NoStrandAMust Dataset

1. Introduction

This dataset was produced during the NoStrandAMust project, by robotics researchers from the DFKI-RIC (German Research Center for Artificial Intelligence - Robotics Innovation Center) between February 2022 and May 2025.

The project aims to study the interactions between robot wheels or legs and the ground to improve real-time simulation capabilities and overall robot autonomy in varying environments for future space missions. 
One part of the project is to collect different sensor data while different robots are moving in a controlled environment over different ground conditions. This dataset consists of cleaned and structured data records from these experiments.

As a test track for the robot experiments, a construction was used that consisted of wooden basins filled with different terrain materials, namely sand, gravel, and an imitation of cooled and solidified lava. The file test_track.jpg shows a bird's eye view of the test track. The four robots whose data are included in the dataset are shown in the file robots.png.

2. Dataset

2.1 Experiment variants

Each of the four robots was subjected to two types of maneuvers on different terrains. The two types of maneuvers include:

1. Straight line: The robot moves in a straight line.
2. lockwise rotation: The robot rotates clockwise in a small diameter circle.

Each of these maneuvers was performed repeatedly on four different terrain basins as follows:

1. Gravel: The basin is completely filled with ~10 mm diameter gravel.
2. Sand: The basin is completely filled with fine sand, 0-1 mm diameter.
3. Lava: Artificial rock consisting of fiberglass and a layer of polyester concrete.
4. Concrete: Solid concrete floor.

For the legged robot Crex, the straight line maneuvers were repeated with two different gaits: 

1. Single leg gait: Only one leg moves at a time, while the other legs remain on the ground for support.
2. Cross gait: The diagonal legs move together - one front leg and the opposite back leg move at the same time.

Some robots did not complete all variants. In most cases, this was due to the operator's decision not to perform the maneuver on the respective terrain type (e.g., lava terrain is difficult and potentially dangerous for some robots). In other cases, the reason was a sensor error (Coyote in the variant with clockwise rotation). The inclusion of concrete floor was considered optional and was only included in a few cases. The following table provides an overview of the variants included for each robot.

2.2 Recorded Data

All systems data was captured from the devices summarized in Table 1.

Table 1: Data included in the dataset

2.3 File Structure

The data is organized in a directory tree structure, with the robot name as the first level folder name, the maneuver as the second level folder name, and the experiment variant as the third level folder name.
The experiment variant is encoded according to the following pattern:

${gait_name}_${terrian_name}_${run_id}

Only the lowest level directory contains the data files for the corresponding run, with the contained data encoded in the filename according to Table 1.

Here is an overview of the file tree:

.
└── ${robot_name}/
    ├── straight_line/
    │   └── ${variant}/
    │       ├── imu_orientation.csv
    │       ├── imu_sensors.csv
    │       .
    │       .
    └── turn_clockwise/
        └── ${variant}/
            ├── imu_orientation.csv
            ├── imu_sensors.csv
            .
            .

3. Acknowledgements

The project on which this publication is based was funded by the German Federal Ministry for Economic Affairs and Energy under the funding code 50RA2122 The authors are responsible for the content of this publication.