Reproduce “GARAD-SLAM: 3D GAussian splatting for Real-time Anti Dynamic SLAM” 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
Mingrui Li, Weijian Chen, Na Cheng, Jingyuan Xu, Dong Li, Hongyu Wang
GARAD-SLAM presents a real-time 3D Gaussian Splatting-based SLAM system that accurately operates in dynamic environments by performing dynamic object removal directly on 3D Gaussians. The system achieves an average ATE of 0.0194m on the TUM RGB-D dataset and generates artifact-free 3D reconstructions at 56.4 FPS.
Key insights
- A tightly coupled approach, where dynamic information from the 3D map informs front-end tracking and robust tracking aids back-end mapping, is effective for dynamic SLAM.
- Directly performing dynamic segmentation on 3D Gaussians, combined with robust verification, can effectively prevent mapping errors and artifacts in dynamic environments.
- A gradual penalty optimization for dynamic Gaussians, rather than immediate removal, helps maintain the integrity and completeness of the static scene reconstruction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.