This work presents a real-time pose graph based Simultaneous Localization and Mapping (SLAM) system for automotive Radar. The algorithm constructs a map from Radar detections using the Iterative Closest Point (ICP) method to match consecutive scans obtained from a single, front-facing Radar sensor. The algorithm is evaluated on a range of real-world datasets and shows mean translational errors as low as 0.62 m and demonstrates robustness on long tracks. Using a single Radar, our proposed system achieves state-of-the-art performance when compared to other Radar-based SLAM algorithms that use multiple, higher-resolution Radars.

Video Attachment to "Real-Time Pose Graph SLAM based on Radar", presented at IEEE Intelligent Vehicles Symposium 2019