ucu-space-robotics/traversability-costmap

Indomitus Rover Traversability Bags

ROS 2 bags (.mcap) recorded on the Indomitus rover of UCU Space Robotics while driving over different terrain: flat ground with bumps, single stones and a stony area. The rides were driven at several speeds.

The dataset is used to train a self-supervised traversability model. The camera images are the model input. The IMU and odometry provide the labels (slip and chassis vibration). Code for preprocessing training and evaluating model are in the GitHub repo.

Contents

Each bag contains:

Topic	Type	Rate
/rover/zed2i/rgb/image_rect_color/compressed	sensor_msgs/CompressedImage	7 Hz
/rover/zed2i/depth/depth_registered/compressedDepth	sensor_msgs/CompressedImage	7 Hz
/rover/zed2i/rgb/camera_info, /rover/zed2i/depth/camera_info	sensor_msgs/CameraInfo	7 Hz
/rover/zed2i/imu/data	sensor_msgs/Imu	100 Hz
/rover/zed2i/odom	nav_msgs/Odometry (ZED visual odometry)	20 Hz
/rover/wheels/odom	nav_msgs/Odometry (wheel odometry)	100 Hz
/rover/odom	nav_msgs/Odometry (EKF fused)	20 Hz
/rover/joint_states	sensor_msgs/JointState	100 Hz
/rover/cmd_vel	geometry_msgs/Twist	50 Hz, when driven
/tf, /tf_static	tf2_msgs/TFMessage

Recordings are grouped into sessions. One session is one recording day at one location, and it has a single camera configuration. Each session has a session.yaml.

Structure

├── README.md
└── raw/
    └── <session>/                # <YYYY-MM-DD>_<location>, e.g. 2026-10-19_ucu-cinema
        ├── session.yaml          # date, location, lighting, surface description, notes
        ├── zed2i_record.yaml     # ZED 2i camera config used for this session
        └── <bag>/                # one ROS 2 bag
            ├── metadata.yaml
            └── <bag>_0.mcap

Bag names follow <ride_type>_<speed>[_<index>]_<YYYYMMDD>_<HHMMSS>, for example mixed_05ms_2_20261019_160254 is the second mixed ride at a maximum of 0.5 m/s, started at 16:02:54 UTC. The ride types, the full naming convention and the train / val split of each bag are described in data/README.md and data/manifest.csv on GitHub.

Download

With the code repo, use scripts/download_data.py. It selects bags by split, session or ride type from the manifest.

Without it, use the hf CLI:

hf download ucu-space-robotics/traversability-costmap --repo-type dataset --local-dir data                                          # everything
hf download ucu-space-robotics/traversability-costmap --repo-type dataset --local-dir data --include "raw/2026-10-19_ucu-cinema/*"  # one session

Reading the bags

The bags can be opened with rosbags without a ROS installation:

from pathlib import Path
from rosbags.highlevel import AnyReader

with AnyReader([Path("data/raw/2026-10-19_ucu-cinema/flat_05ms_1_20261019_152249")]) as reader:
    conns = [c for c in reader.connections if c.topic == "/rover/zed2i/imu/data"]
    for conn, t, raw in reader.messages(connections=conns):
        msg = reader.deserialize(raw, conn.msgtype)
        print(t, msg.linear_acceleration)

With ROS 2 installed, use ros2 bag play <bag folder> or ros2 bag info <bag folder>.