Driverless Vehicles Require Cocktail of Positioning Tech

In order for autonomous cars to truly allow hands-free driving in all types of scenarios, improving sensor-based accuracy is a critical challenge to ensure the vehicles know exactly where they are.

Increasingly, fused vehicle sensor information, like car-mounted sensors, cameras, radar and global positioning engines (GNSS), will help boost localization performance and map learning capabilities. “There are two major challenges for positioning systems in autonomous vehicle use cases,” said Jon Auld, vice-president of engineering and safety critical systems at Hexagon Positioning Intelligence. “The first is how to effectively and efficiently take advantage of all the various sensors that are used in an autonomous vehicle to deliver a highly accurate and highly available localization solution.”

He explained GNSS is a key component of providing absolute position but it has availability challenges and needs to be used in conjunction with other sensors to provide an overall solution that still maintains that accurate solution but in very challenging environments. “The second challenge is to deliver a safety parameter, typically known as a protection level, with the position solution to allow the overall system to trust the localization data when integrating it into a safety critical autonomous driving system,” Auld said.

Currently radar, LiDAR and cameras are used to provide the distance to objects that surround the vehicle and, when integrated with complimentary technology such as ultrasonic, inertial motion, digital maps, GNSS can act as the sixth sense to deliver the positioning performance required by autonomous vehicles. “This is all about satellite availability and connectivity to the correction data source,” Auld explained. “In open-sky conditions we will have good satellite visibility and, therefore, good solution availability.”

He noted that in conditions of urban canyons there is a need to extend the less available satellite solution with other relative sensors such as inertial units, cameras and LiDAR and other technologies. “The overall positioning solution can combine the inputs from all these to extend the availability of the high accuracy solution in conditions of lower availability GNSS,” Auld said.

Dan Dempsey, senior director of business development at inertial navigation specialist Aceinna, explained densely populated cities with many tall buildings present multi-path and GNSS signal loss issues. “The tall buildings of a downtown setting will restrict the angle of visibility to the sky,“ Dempsey explained. “A potential location on the top of a tall building can see 50 satellites in the sky. Down at street level, the vehicle may only be able to see two or three satellites that are directly over-head.”

He said this limited number of satellites greatly increases the chances that the signal fix will be dropped. Additionally, the satellite signals may reflect off buildings as they descend to the car’s receiver at street level. The reflection may result in a satellite signal being received twice, one directly and the other with a time delay after reflection off a building. The receiver may not know which signal to accept and could result in errors in the perceived position of the vehicle. “To get precision down to the centimeter-level accuracy required by sophisticated ADAS systems and autonomous cars, one must correct for the ever-changing atmospheric distortions of the incoming GNSS signals,” Dempsey said.

Along with real-time kinematic correction data vehicles requiring highly accurate positioning or velocity can deploy post-processing technology. Post-processing maximizes the accuracy of the solution by processing previously stored GNSS and inertial measurement unit (IMU) data forward and reverse in time and then combining the results.

For quite some time, inertial navigation systems, which use rotation and acceleration information from IMUs have complemented GNSS to provide a more robust positioning system. When GNSS signals are compromised owing to line-of-sight obstructions to the satellites, the INS, which is initialized and calibrated by good GNSS, can continue to estimate positions and velocities with a high degree of accuracy for short windows of time until line-of-sight is re-established.

“Ultimately, the goal is the same – position the vehicle accurately on the road and detect, recognize and avoid objects in the path of the vehicle,” said Mike Housholder, director of location software and services business development at TDK InvenSense. “The underlying requirements don’t change, however, the conditions and specific complexities are different.”

Housholder explained that at one level, scaling up and system cost are concerns for this to reach mass-market. “At the macro level, the combination of always-available precise positioning, dependable object recognition, reliable vehicle connectivity, and fail-proof decision making from an advanced AI that is equal to or better than a human’s across all types of driving scenarios remains the most daunting challenge for the industry to tackle,” he concluded.