Multiscale, Beamlet-Based Data Processing for the Solution of Shortest-Path Problems with Applications to Embedded Vehicle Autonomy

该奖项的研究目标是将多分辨率分析思想应用于自动驾驶车辆的路径和运动规划。特殊的基函数被用来有效地编码车辆运行环境中的所有可能的路径。该方法背后的关键思想是路径具有比周围空间更低的维度，因此，比标准2D和3D单元分解更有效地编码问题数据是可能的。使用的主要数学工具是beamlet，它的独特性质除了尺度和局部性外，还捕捉方向性。其结果是降低了嵌入式控制系统的算法计算复杂度。交付成果包括对环境中的障碍物进行编码的新软件，通过数值模拟对方法进行演示和验证，在同行评议的出版物中以及通过在国内和国际会议上的演讲记录研究结果，以及工程学生教育。如果成功，结果将使当前基于动态规划的搜索算法减少一个数量级，从而提高航空航天、军事和工业应用中遇到的一大类小规模车辆和系统的智能水平，这些车辆和系统必须在其性能范围内运行(例如，无人机、高速自主轮式车辆、机器人等)。而且它们的机载计算资源有限。所提出的直线编码方案可能会产生具有自动图像边缘检测能力的新型传感器，从而推动图像处理技术的发展。通过利用与当地行业建立的合作伙伴关系，成果将过渡到商用自动驾驶(主要是空中)车辆。研究生和本科生将通过课堂教学和参与拟议的研究活动，从这项研究的结果中受益。

The research objective of this award is to apply multiresolution analysis ideas for path- and motion planning of autonomous vehicles. Special basis functions are used to efficiently encode all possible paths inside the environment the vehicle operates in. The key idea behind the approach is the fact that paths have lower dimensionality than the ambient space, and hence, a more efficient encoding of the problem data than standard 2D and 3D cell decompositions is possible. The main mathematical tool used is beamlets, whose unique properties capture directionality, in addition to scale and locality. The result is reduced computational complexity algorithms for embedded control systems. Deliverables include new software to encode obstacles in the environment, demonstration and validation of the approach via numerical simulations, documentation of research results in peer-reviewed publications and via presentations at national and international conference, and engineering student education.If successful, the results will allow the reduction of current dynamic-programming based search algorithms by an order of magnitude, thus allowing increased levels of intelligence for a large class of small-scale vehicles and systems encountered in aerospace, military, and industrial application, which must operate at the limits of their performance envelope (e.g., unmanned aerial vehicles, high-speed autonomous wheeled vehicles, robots, etc), and which have limited on-board computational resources. The proposed line encoding scheme may lead to new sensors with built-in capability of automatic image edge detection, promoting the state-of-the-art in image processing technology. By leveraging established partnerships with local industry, the results will be transitioned to commercial autonomous (primarily aerial) vehicles. Graduate and undergraduate students will benefit from the results of this research through classroom instruction and through their involvement in the proposed research activities.