InPipe-SLAM: A Multi-Sensor Benchmark Dataset for SLAM Evaluation in Straight and Elbowed Pipe Environments

InPipe-SLAM is a multi-sensor benchmark dataset designed for the evaluation of localization, odometry, and SLAM algorithms in confined pipe environments. The dataset targets the specific challenges of in-pipe robotic navigation, including cylindrical geometric degeneracy, limited visual texture, repetitive structures, reduced sensor overlap, occlusions, and trajectory drift in straight and elbowed pipe sections. The dataset contains real-world ROS 2 recordings collected with a custom mobile in-pipe robot equipped with RGB-D sensing, LiDAR sensing, wheel encoders, and inertial measurements. The released sequences cover straight pipe sections and elbow geometries, including moderate and sharper curvature configurations. These scenarios allow the evaluation of SLAM robustness under both weakly constrained straight motion and curved pipe transitions. Ground-truth and reference trajectories are provided when available using external tracking, fiducial markers, and/or odometry-based reference generation depending on the sequence. The dataset also includes calibration information, topic descriptions, and documentation required for reproducible benchmarking. The dataset is intended to support the comparison of visual, LiDAR-based, LiDAR-inertial, RGB-D, and wheel-odometry-based localization methods. It can be used with standard trajectory evaluation metrics such as Absolute Trajectory Error (ATE) and Relative Pose Error (RPE). The dataset is distributed as compressed ROS 2 bag archives. Users are encouraged to cite this dataset when using it for research, benchmarking, or publication.