MotionGS-SLAM: Event-Modulated Gaussian Splatting for Motion-Blur Robust SLAM

The newly developed SLAM system, MotionGS-SLAM, aims to tackle the limitations of vision-based SLAM systems when faced with motion blur. Instead of trying to retrieve clear images from compromised data, this system reinterprets motion blur as a forward problem, simulating the blur creation within the rendering process. It utilizes the microsecond temporal resolution and motion blur resistance of event cameras, introducing a Gaussian kernel that adjusts rasterization dynamically according to motion signals. This dual-modulation approach converts 2D Gaussian projections into elliptical shapes that are anisotropic and aligned with motion, while also adaptively altering the density of exposure integral sampling. The research can be found on arXiv under the identifier 2608.15024.