NeurIT: Pushing the Limit of Neural Inertial Tracking for Indoor Robotic IoTDataset (tracking record of wheeled robot)

NeurIT Dataset is open-sourced for public research usage. It is collected using the customized robotic platform across three buildings. We collect the training, validation, and test-seen sets in Building A, and build the test-seen and test-unseen set in Building B and C. During data collection, the robot moves at varying speeds up to the maximum value (1.5m/s). The dataset contains 110 sequences, totaling around 15 hours of tracking data that corresponds to a travel distance of about 33.7 km. Each sequence of data lasts 6~10 minutes, containing both IMU data (acceleration, gyroscope, magnetometer) and the ground truth trajectory. The ratio of the training set, validation set, test-seen set, and test-unseen set is 15:3:3:4.