Reliable Localization Using Multi-sensor Fusion for Automated Valet Parking Applications

Multi-sensor fusion is the key to enabling reliable and accurate localization for automated valet parking (AVP) systems that can autonomously navigate in highly structured indoor environments. The current state-of-the-art in AVP systems depends on light detection and ranging (LiDAR) systems, visible-light cameras, or infrastructure-based solutions. LiDAR systems can provide high positioning accuracy across a variety of operating conditions. However, they are generally more expensive, have moving parts that create more room for error, and the processing of its 3D point cloud is computationally demanding. As for cameras, the performance of any camera-based localization system is susceptible to variations in lighting conditions and fail altogether under degraded visual environments. This paper presents a real-time, radar-based localization library developed by Profound Positioning Inc. (PPI). PPI’s localization library utilizes a multi-sensor fusion approach for real-time vehicle localization in structured environments such as underground parking lots. PPI’s algorithm integrates the onboard motion sensors, a set of mid-range automotive radars, and two-dimensional (2D) high definition (HD) maps. PPI’s radar-based localization library is capable of maintaining a decimeter-level accuracy, and it is validated and evaluated using real-life scenarios in an indoor parking lot.