About MRCD

Our dataset was recorded in ROS2 Humble. We recorded a total of 8 sequences in a campus environment. The dataset features a variety of challenging outdoor sequences for outdoor mobile robots. All sequences can be downloaded from the Download page.

Hardware Specifications

ROS2 TF-Tree

Available Topics

Topic Name Type Frequency (Hz)
/ground_truth nav_msgs/msg/Path 30
/platform/dynamic-joint_states control_msgs/msg/DynamicJointState 50
/platform/joint_states sensor_msgs/msg/JointState 50
/platform/odom nav_msgs/msg/Odometry 50
/platform/odom/filtered nav_msgs/msg/Odometry 50
/cmd_vel_teleop geometry_msgs/msg/Twist 20
/teleop/joy sensor_msg/Joy 180
/tf tf2_msgs/msg/TFMessage 228
/tf_static tf2_msgs/msg/TFMessage 0.001
/scan sensor_msg/LaserScan 10
/velodyne_points sensor_msgs/msg/PointCloud2 10
/velodyne_packets velodyne_msgs/msg/VelodyneScan 10
/rs_front/rs_front/aligned_depth_to_color/camera_info sensor_msgs/msg/CameraInfo 15
/rs_front/rs_front/aligned_depth_to_color/image_raw sensor_msgs/msg/Image 15
/rs_front/rs_front/color/camera_info sensor_msgs/msg/Image 30
/rs_front/rs_front/color/image_raw sensor_msgs/msg/Image 30
rs_front/rs_front/color/metadata realsense2_camera_msgs/msg/Metadata 30
/rs_front/rs_front/depth/camera_info sensor_msgs/msg/CameraInfo 15
/rs_front/rs_front/depth/color/points sensor_msgs/msg/PointCloud2 15
/rs_front/rs_front/depth/image_rect_raw sensor_msgs/msg/Image 15
/rtk_gnss/tcpfix sensor_msgs/msg/NavSatFix 5
/rtk_gnss/tcpvel sensor_msgs/msg/TwistStamped 5
/zed_node/imu/data_raw sensor_msgs/msg/Imu 200
/zed_node/imu/mag sensor_msgs/msg/MagneticField 50
/zed_node/left_raw/camera_info sensor_msgs/msg/ CameraInfo 30
/zed_node/left_raw/image_raw_color sensor_msgs/msg/Image 30
/zed_node/right_raw/camera_info sensor_msgs/msg/ CameraInfo 30
/zed_node/right_raw/image_raw_color sensor_msgs/msg/Image 30
/zed_node/odom nav_msgs/msg/Odometry 30
/zed_node/path_map nav_msgs/msg/Path 2
/zed_node/path_odom nav_msgs/msg/path 2
/zed_node/point_cloud/cloud_registered sensor_msgs/msg/PointCloud2 10
/zed_node/pose geometry_msgs/msg/PoseStamped 30
/zed_node/pose_with_covariance geometry_msgs/msg/PoseWithCovarianceStamped 30

Ground Truth

As you see in the table above, MRCD contains a topic for discrete gound truth. Our ground truth is based on B-splines (for further information we recommend the works of Sommer & Usenko and Nguyen). For external downloads, both continuous- and discrete-time ground truth sampled with 10 Hz, please visit our Download page.

Discrete time ground truth is recommended for traditional benchmark evaluation. For this purpose, you can use sampled-10Hz.csv files. A snippet is given below.

# t, x, y, z, qx, qy, qz, qw
1733232244.216223,-11.381495,419.831974,13.050092,-0.001421,0.002207,0.986825,0.161768
1733232244.316222,-11.380007,419.838220,13.049208,-0.001452,0.002527,0.986845,0.161641
1733232244.416222,-11.376395,419.840031,13.049060,-0.001391,0.002309,0.986843,0.161660
1733232244.516222,-11.376447,419.839133,13.048638,-0.001201,0.002303,0.986867,0.161513

Continuous-time Ground Truth is provided in the form of B-splines in files named spline.csv, which contain a list of control points. Please find a snippet below. For reading the spline.csv files, we refer to the example Python script in MCD. In addition, please find our docker image for running the script in our Download page.

Dt: 0.01, Order: 4, Knots: 18630, MinTime: 1733232244.216222, MaxTime: 1733232430.486222, OtrItr: 0
0,1733232244.216222,-11.38167635118966,419.831709573794,13.04996049864827,-0.001487913908807536,0.002282587900610606,0.9867893981649208,0.1619853683947698
1,1733232244.226222,-11.38149436097049,419.8319735971247,13.05009187779725,-0.00142456389288367,0.002201660148280774,0.9868312052885267,0.1617321723687045
2,1733232244.236222,-11.38131456689262,419.8322401513307,13.0502227986052,-0.001339607412260437,0.002152985618908402,0.9868380617679345,0.1616917126850109
3,1733232244.246222,-11.38113488883607,419.832506908778,13.05035379930804,-0.001305077430043593,0.002090431447356799,0.9868645309147559,0.1615311873653645

Survey-Grade Prior Map

Ground truth trajectories are derived from a survey-grade prior map of the campus environment. The map is constructed via stitching of pointclouds that have been recorded on campus. As the campus consists of a northern and southern area, two separate pointclouds have been created. The map is available for download from the Download page.


Northern campus pointcloud
Southern campus pointcloud