的救援機器人系統，針對不同任務對移動性能的不同要求設計了三種移動平臺，為滿足不同的救援操作要求設計了兩種模組化關節以及快速連接器。同時，文中為實際的地震救援任務提出了一套救援機器人系統解決方案。 / Natural and man-made disasters nowadays still present a large amount of risk. Disaster response is an important phase of disaster management, and the enhancement of its effectiveness and accountability has attracted an increasing amount of attention. Robots can help rescuers in doing this task because of its wide range of applications. In general, the rescue robot concept assumes one or more targeted tasks while design, and one or a set of robot(s) is/are designed by integrating different functions to accomplish those tasks. Once the design of a robot is finished, its function cannot be changed. However, this kind of design is environment-dependent, as once a disaster environment changes, the execution performance of the robot will reduce. Furthermore the function-integrated design concept may cause internal constraints between functions, and fail to provide a targeted solution for different disaster environments. / This dissertation introduces a novel design concept, based on which a requirement-oriented rescue robot system is developed. This design concept adopts a distributed strategy, according to which tasks are no longer seen as a whole but divided into two parts: traversability and operation. Several functional modules are designed to meet the different requirements of the two parts separately, and the entire robot system can be assembled using different functional modules according to the real-time requirements of the disaster environment. Compared with the traditional rescue robot system, this system can provide a more targeted solution for different disaster situations, and is more adaptable and flexible. / This dissertation details the basic functional modules, including three kinds of mobile bases for traversability and two sets of modular joints for operation, and analyzes a quick connector that makes the connection easier and more convenient. Several possible combinations of the rescue robot system are displayed to show how to construct a rescue robot system according to different requirements. This kind of rescue robot system can provide targeted solutions to different disaster tasks. Robustness is also enhanced, as the replacement of the functional modules is flexible and easy to overhaul. Furthermore, the functional modules can be decomposed and reused to make the robot system more economical. This dissertation makes several contributions. It presents a systematic solution for rescue robot, develops three mobile bases for high traversability and two kinds of modular joints and a quick connector for rescue operation. Furthermore, it also develops a rescue robot system for missions in earthquake. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Yang, Yong. / Thesis (Ph.D.) Chinese University of Hong Kong, 2014. / Includes bibliographical references (leaves 226-236). / Abstracts also in Chinese.

Robot localization is one of the most important subjects in the Robotics science. It is an interesting and complicated topic. There are many algorithms to solve the problem of localization. Each localization system has its own set of features, and based on them, a solution will be chosen. In my thesis, I want to present a solution to find the best estimate for a robot position in certain space for which a map is available. The thesis started with an elementary introduction to the probability and the Gaussian theories. Simple and advanced practical examples are presented to illustrate each concept related to localization. Extended Kalman Filter is chosen to be the main algorithm to find the best estimate of the robot position. It was presented through two chapters with many examples. All these examples were simulated in Matlab in this thesis in order to give the readers and future students a clear and complete introduction to Kalman Filter. Fortunately, I applied this algorithm on a robot that I have built its base from scratch. MCECS-Bot was a project started in Winter 2012 and it was assigned to me from my adviser, Dr. Marek Perkowski. This robot consists of the base with four Mecanum wheels, the waist based on four linear actuators, an arm, neck and head. The base is equipped with many sensors, which are bumper switches, encoders, sonars, LRF and Kinect. Additional devices can provide extra information as backup sensors, which are a tablet and a camera. The ultimate goal of this thesis is to have the MCECS-Bot as an open source system accessed by many future classes, capstone projects and graduate thesis students for education purposes. A well-known MRPT software system was used to present the results of the Extended Kalman Filter (EKF). These results are simply the robot positions estimated by EKF. They are demonstrated on the base floor of the FAB building of PSU. In parallel, simulated results to all different solutions derived in this thesis are presented using Matlab. A future students will have a ready platform and a good start to continue developing this system.

