|Abstract:||Autonomous driving is predicated on the ability to measure a vehicle’s position, velocity, attitude and the environment around it. This complex task requires a large and varied array of sensors as well as algorithms to exploit their measurements. A major component of this is absolute positioning. Two technologies in particular are heavily relied upon to deliver the absolute position: GNSS and inertial sensors. Current production vehicles already contain both GNSS and inertial sensors, but these are used (and are therefore specified) for functions such as infotainment and stability control. Performance needed for these functions is significantly below what is required for safe lane-level navigation. This means the current sensors, and therefore positioning capabilities, are unable to satisfy the more rigorous requirements needed for autonomy. Future vehicles will need to include more advanced versions of both sensors, along with corrections and advanced fusion techniques, all at a cost level that remains amenable to mass-production. This is a significant challenge facing mass market autonomous driving. There are already many autonomous test vehicles on the roadways, though these provide mostly level 4 autonomy or lower , meaning they still require a human driver to be present and operate only in certain conditions. Additionally, these test vehicles are typically using sensors which are larger and more expensive than what is acceptable for mass production. The task of these test vehicles is to prove, and gain public acceptance of, the technology while mass marketable versions are developed. This paper focuses on the absolute positioning component using GNSS and INS technologies. Specifically on the inertial fusion using currently achievable performance using mass market sensors. A companion paper: “Precise Positioning for Automotive with Mass Market GNSS Receivers,”  focuses on GNSS capabilities. The synergy between GNSS and INS has been well documented and is now widely used in growing industries such as mobile mapping. Noise inherent in GNSS positions can be smoothed by the continuous INS solution and at the same time, low frequency inertial errors are controlled by the GNSS. However, when GNSS is unavailable, the INS errors grow rapidly when unconstrained. The speed at which the INS errors grow is much larger when using lower-performance mass market sensors, making the task of mitigating these conditions more difficult for the fusion solution. Clear trends can be seen in the results showing that the type of sensor has a significant impact on overall accuracy, but so too does the fusion technology used. The effects of this are apparent though modest in the suburban highway scenario, but are much larger in the more difficult urban canyon scenario. To evaluate the achievable performance combining mass market sensors and NovAtel’s SPAN inertial fusion technology, live data was collected and analyzed. These datasets were collected using different IMUs in two major scenarios; suburban highway and urban canyon. Additionally, a simplified loosely coupled inertial filter was run on the same data to evaluate the effect of the fusion algorithms on the resultant solution as final system performance depends on both the sensors and the integration. Results are compared against a post-processed solution using a survey grade dual frequency receiver paired with a navigation grade inertial measurement unit (IMU). Percentage of solutions with accuracy under the target of “lane level” positioning will be examined in each environment. It is clear in the results that autonomous highway driving can be very nearly achieved with GNSS and INS sensors alone, even with mass market sensors, when the right fusion technology is used. But in urban canyons, where GNSS satellites are often obstructed, or only visible via reflection, this is not the case. While sophisticated fusion technologies offer vast improvements, more development and likely more sensors will be required to meet the required accuracies at the extremely high confidence levels required for autonomy.|
Proceedings of the 31st International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2018)
September 24 - 28, 2018
Hyatt Regency Miami
|Pages:||1497 - 1509|
|Cite this article:||
Dixon, Ryan, Bobye, Michael, Kruger, Brett, Sinha, Anil, "GNSS/INS Fusion for Automotive with Mass Market Sensors," Proceedings of the 31st International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2018), Miami, Florida, September 2018, pp. 1497-1509.
ION Members/Non-Members: 1 Download Credit