The HelmDyn (Helmet Dynamic) dataset composes 10 sequences and is recorded for testing RESPLE on wearable platforms. A Livox Mid360 is mounted on a helmet and operated in a 12m × 12m × 8 m cubic space along with dynamic movements combining walking, running, jumping, and in-hand waving. Ground truth trajectories are acquired using a high-precision (submillimeter), low-latency motion capture system consisting of 12 Oqus 700+ and 8 Arqus A12 Qualisys cameras with passive markers. Check out the Research Arena Visionen at Linköping University in Sweden.

The R‐Campus (RUG Campus) dataset has 1 sequence and was recorded on a wheel-legged platform. A Livox Avia is mounted on a DIABLO robot which was operated within the Zernike campus of the University of Groningen, the Netherlands. The robot started and ended at the same point, enabling us to compute the end-to-end error.

The TudoRun (Tudor Run) dataset consists of 8 sequences. A Livox Mid360 is mounted on a Unitree Go2 quadruped robot operated indoor with dynamic motions. The first 3 sequences (TudoRun01-03) are recorded within a test field of about 10m × 4m covered by a motion capture system combining 8 Qualisys Miqus M3 cameras. The remaining 5 sequences (TudoRun04-08) started and ended in this test field but included more movements in a larger adjacent hall (no motion capturing). Ground truth was recorded only within this test field using the motion capture system with passive markers.

The TudoRun dataset serves as a supplementary dataset for RESPLE and was not included for evaluation in the original paper. Therefore, we provide the results above for comparision with a few representative LO or LIO systems.