Providing mobile robots with the ability to manipulate objects has, despite decades of research, remained a challenging problem. The problem is approachable in constrained environments where there is ample prior knowledge of the environment layout and manipulatable objects. The challenge is in building systems that scale beyond specific situational instances and gracefully operate in novel conditions. In the past, researchers used heuristic and simple rule-based strategies to accomplish tasks such as scene segmentation or reasoning about occlusion. These heuristic strategies work in constrained environments where a roboticist can make simplifying assumptions about everything from the geometries of the objects to be interacted with, level of clutter, camera position, lighting, and a myriad of other relevant variables. The work in this thesis will demonstrate how to build a system for robotic mobile manipulation that is robust to changes in these variables. This robustness will be enabled by recent simultaneous advances in the fields of big data, deep learning, and simulation. The ability of simulators to create realistic sensory data enables the generation of massive corpora of labeled training data for various grasping and navigation-based tasks. It is now possible to build systems that work in the real world trained using deep learning entirely on synthetic data. The ability to train and test on synthetic data allows for quick iterative development of new perception, planning and grasp execution algorithms that work in many environments. To build a robust system, this thesis introduces a novel multiple-view shape reconstruction architecture that leverages unregistered views of the object. To navigate to objects without localizing the agent, this thesis introduces a novel panoramic target goal architecture that takes previous views of the agent to inform a policy to navigate through an environment. Additionally, a novel next-best-view methodology is introduced to allow the agent to move around the object and refine its initial understanding of the object. The results show that this deep learned sim-to-real approach performs best when compared to heuristic-based methods in terms of reconstruction quality and success-weighted-by-path-length (SPL). This approach is also adaptable to the environment and robot chosen due to its modular design.

This thesis seeks to advance the field of microrobotics by leveraging Bayesian principles and computational tools to design system parameters for information gain and microrobot propulsion. Inspired by living cells, the development of mobile robots on the micron scale (microrobots) promises new capabilities for advancing human health, renewable energy, and environmental sustainability. To help pave the way towards this goal we develop practical recipes for applying computational and analytical tools to physics-based dynamical models of our microrobot experiments. We apply methods of criticism and validation to identify robust models for the motion of magnetic particles at curved interfaces, and identify optimal conditions for propulsion in our model system. We then develop tools for identifying optimal experimental conditions for most efficiently learning model parameters. By studying microscale actuation in depth, we seek to provide a roadmap of how to apply these computational tools to other microrobot design challenges, accelerating the scientific process. In Chapter 1, we focus on the actuation of magnetic particles adsorbed at curved liquid interfaces by external fields, a phenomenon that can be utilized for applications such as droplet mixing or propulsion. To optimize these behaviors, the development and validation of predictive models are essential. We employ Bayesian data analysis as a principled approach to infer model parameters from experimental observations, assess the capabilities of candidate models, and select the most plausible among them. Specifically, we identify and validate a dynamical model which accounts for the effects of gravity and tilting of the particle, a Janus sphere, at the interface. We show how this favored model can predict complex particle trajectories with micron-level accuracy across the range of driving fields considered. Chapter 2 builds on this modeling to develop the optimal properties of a mobile liquid droplet, driven by an adsorbed magnetic particle. This configuration enables the design of responsive emulsions, which can be actuated by a magnetic field. This work develops the properties of such a swimmer and validates our findings with an experimental realization of a ferromagnetic ellipsoid adsorbed onto a stationary water droplet in decane. Accounting for geometric differences, the model developed in the previous chapter is demonstrated to be accurate for this new system. We find that the configuration of the magnetic moment of our ellipsoid prohibits swimming of the assembly, but if it can be modified during fabrication, propulsion is possible. In Chapter 3 we show how automated experiments based on Bayesian inference and design can accurately and efficiently characterize another microscale propulsion system, the acoustic field within resonant chambers used to propel acoustic nanomotors. Repeated cycles of observation, inference, and design are guided by a physical model that describes the rate at which levitating particles approach the nodal plane. We show how this iterative process serves to discriminate between competing hypotheses and efficiently converges to accurate parameter estimates using only a few automated experiments. This work demonstrates how Bayesian methods can learn the parameters of nonlinear hierarchical models used to describe video microscopy data of active colloids. Finally, the forward-looking perspective in Chapter 4 illustrates how best to leverage these techniques and models to provide a path forward for self-guided microrobots. Existing microrobots based on field-driven particles rely on knowledge of the particle position and the target destination to control particle motion through fluid environments. These external control strategies are challenged by limited information and global actuation, where a common field directs multiple robots with unknown positions. We discuss how time-varying magnetic fields can be used to encode self-guided behaviors of magnetic particles conditioned on local environmental cues. Programming these behaviors is framed as a design problem: we seek to identify the design variables (e.g. particle shape, magnetization, elasticity, stimuli-response) that achieve the desired performance in a given environment. We discuss strategies for accelerating the design process using the methods developed in this thesis—including automated experiments, computational models, and statistical inference—as well as other approaches such as machine learning. Based on the current understanding of field-driven particle dynamics and existing capabilities for particle fabrication and actuation, we argue that self-guided microrobots with potentially transformative capabilities are close at hand. This research offers a unique contribution by demonstrating the practicality and efficiency of Bayesian computational methods in microrobot design, and provides a template that is applicable anywhere that physics-based dynamical models can be used to guide experimental effort.