Drift-Free Visual Compass Leveraging Digital Twins for Cluttered Environments

2nd Space Robotics Workshop @ IEEE SCC 2025
Jungil Ham1*, Ryan Soussan2, Brian Coltin2, Hoyeong Chun1, Pyojin Kim1†
1 Gwangju Institute of Science and Technology (GIST), Korea 2 NASA Ames Research Center (KBR Inc.), USA
Paper Video Code

Abstract

We propose a Digital Twin-Based Outlier Rejection (DTOR) method for drift-free 3-DoF rotational motion estimation of free-flying robots in cluttered environments such as the International Space Station (ISS). By matching detected line features in RGB images with rendered images from a clutter-free 3D CAD model of the ISS, we effectively filter out clutter-induced outliers and extract reliable structural lines aligned with the Manhattan World (MW). Combined with a dominant plane extracted from the depth image, our method estimates the camera’s absolute rotation using only a single line and plane, the theoretical minimal configuration. Extensive evaluations on the Astrobee dataset demonstrate that our method significantly reduces rotational drift and outperforms state-of-the-art SLAM and visual compass systems, especially under extreme clutter and occlusion.

Video

BibTeX

@inproceedings{ham2025drift,
  author    = {Jungil Ham and Ryan Soussan and Brian Coltin and Hoyeong Chun and Pyojin Kim},
  title     = {Drift-Free Visual Compass Leveraging Digital Twins for Cluttered Environments},
  booktitle = {Proceedings of the 2nd Space Robotics Workshop (SRW) at IEEE SCC},
  year      = {2025},
  address   = {Pasadena, CA, USA}
}