LightLoc++: Sensor-Robust Representation Learning for Efficient Outdoor LiDAR Localization
LightLoc++ improves outdoor LiDAR localization by decoupling scene representation learning from prediction heads to save training time. It uses a synchronized multi-LiDAR dataset called SULID to create sensor-invariant backbone features. This allows the system to achieve state-of-the-art performance across different LiDAR sensors while requiring only minimal new-scene training, making it practical for real-time deployment.