The Intelligent Systems, Automation and Robotics Laboratory

Marzaglia Dataset

Marzaglia data set was collected with a Formula 3 car at the Marzaglia racing track (in Modena). The car was equipped with a RealSense D455, composed of an IMU, recording at 250 Hz, and a stereo camera, which provides images at 30 FPS with a resolution of 640 × 480 pixels and a field of view of 90°. Images are cropped to 640 × 430 to remove the vehicle from the picture. Pose ground truth is obtained by collecting GPS measurements with an Emlid Reach M2, while RTK corrections are provided by an Emlid Reach RS2. With this sensor setup we recorded three sequences, each one associated with one of the three laps around the circuit we performed by driving the vehicle. The scenarios proved to be particularly challenging due to the lack of features and the high IMU noise caused by the strong vibrations of the vehicle. Moreover, it is important to mention that the ground truth was collected by only using GPS measurements, thus excluding the rotation.

We provide the full sequence

Available data includes:
– Rosbag containing IMU and images of both the right and left cameras;
– Position ground-truth in TUM format;
– Calibration files.

The dataset is available at the following link:

For a benchmark of VO-VIO approaches on the Marzaglia Dataset, refer to the following publication:

If you use Marzaglia dataset in an academic work, please cite:

@article{legittimo23benchmark, AUTHOR = {Legittimo, Marco and Felicioni, Simone and Bagni, Fabio and Tagliavini, Andrea and Dionigi, Alberto and Gatti, Francesco and Verucchi, Micaela and Costante, Gabriele and Bertogna, Marko}, TITLE = {A benchmark analysis of data‐driven and geometric approaches for robot ego‐motion estimation}, YEAR = {2023}, JOURNAL = {Journal of Field Robotics} }