Robotics

Mag4D-SLAM: First large-scale outdoor geomagnetic SLAM dataset

18 km of synchronized multi-modal data with day/night traversals for robust localization.

Deep Dive

Mag4D-SLAM fills a critical gap in robotics research by providing the first large-scale outdoor dataset dedicated to geomagnetic SLAM. Previous magnetic datasets were limited to small indoor spaces and lacked the multi-modal synchronization, repeated traversals, and high-precision 6-DoF ground truth needed for systematic study. The dataset comprises 14 sequences totaling over 18 km of data from LiDAR, camera, IMU, tri-axis magnetometer, and GNSS, all with SE(3) ground-truth poses. Traversals were performed along structured campus trajectories under paired day/night conditions and in both forward and reverse directions, enabling analysis of magnetic field repeatability across sessions.

Beyond its scale, Mag4D-SLAM is designed to support three core research areas: drift-free global heading estimation, location-discriminative magnetic signatures for cross-session place recognition, and magnetic loop closure. The dataset opens new questions about how geomagnetic sensing can complement visual and LiDAR modalities or serve as a fallback under illumination changes, structural repetition, and GNSS denial. Accepted to IROS 2026, Mag4D-SLAM promises to accelerate progress in long-term autonomous navigation, especially for robots operating in challenging environments where traditional sensors fail.

Key Points
  • 14 sequences covering over 18 km with synchronized multi-modal sensors (LiDAR, camera, IMU, magnetometer, GNSS).
  • Captured under paired day/night conditions in forward and reverse directions for repeatability analysis.
  • Includes high-precision SE(3) ground-truth poses for rigorous SLAM evaluation.

Why It Matters

Enables reliable robot navigation in GNSS-denied areas using Earth's magnetic field as a robust, drift-free reference.

📬 Get the top 10 AI stories daily