Reproduce “DLO-Splatting: Tracking Deformable Linear Objects Using 3D Gaussian Splatting” 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

Holly Dinkel, Marcel Büsching, Alberta Longhini, Brian Coltin, Trey Smith, Danica Kragic, Mårten Björkman, Timothy Bretl

Researchers at the University of Illinois Urbana-Champaign, KTH Royal Institute of Technology, and NASA Ames Research Center developed DLO-Splatting, a hybrid algorithm for tracking the 3D shape of Deformable Linear Objects (DLOs) from multi-view RGB images. The method combines physics-based prediction using Position-Based Dynamics with a 3D Gaussian Splatting-based visual update, demonstrating improved tracking of grasped rope tips during complex knotting maneuvers.

Key insights

  • Combining first-principles physics (PBD) for prediction with rendering-based visual updates (3D Gaussian Splatting) offers a robust approach for DLO tracking that doesn't require learning object dynamics from data.
  • Decoupling the geometric model (node chain) from the appearance model (Gaussians) in the 3D Gaussian Splatting approach allows for faster visual optimization crucial for dynamic DLO state estimation.
  • Integrating gripper state information helps maintain accurate tracking of actuated DLO parts, which is beneficial in manipulation tasks involving complex hand-object interactions.

Summary by alphaXiv.

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