Reproduce “Robust Incremental Smoothing and Mapping (riSAM)” 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

Daniel McGann, John G. Rogers III, Michael Kaess

A new algorithm, Robust Incremental Smoothing and Mapping (riSAM), provides an online, incremental approach to Simultaneous Localization and Mapping (SLAM) that effectively handles outlier measurements. It demonstrated accuracy comparable to state-of-the-art batch robust SLAM while operating in real-time and achieving a 24% runtime reduction on one dataset with specific parameter adjustments.

Key insights

  • Full convergence at each step of Graduated Non-Convexity (GNC) can be approximated with a single update per control parameter step for online SLAM, maintaining robustness while drastically improving efficiency.
  • A novel Scale Invariant Graduated (SIG) kernel, featuring a finite μ range and scale-independent convexity, ensures predictable GNC iteration counts and consistent behavior across varying measurement scales.
  • A specialized Dog-Leg Line Search, modified for graduated non-convexity, addresses numerical instabilities and facilitates robust convergence in the presence of outliers, outperforming standard step-acceptance criteria.

Summary by alphaXiv.

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