RLPR: Radar-to-LiDAR Place Recognition via Two-Stage Asymmetric Cross-Modal Alignment for Autonomous Driving
Researchers have developed RLPR, a novel framework for radar-to-LiDAR place recognition designed to enhance all-weather autonomous driving capabilities. The system addresses the challenge of integrating radar data, which is resilient to adverse weather, with existing LiDAR maps, overcoming limitations in feature extraction and data scarcity. RLPR employs a dual-stream network for sensor-agnostic feature extraction and a two-stage asymmetric cross-modal alignment strategy to effectively map radar scans into LiDAR environments, demonstrating state-of-the-art accuracy and generalization. AI