Reproduce “On-Manifold Preintegration for Real-Time Visual-Inertial Odometry” 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
Christian Forster, Luca Carlone, Frank Dellaert, Davide Scaramuzza
The authors present a monocular visual-inertial odometry system that compresses high-rate IMU readings between camera keyframes into relative-motion constraints for incremental factor-graph smoothing, using rotations on SO(3). On a 430 m indoor route, its average relative drift over 360 m is 0.3 m, versus 0.7 m for OKVIS and MSCKF; this system-level comparison does not isolate preintegration’s effect.
Key insights
- Rigorous on-manifold IMU preintegration, unlike Euler-angle methods, provides geometrically consistent noise propagation, avoids singularities, and maintains invariance under rigid body transformations.
- Combining on-manifold preintegration with incremental smoothing algorithms (iSAM2) and a structureless vision model enables real-time, highly accurate visual-inertial odometry by achieving full-smoothing accuracy with filtering-like speed.
- First-order a-posteriori bias correction for preintegrated measurements maintains computational efficiency without requiring full re-integration when bias estimates change during optimization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.