M2DCD-SLAM: A Multi-Modal Campus Dataset for Robust SLAM Evaluation in Semi-Structured Outdoor Environments

M2DCD-SLAM is a multi-modal campus dataset for robust SLAM evaluation in semi-structured outdoor environments. The dataset contains ten complete handheld multi-modal sequences. Six general campus sequences cover representative outdoor environments, including college building areas, library surroundings, academic building areas, vegetation-rich paths, bridge structures, pond-side sections, and waterfront routes. Two additional route pairs provide dynamic and reference traversals of the same campus routes, with pedestrian and cyclist interactions identified in the dynamic sequences. Each sequence provides a ROS bag containing RGB images from an Intel RealSense D455 camera, Livox MID-360 LiDAR measurements, and IMU measurements from the built-in IMU of the Livox MID-360. The dataset also provides raw RTK-GNSS records, RTK trajectories converted to local ENU coordinates in TUM format, calibration and configuration files, FAST-LIVO2 baseline trajectories, trajectory-evaluation results, and auxiliary scripts for RTK-to-TUM conversion and trajectory evaluation. The RTK-GNSS trajectories correspond to the RTK antenna position. The measured LiDAR-to-RTK lever arm is provided separately for users who require LiDAR-center compensation. Prior to public release, all RGB image streams were manually reviewed for identifiable personal information. Identifiable faces, readable license plates, and personal identifiers in the six general campus sequences were anonymized by image blurring. The four paired-route sequences were also manually reviewed, and no content requiring additional blurring was identified.