Ground truth poses

Ground-truth Poses

For each scene, we provide the ground-truth trajectories in CSV format. These trajectories include the translational (tx) and rotational (ro) components that make up the homogenous transformation matrix representing the robot's poses.

Each row in the CSV file corresponds to a single timestamped pose and is formatted as follows:

• Time (timestamp): The timestamp of the data point that indicates when the pose was recorded. It is provided in the standard ROS Unix epoch.

• Position Components (position_x, position_y, position_z): This is the absolute position coordinates of the robot in 3D space. It describes its coordinates in the global and fixed reference frame.

• Orientation Components (orientation_x, orientation_y, orientation_z, orientation_w): The quaternion representation of the robot's orientation. This captures the robot's rotation in space.

Example Pseudo ground-truth data

Geographic map and the reconstructed maps in one of the scenes in EnvoDat. (a) Satellite image with annotated trajectory and waypoints P1 - P5, where we slowed the robot for multiple measurements to refine pose estimates. (b) Dense LiDAR inertial odometry pseudo-ground truth reconstruction. (c) Reconstructed map from raw LiDAR point clouds for scale precision. (d) Feature point distribution from the reconstructed map, showing high-density regions where the robot slowed and low-density areas of faster movement. The robot’s trajectory, overlaid on the density bars, is colour-coded based on the cumulative distance transitioning from blue (start) to red (end). This data supports the correlative analysis of the impact of feature densities on SLAM algorithm performance.