GNSS/INS Sensor Fusion with On-Board Vehicle Sensors
GNSS/INS Sensor Fusion with On-Board Vehicle Sensors
复制标题
GNSS/INS 传感器与车载传感器融合
DOI：
10.33012/2020.17611
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
J. Jacox
中科院分区：
文献类型：
--
作者：
Ryan Dixon;M. Bobye;Brett Kruger;J. Jacox
自动驾驶汽车应用的一个关键要求是可靠、准确和鲁棒的定位（又名定位）解决方案。没有可靠的定位，关键的导航、规划和决策操作就无法进行。这意味着准确的定位必须无处不在-换句话说，在车辆预期运行的任何时间和任何地方都可靠可用。虽然全球导航卫星系统通常为绝对定位提供基础，但在无法直接观测到足够多的卫星时，它总是存在固有的可用性问题。
为了解决故障模式，可以通过称为传感器融合的技术将额外的互补传感器添加到整体导航解决方案中。传感器（例如惯性测量单元（伊穆斯）、相机、LiDAR、RADAR等）可以以减轻每个传感器的个别缺点并且提高整体鲁棒性和可靠性的方式来选择。虽然目前的自动驾驶汽车应用采用传感器融合技术，他们往往依赖于高性能的传感器，以满足精度要求。这些高性能传感器往往会导致比商业生产可接受的成本负担高得多的成本负担，因此使大规模自主变得过于昂贵。
本文将集中在开发成本较低的传感器已经在大多数现代车辆。这些传感器包括目前用于动态稳定性控制和车轮滑移检测的低分辨率里程计（ECU）和消费级伊穆斯单元。一种新的方法相结合的车辆速度，转向角，传输设置和多个里程计输入将沿着可实现的结果，同时在GNSS拒绝环境下运行。测试轨迹将模拟一个典型的停车场结构，有许多拐角和短的直线段。过滤器所需的唯一先验信息是轮距和轴距（车轮间距）。
与独立的GNSS/INS解决方案相比，在GNSS中断长达30分钟期间观察到90%的性能改进。此外，在相同的30分钟停电条件下，将多里程测量与单里程测量停电进行比较时，观察到高达50%的改善。除了GNSS中断性能之外，还将展示如何使用滤波器的额外输入来提高定位系统的保护级别，以允许更频繁地参与自主导航系统。
A key requirement of autonomous vehicle applications is a reliable, accurate, and robust positioning (aka localization) solution. Key navigation, planning and decision operations cannot happen without dependable positioning. This means that accurate positioning must be ubiquitous - in other words, reliably available at all times and in all places the vehicle is expected to operate. While Global Navigation Satellite Systems (GNSS) commonly provide the basis for absolute positioning, it always suffers from the inherent problem of availability whenever a direct view of enough satellites is not possible.
To address the failure mode, additional complementary sensors can be added to the overall navigation solution through a technique known as sensor fusion. Sensors such as inertial measurement units (IMUs), cameras, LiDARs, RADAR, etc. can be selected in such a way that the individual shortcomings of each sensor are mitigated, and the overall robustness and reliability are improved. Although current autonomous vehicle applications employ sensor fusion techniques, they tend to rely on highperformance sensors to meet the accuracy requirements. These high-performance sensors tend to induce a much higher cost burden than would be acceptable for commercial production, and therefore make mass autonomy too expensive.
This paper will focus on the exploitation of the lower cost sensors already available on most modern vehicles. These sensors include low resolution odometry (DMI) and consumer grade IMUs currently used for dynamic stability control and wheel slip detection. A novel approach for combining vehicle speed, steering angles, transmission settings and multiple odometry inputs will be presented along with achievable results while operating under a GNSS denied environment. The test trajectory will mimic a typical parking structure with many corners and short straight segments. The only apriori information required for the filter is the wheel track and wheelbase (separation of wheels).
A 90% performance improvement compared to the stand-alone GNSS/INS solution was observed during GNSS outages up to 30 minutes. Furthermore, up to a 50% improvement was observed when comparing between the multi-odometry vs single odometry outages during the same 30-minute outage condition. Beyond GNSS outage performance, it will be shown how the use of the extra input to the filter can improve protection levels of the positioning system to allow for more frequent engagement of the autonomous navigation system.