Real-time Onboard Multi-sensor Navigation Systems for Unmanned Aerial Vehicles in GPS-challenging Environments

无人机在遥感、态势感知、移动测绘等领域的应用越来越广泛。它们具有携带传感器、处理器、发射器和成像设备的能力，并具有独特的全方位机动特性。无人机的安全运行需要精确的定位和导航系统。目前无人机定位技术主要由全球导航卫星系统(GNSS)和高速率星上惯性测量单元(IMU)组成。不幸的是，由于卫星信号阻塞和IMU误差，GNSS/IMU技术在密集的城市和室内地区等闭塞环境中严重失败。这一限制目前阻碍了无人机在复杂的障碍性环境中执行低空任务。 为了填补这一空白，在过去的十年里，利用车载视觉/距离传感器进行准确的定位和导航得到了广泛的研究。然而，有几个挑战阻碍了当前系统的安全采用和商业利用。一个主要的缺点是缺乏稳健性，因为现有的传感器融合方法在动态变化的环境中处理具有不同高度非线性噪声特性的海量异质传感器流的能力有限。另一个主要缺点是昂贵的计算。基于视觉/距离的组合定位系统必须以极高的速率处理海量数据扫描，以确保对无人机平台的稳定控制。 小型无人机还面临着额外的挑战。它们真正的三维机动和悬停对估计方法和实时性能提出了几个挑战。此外，转子表现出的高振动和电磁干扰会立即导致难以建模的传感器误差。加速度计和陀螺仪在高振动下都会受到很大的噪声，磁力计在靠近电磁干扰源的地方不能正常工作。 在申请人以前多传感器定位系统工作的基础上，这项研究计划将专注于提高小型无人机平台在复杂障碍的GNSS挑战/拒绝环境中的定位和导航能力。该研究将探索将人工智能方法和模糊理论与传统的信号处理和估计技术相结合，开发适用于无人机动力学的新的运动模型，以及能够处理异质低成本噪声传感器的新型非线性自适应传感器融合方法。车载实时处理将使用支持图形处理单元“GPU”的处理器进行优化。研究成果可以很容易地应用于其他新兴技术，如自动驾驶汽车、自动机器人和增强现实。

Unmanned Aerial Vehicles (UAVs) are being increasingly utilized in wide range of applications in remote sensing, situation awareness, and mobile mapping. They have the ability to carry sensors, processors, transmitters, and imaging equipment and they have unique omnidirectional maneuvering characteristics. Safe operation of UAVs requires accurate positioning and navigation systems. Present UAV positioning technology is dominated by Global Navigation Satellite Systems (GNSS) integrated with high-rate onboard inertial measurement units(IMU). Unfortunately, GNSS/IMU technology fails significantly in obstructed environments such as dense urban and indoor areas due to satellite signal blockage and IMU errors. This limitation currently prevents UAVs from being used to perform low-altitude tasks in complex obstructed environments. To fill this gap, utilization of onboard vision/range sensors for accurate positioning and navigation has been extensively studied in the last decade. However, there are several challenges that prevent current systems from being safely adopted and commercially utilized. One major drawback is lack of robustness due to the limited ability of current sensor fusion methods to handle massive streams of heterogeneous sensors with different highly nonlinear noise characteristics in a dynamically changing environment. Another major drawback is the expensive computation. Vision/range-based integrated positioning systems have to process massive scans of data at extremely high rates to assure stable control of UAV platforms. Small scale UAVs have additional challenges. Their truly 3D maneuvering and hovering impose several challenges on the estimation methods and real-time performance. In addition, the high vibration and electromagnetic interference exhibited by rotors cause immediate sensors errors that are difficult to model. Both accelerometers and gyroscopes suffer from large noise under high vibration and magnetometers do not work properly in close proximity to electromagnetic interference sources. Building on the applicant's previous work in multi-sensor positioning systems, this research program will focus on advancing the positioning and navigation capabilities of small scale UAV platforms in complex obstructed GNSS-challenging/denied environments. The research will explore the integration of artificial intelligence methods and fuzzy theory with conventional signal processing and estimation techniques to develop new motion models that are suitable for UAV dynamics and novel nonlinear adaptive sensor fusion methods that can handle heterogeneous low-cost noisy sensors. Onboard real-time processing will be optimized using Graphical-Processing-Unit “GPU”-enabled processors. The research outcomes can be readily applied in other emerging technologies such as self-driving cars, autonomous robots, and augmented reality.