Reproduce “MOVIN: Real-time Motion Capture using a Single LiDAR” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Deok-Kyeong Jang, Dongseok Yang, Deok-Yun Jang, Byeoli Choi, Taeil Jin, Sung-Hee Lee

Researchers from KAIST and MOVIN Inc. developed MOVIN, a system that achieves real-time full-body motion capture with accurate global translation using only a single LiDAR sensor. It demonstrated an average joint position error of 6.21 cm, outperforming vision-based baselines while providing robust root position and velocity tracking.

Key insights

  • A single LiDAR sensor can provide sufficient 3D data for accurate full-body motion capture, including robust global translation, addressing limitations of camera-based systems.
  • Incorporating historical point cloud data within an autoregressive CVAE framework significantly improves temporal coherence and robustness against self-occlusions during motion capture.
  • The Mixture-of-Expert decoder empirically enhances pose construction and reduces visual artifacts, contributing to more plausible and natural generated movements.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.