Reproduce “CT-ICP: Real-time Elastic LiDAR Odometry with Loop Closure” 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
Pierre Dellenbach, Jean-Emmanuel Deschaud, Bastien Jacquet, François Goulette
CT-ICP introduces a real-time, LiDAR-only odometry method that uses an elastic intra-scan distortion model and inter-scan discontinuity for robust pose estimation, especially during high-frequency motions. Integrated with a novel pure LiDAR loop closure, the system achieves global consistency, notably reducing Absolute Trajectory Error from 29.87m to 1.07m on KITTI-360 and outperforming baseline methods on uncorrected and dynamic datasets.
Key insights
- Modeling intra-scan motion elastically with continuous pose interpolation between two scan-specific poses significantly improves precision on raw LiDAR data by compensating for motion during scan acquisition.
- Allowing discontinuity between adjacent scan poses, while constrained, enhances robustness to abrupt high-frequency sensor movements that rigid continuity models struggle with.
- Leveraging aggregated local point clouds stored in a sparse voxel grid for calculating normals and planarity weights provides more stable and reliable information for the ICP objective function.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.