本文为美国俄亥俄州立大学（作者：J.N. (Nikki) Markiel）的博士论文，共216页。
生物驾驭环境的能力是生命的一大奥秘。人类，甚至从很小的时候，就可以获取周围环境的数据，确定物体是可移动的还是固定的，并在无穷小的时间跨度内识别出开阔的空间，分离出静止和非静止的物体，以最小的努力向另一个位置运动。在很长的一段时间内，人类可以仅仅基于地标的相对位置来回忆物体的位置，并重复导航任务。尽管在过去的半个世纪里进行了大量的研究工作，我们在自动驾驶汽车上模拟这一复杂过程的能力仍然不完整。
自主车辆依靠各种电子传感器来获取有关其环境的数据；挑战是将这些数据转换为支持导航目标的信息。历史上，许多传感器数据仅限于二维（2D）情况；最近的技术发展，如激光测距和3D声纳正在将数据采集扩展到全三维（3D）采集。本文的目标是开发一种算法，以支持在未知环境下，在动态运动目标存在的情况下，将三维测距数据转换为导航求解。该算法反映了利用三维测距技术实现自主导航的最初尝试之一，并提供了一个能够完成以下目标的系统：•环境中静态和非静态元素的分离•基于静态元素距离测量的导航
本研究扩展了三个主题的知识体系。
1） 第一种是开发一种通用方法，以从后续数据集中的m个特征中识别初始数据集中的n个特征，前提是这两个数据集都是通过3D测距传感器获取的。在试图链接重叠的数据集时实现这一目标，特别是在二维数据集方面，一直是一个难题。
2） 其次，提出了一种新的分割离散三维距离测量的方法。
3） 最后，本研究提出了一种方法，以支持先前因缺乏位置更新而无法在自主车辆上使用的环境中导航。这个问题在导航领域是众所周知的；虽然全球定位系统（GPS）提供了极好的位置信息，但它们的信号在各种条件下都可能变得不可用，例如室内或地下位置、密集的城市环境或干扰的信号环境。
目前对机器人操作的研究很少涉及未知环境下的操作概念，而且几乎从不尝试在非静态物体存在的情况下导航。将导航解决方案扩展到这些限制之外的能力进一步拓展了自主导航的可能性，并推进了导航领域的发展。目前的算法不能为不确定的时间段提供导航解算，但可以在不借助GPS定位的情况下，扩大导航的可行范围。虽然这项研究不可能宣称能解决自主导航的问题，但它代表了朝着开发机器模拟认知导航的愿景迈出的重要一步。由于我们只能站在巨人的肩膀上看得更远，希望这项研究有一天能使另一位研究人员看到真正的自主导航的成就。
The ability of living creatures to navigatetheir environment is one of the great mysteries of life. Humans, even from anearly age, can acquire data about their surroundings, determine whether objectsare movable or fixed, and identify open space, separate static and non-staticobjects, and move towards another location with minimal effort, ininfinitesimal time spans. Over extended time periods humans can recall thelocation of objects and duplicate navigation tasks based purely on relativepositioning of landmarks. Our ability to emulate this complex process inautonomous vehicles remains incomplete, despite significant research effortsover the past half century. Autonomous vehicles rely on a variety of electronicsensors to acquire data about their environment; the challenge is to transformthat data into information supporting the objective of navigation.Historically, much of the sensor data was limited to the two dimensional (2D)instance; recent technological developments such as Laser Ranging and 3D Sonarare extending data collection to full three dimensional (3D) acquisition. Theobjective of this dissertation is the development of an algorithm to supportthe transformation of 3D ranging data into a navigation solution within unknownenvironments, and in the presence of dynamically moving objects. The algorithmreflects one of the very first attempts to leverage the 3D ranging technologyfor the purpose of autonomous navigation, and provides a system which enablesthe ability to complete the following objectives: ·Separation of static and non-static elements in the environment ·Navigation based upon the range measurements of static elements This researchextends the body of knowledge in three primary topics. 1) The first is thedevelopment of a general method to identify n features in an initial data setfrom m features in a subsequent data set, given that both data sets areacquired via 3D ranging sensors. Accomplishing this objective, particularlywith respect to 2D datasets, has long been a difficult proposition whenattempting to link overlapping data sets. 2) Secondly, an innovativemethodology to segment a set of discrete 3D range measurements is presented. 3)Finally, the research develops a methodology to support navigation inenvironments previously infeasible for autonomous vehicles due to lack ofposition updates. This problem is well known in the navigation field; whileGlobal Positioning Systems (GPS) provide excellent positional information,their signals can become unavailable in a wide variety of conditions, such asindoor or underground localities, dense urban settings, or jammed signalenvironments. Current research in robotic manipulation rarely addresses theconcept of operations within an unknown environment, and virtually neverattempts navigation in the presence of non-static objects. The ability toextend the navigation solution beyond these limitations extends the possibilitiesfor autonomous navigation and advances the field of navigation. The currentalgorithm cannot provide a navigation solution for an indefinite time period;it can extend the feasible extent of navigation without benefit of GPSpositioning. While this research could not possibly claim to solve the problemof autonomous navigation, it represents an important step towards the vision ofdeveloping a machine to emulate cognitive navigation. As we see farther only bystanding on the shoulders of giants, it is hoped that this research willsomeday enable another researcher to see the achievement of true autonomousnavigation.
- 引言
- 三维数据与算法
- 算法实现
- 支撑技术回顾
- 实验数据采集
- 本文算法在实验数据和结果中的应用
- 结论与观察
附录 绝对定向：霍恩闭式解
更多精彩文章请关注公众号：