Integrity, protection levels, and alert limits; these three concepts, along with the rigorous testing of measurements involved, answer the questions, “Are we safe?” and “How confident are we in our safety?” for fully autonomous vehicles in development. Delivering confidence in autonomous car safety is a major hurdle for manufacturers.
Much research has been done to understand and realize safety in autonomous vehicles, whether the solution is still in the development stage, or when the solution is deployed across weather conditions. But testing a positioning solution in real scenarios and environments is critical to encouraging the future adoption of autonomous cars.
The Hexagon | NovAtel® Safety Critical Systems team had the opportunity to test and demonstrate confidence in safety at the Consumer Electronics Show 2020 show in January in Las Vegas. Team members drove a research car around the city as its positioning system continuously adhered to lane-level accuracy and provided positive proof of confidence in its safety levels.
As the research car drove public roads north of the Las Vegas strip, real-time visualization screens mapped the vehicle’s path, documenting the solution status, protection levels, and alert limits.
The team ensured safety with the software positioning engine from NovAtel using TerraStar X correction technology, an ST Teseo V/APP automotive-grade receiver chipset, an automotive-grade Inertial Measurement Unit (IMU), and a NovAtel GNSS-1500 automotive-grade antenna.
The software positioning engine from NovAtel processes raw code- and carrier-phase GNSS measurements from the receiver chipset, combines acceleration and angular rate measurements from the IMU, and relies on TerraStar X correction data to calculate a confident vehicle position.
The solution produces the lane-level accuracy of less than half a meter to form the foundation of future driverless development. The convergence time is less than one minute, translating to quick get-up-and-go use when starting the vehicle, another critical requirement.
How we measure safety in autonomous vehicles
NovAtel developed this safety-keyed approach with partners Stanford University and the Illinois Institute of Technology, two research institutions closely involved in formulating similar safety concepts in aviation. Together they developed threat models and safety monitors to quantify and define error bounds and protection levels to ensure integrity for positioning in autonomous automotive applications.
This Stanford Diagram indicates Integrity Performance, and provides a robust methodology for engineers to test protection levels, integrity, and confidence in the autonomous solution.
The concepts and models employ the following key terms:
Integrity is the measure of the trust given to the solution’s accuracy of navigation information. It means quantifying a position and putting an error bound around it with a numerical guarantee. Read more about integrity in our 2019 Velocity magazine issue.
The Horizontal Protection Level (HPL), a predicted upper bound on the maximum possible error, describes the measure of confidence in the lane-level accuracy of the vehicle’s position.
The Alert Limit (AL) refers to the maximum allowable position error for safe operation. If the performance level exceeds the alert limit, the positioning solution is no longer guaranteed to be flagged safe for use in the application.
Misleading Information (MI) is a position error greater than the protection level, but less than the alert limit. Hazardously Misleading Information (HMI) describes a position error greater than both the protection level and the alert limit.
The Integrity Risk (IR) outlines how often position errors can exceed the protection level. In automotive applications, the integrity risk defines what performance level is necessary based upon the vehicle’s protection level outputs in a safety critical application. Learn how NovAtel safety critical teams test integrity risk in this presentation and paper from the 2019 ION conference.
With these standards in place, critical safety teams can ensure confidence in autonomous car safety.
The team ensured safety with the software positioning engine from NovAtel using TerraStar X correction technology, an ST Teseo V/APP automotive-grade receiver chipset, an automotive-grade Inertial Measurement Unit (IMU), and a NovAtel GNSS-1500 automotive-grade antenna.
How we demonstrated autonomous vehicle safety
The positioning solutions delivered by the software positioning engine from NovAtel, computed by integrating measurements from the Teseo receiver, IMU data, and Terrastar X GNSS corrections, were more consistent and more accurate than those from the automotive-grade receiver also onboard the vehicle. The positioning engine’s errors remained under 0.5 meters in open-sky environments. Overall, the integrated solution delivered a 70% accuracy improvement over industry-standard chipset solutions.
The team tested their solution by traveling a more complicated path around Las Vegas through challenging GNSS-denied environments like underpasses. When the GNSS receiver has an obstructed view of the sky, protection levels adjust to allow for an improvement in estimating the maximum possible error or protection level. In a changing environment, this adjustment provides the user with a position solution that has both accuracy and integrity.
During testing in these challenging conditions, the system mitigated position errors through integrated advanced fusion algorithms using the IMU measurements to maintain an accurate solution. These errors were consistently less than those reported by the industry-standard chipset solution.
Where autonomous car safety takes us
Throughout the Las Vegas demonstration, the software positioning engine proved robust and capable of bounding the position error, reporting no instances of HMI or MI. Confidence in autonomous car safety was tested, demonstrated, and assured.
Built with advanced sensor fusion techniques, the software positioning engine from NovAtel and TerraStar X correction technology sets a new standard for safe, autonomous vehicles. Continuous lane-level positioning from highway conditions to the most challenging environments is not only possible, but it is also correctly verified within realistic error bounds and protection levels.