Mostafa Sakr, Adel Moussa, Walid Abdelfatah, Mohamed Elsheikh, Aboelmagd Noureldin and Naser El-Sheimy, Profound Positioning Inc., Canada

View Abstract Sign in for premium content


This paper demonstrates an integrated radar-based localization system developed by Profound Positioning Inc. (PPI) that supports the Society of Automotive Engineers (SAE) driving automation level 4 applications with a focus on automated valet parking in degraded visual environments in covered parking garages. The system integrates automotive radars and dead reckoning technologies supported by high-definition (HD) maps to offer decimeter level positioning accuracy for robust automated valet parking. The system addresses the current limitations in the state-of-the-art localization systems for autonomous vehicles in structured areas. Profound Positioning Inc. distributes the localization module to its customers as a software library with a simple application programming interface (API) for easy integration into various localization-based autonomous applications. The current solution for vehicle-based navigation systems is dominated by global navigation satellite systems (GNSS), which could provide positioning services with various levels of accuracies in an open sky environment. However, the GNSS solution is regularly disturbed in urban canyons, under thick foliage, or inside underground or covered areas. For human vehicle operators, the intermittent nature of GNSS positioning does not pose a significant challenge on their navigation abilities, and the operator will be able to drive the vehicle without reliable positioning information safely. However, for the next generation of automated and connected vehicles, reliable and precise location information is crucial to the proper operation of the vehicle. Researchers in academia and the industry have dedicated a significant effort to develop localization solutions that can operate reliably under a variety of weather and road conditions. The two most prominent technologies for localization depend on light detection and ranging (LiDAR) scanners or visible light cameras, in conjunction with detailed HD road maps and specialized geospatially annotated maps for LiDAR or camera features. However, localization using LiDAR or camera technologies introduce another set of limitations and constraints to the system. For example, vision-based localization systems are affected by degraded and challenging weather conditions, such as the presence of fog, rain or snow and typical cameras will fail altogether under low or no light conditions. The current generation of LiDAR scanners are still expensive and bulky and contains mechanical moving parts, increasing the challenges of integrating these sensors into mainstream vehicles. While LiDAR sensors can operate under low light conditions, they still fail under extreme weather conditions. Profound Positioning radar-based localization system was tested and validated in single and multi-floor underground parking lots. In the first case, the vehicle travels on a single floor and performs a parking maneuver on the far end of an underground parking location, then returns to the initial point. In the second case, the vehicle starts on the lower floor of a multi-floor parking building, then travels to the upper-floor before returning to the initial floor and performing a reverse parking maneuver. In both scenarios, the localization solution was consistent with the stringent requirements of driving automation level 4 applications. This paper presents an innovative radar-based localization system developed by Profound Positioning Inc., that integrates radar and dead reckoning solutions, for autonomous vehicles navigating structured environments in the absence of the GNSS solution. The methodology alleviates the limitations of LiDAR-based and camera-based systems by exploiting the unique characteristics of automotive radar sensors. The system is capable of handling single and multi-floor scenarios by overcoming the common shortcomings of any radar-based system, which are the sparsity of the point cloud and the poor performance at low speed by integrating automotive radars and dead reckoning technologies. In the demonstrated results, the system can maintain the strict requirements of driving automation level 4 applications.