MotionGS-SLAM: Event-Modulated Gaussian Splatting for Motion-Blur Robust SLAM
ZHIQIANG HU, Shouren Huang, Masatoshi Ishikawa
AI summary
Problem
Current vision-based SLAM systems fail under severe motion blur because they treat blur as an artifact to be removed rather than modeling its physical formation, leading to degraded tracking and corrupted 3D maps.
Approach
The method replaces deblurring with a generative forward model that uses high-temporal-resolution event camera data to dynamically adapt 3D Gaussian splatting kernels, aligning their shape and temporal sampling with local motion cues to physically simulate blur during rendering.
Key results
- Proposes a physics-based SLAM framework that models blur formation instead of removing it
- Introduces an event-modulated Gaussian kernel with dual spatial and temporal modulation via a 4D spatio-temporal hash grid
- Achieves significant improvements in trajectory accuracy and map quality over state-of-the-art methods on synthetic and real-world sequences
- Enables real-time tracking and joint optimization of camera poses and 3D Gaussian maps without depth sensors
Why it matters
It enables reliable real-time navigation and dense scene reconstruction for robots and AR systems operating in high-motion or low-light conditions where traditional visual SLAM fails.
Abstract
Current Vision-based SLAM systems fail catas- trophically when motion blur corrupts the visual input, as they attempt the ill-posed inverse problem of recovering sharp content from degraded observations. We present MotionGS- SLAM, which fundamentally reimagines motion blur handling through a paradigm shift: rather than removing blur artifacts, we reformulate the challenge as a well-constrained forward problem that generatively models blur formation within the rendering pipeline. By leveraging event cameras’ microsecond temporal resolution and immunity to motion blur, we introduce a novel event-modulated Gaussian kernel that dynamically adapts each Gaussian’s rasterization based on precise motion cues. Our dual-modulation mechanism transforms 2D Gaussian projections from isotropic dots into anisotropic, motion-aligned elliptical brush strokes (spatial modulation) while adaptively varying exposure integral sampling density based on local velocity (temporal modulation). This physics-based approach enables joint optimization of intra-exposure camera trajectories and 3D scene geometry through blur-aware photometric and event-based constraints. Extensive experiments demonstrate significant improvements over state-of-the-art methods in tra- jectory accuracy and map quality under severe high-motion conditions.