IMU-Based Tip-Over Dataset for Space Exploration Rover Dynamics and Stability Analysis

This dataset provides a comprehensive collection of Inertial Measurement Unit (IMU) sensor data captured from a space exploration rover under varying conditions of terrain, speed, and inclination. The primary goal of this dataset is to enable the study of dynamic stability, specifically the detection and analysis of tip-over events, which are critical for safe and efficient operation of autonomous rovers in extraterrestrial environments. The dataset was captured on rocky and sandy terrain along with a dataset captured in Tenerife, Spain, in a volcanic region on sandy and rocky terrain, which replicates extraterrestrial surfaces like Mars. This dataset is provided by the Robotics Innovation Center, DFKI GmbH. Column Description time.microseconds Timestamp of each IMU measurement in microseconds, used to preserve temporal ordering and synchronize sensor streams. acc.data[0], acc.data[1], acc.data[2] Three-axis linear acceleration measured by the IMU (x, y, z). These capture translational dynamics and terrain-induced impacts. gyro.data[0], gyro.data[1], gyro.data[2] Three-axis angular velocity measurements (roll, pitch, and yaw rates), providing information about the rover's rotational motion and stability evolution. mag.data[0], mag.data[1], mag.data[2] Three-axis magnetometer measurements are used for orientation estimation. These signals were recorded but were not used as network inputs in the proposed framework. roll, pitch, yaw Orientation angles estimated from the IMU, primarily used for visualization and generation of ground-truth labels rather than as learning inputs. tip_over Binary ground-truth label indicating the actual tip-over event (1 = tip-over, 0 = normal operation). Used for evaluation and supervised classification. precursor_50hz Early warning label generated by shifting the tip-over label 25 samples earlier, corresponding to 0.5 s before the actual tip-over at a sampling frequency of 50 Hz. Used as the target for the 0.5 s prediction head during training. precursor_100hz Early warning label generated by shifting the tip-over label 50 samples earlier, corresponding to 0.5 s before the actual tip-over at a sampling frequency of 100 Hz. Used when training models on data sampled at 100 Hz.