Enabling Confidence in Autonomous Vehicles
Lane-Level Accuracy with Safety and Integrity Demonstrated at CES 2020
Conducted by NovAtel’s Safety Critical Systems team during the Consumer Electronics Show (CES) January 7-10,2020, a research development vehicle platform demonstrated continuous lane-level positioning, with a proven solution to address safety and integrity.
“Integrity is essential to the autonomous future,” explained Michael Ritter, President of Hexagon’s Autonomy & Positioning division, which includes NovAtel and AutonomouStuff brands. “We’ve got a long history of delivering navigation integrity. We’re taking a similar approach to integrity for autonomous driving—because in reality we have to be even more accurate than in aviation.”
Truth and Consequences
The NovAtel software positioning engine demonstration was built using a Hexagon | AutonomouStuff Research Platform.
The key components included:
- The NovAtel software positioning engine using TerraStar X correction technology;
- An ST Teseo V/APP automotive grade receiver chipset;
- Analog Devices’ ADIS 16505 inertial measurement unit(IMU); and
- The NovAtel GNSS-1500 automotive grade antenna.
The combination delivers an improvement of up to 70% in accuracy over industry standard chipset solutions.
The NovAtel software positioning engine processes raw code and carrier phase GNSS measurements from the receiver chipset, acceleration and angular rate measurements from the IMU, and TerraStar X correction data to calculate the position of the vehicle with lane-level accuracy. With the utilization of TerraStar X technology, the solution convergence time is reduced to under one minute.
In addition to the NovAtel software positioning engine, the demonstration platform incorporated a high-accuracy reference truth system that benchmarked the solutions from the automotive grade receiver and the software positioning engine output.
The reference system consisted of a NovAtel PwrPak7D™ receiver, a μIMU-IC IMU, and two NovAtel GNSS-502 antennas. The reference system utilized HxGN SmartNet network RTK corrections to compute the most accurate reference trajectory for ground truthing.
On the Road
During CES 2020, the research vehicle was operated on public roads north of the Las Vegas strip.
Inside the vehicle, real-time visualization screens mapped the vehicle path, documenting the solution status, protection levels and alert limits. The graph in the upper section of Figure 1plots the error from the ST Teseo chipset measurement engine (ME) relative to the truth system. The green line is the associated errors from the NovAtel software positioning engine, which computes a solution through a combination of GNSS corrections, IMU data and the measurements from the Teseo receiver. The result is more consistent and accurate measurements with errors always less than 0.5 meters in open-sky environments.
The visualization screen provided participants of the demonstration with clear verification of the INS status(in this case Good) and the number of visible satellites. Note the horizontal protection level (HPL) is at 2.1 m, below the 3.0m alert limit (AL), thus the vehicle lane-level accuracy remains within the safe green bubble.
Figure 2 reflects the NovAtel software positioning engine performance compared to standard automotive-grade GNSS for an entire demonstration path around Las Vegas. The points of significant error growth in the graph are indicative of the vehicle traveling through challenging GNSS-denied environments, e.g. underpasses. The NovAtel positioning engine errors are constrained during these times, especially when compared to the traditional automotive grade hardware, through integrated advanced fusion algorithms, which use the IMU measurements to maintain an accurate solution.
The Margin of Error
Integrity is about quantifying a position and putting an error bound around it with a numerical guarantee—essentially the measure of trust in the solution. NovAtel provides integrity of its software positioning engine through protection levels.
The company has developed its approach through a research partnership with some of the leaders in integrity in academia, including Stanford University and the Illinois Institute of Technology. Through the partnership, threat models and safety monitors have been developed, much like those used to define aviation integrity, to quantify and define error bounds or protection levels and assure integrity in GNSS measurements for autonomous automotive applications.
As noted earlier, in automotive applications the PL, or the upper bound on the predicted maximum possible error in position, is the measure of confidence in the lane-level accuracy of the vehicle’s position.
The dataset from the Las Vegas demonstration (Figure 3) is a comparison of position error versus protection levels along the Las Vegas public highway route.
If the protection level exceeds the alert limit (AL), the positioning solution is no longer deemed safe for use in that application. The associated integrity risk is defined as how often position errors exceed the PL. In automotive applications, the PL allows the user to be confident in the lane-level accuracy of the vehicle’s position.
The areas where the protection level is the largest in Figure 3 are related to instances where the vehicle has encountered a challenging GNSS environment, traveling under an overpass, and obstructing the view of the sky. In these scenarios the estimation of the maximum possible error, or protection level, must increase as GNSS signals become obstructed or unavailable.
However, despite the larger protection level, the position error of the software positioning engine remains constrained due to NovAtel’s advanced algorithms in dead reckoning and inertial sensor fusion. The Google Earth imagery (Figure 4) shows the conditions at one of these instances (t=509625) where the vehicle is travelling under a series of underpasses along the highway route. The protection levels respond accordingly as the environment changes, providing the user with a position solution that is not only accurate, but has integrity.
Throughout the Las Vegas demonstration, NovAtel’s software positioning engine proved robust and capable of bounding the position error reporting no hazardously misleading
information (HMI) or misleading information (MI).
Built on advanced sensor fusion techniques, NovAtel’s software positioning engine sets a new standard for moving forward on the road to safe autonomous vehicles. Through this engine, continuous lane-level positioning from highway conditions to the most challenging of environments is not only possible, but defined within realistic error bounds and
Stay tuned for more advances in 2020 on the road to truly autonomous driving under all conditions…with confidence.
Read the full PDF here: