elevation-mapping

Learning robot-specific traversability from a short manual drive — No manual labels needed. [RA-L '24]

C++17/ROS 2 robot centric elevation mapping with probabilistic confidence bounds. Implements Kalman filtered height fusion, 3D covariance propagation, Mahalanobis based multi surface handling, empirical CDF map fusion, and ray tracing visibility checks. Includes wheeled odometry noise model, Gazebo sim, and RMSE/coverage evaluation framework.

Preprocessed and split version of the SSBH remote sensing dataset for building height estimation, including RGB composites, height maps, and building masks with train/valid/test manifests and ready-to-train scripts.