VBGS-SLAM: Variational Bayesian Gaussian Splatting Simultaneous Localization and Mapping
Yuhan Zhu, Yanyu Zhang, Jie Xu, Wei Ren
AI summary
Problem
Current 3DGS-based SLAM systems rely on deterministic gradient optimization, making them sensitive to initialization, prone to catastrophic forgetting, and computationally heavy for long sequences. They also lack a principled way to quantify uncertainty in poses and map parameters.
Approach
The method formulates SLAM as a generative probabilistic model where camera poses and Gaussian map parameters are treated as latent variables. It uses variational Bayesian inference with conjugate priors to derive closed-form updates that jointly optimize the map and pose while propagating uncertainty.
Key results
- State-of-the-art tracking accuracy and robustness on long sequences
- Efficient closed-form updates outperform gradient-based optimization pipelines
- High-quality novel view synthesis across synthetic and real-world datasets
- Explicit uncertainty modeling mitigates drift and catastrophic forgetting
Why it matters
Enables more reliable and efficient real-time 3D reconstruction and localization for robotics and autonomous systems by replacing fragile deterministic optimization with principled uncertainty quantification.
Abstract
3D Gaussian Splatting (3DGS) has shown promis- ing results for 3D scene modeling using mixtures of Gaus- sians, yet its existing simultaneous localization and mapping (SLAM) variants typically rely on direct, deterministic pose optimization against the splat map, making them sensitive to initialization and susceptible to catastrophic forgetting as map evolves. We propose Variational Bayesian Gaussian Splatting SLAM (VBGS-SLAM), a novel framework that couples the splat map refinement and camera pose tracking in a genera- tive probabilistic form. By leveraging conjugate properties of multivariate Gaussians and variational inference, our method admits efficient closed-form updates and explicitly maintains posterior uncertainty over both poses and scene parameters. This uncertainty-aware method mitigates drift and enhances robustness in challenging conditions, while preserving the effi- ciency and rendering quality of existing 3DGS. Our experiments demonstrate superior tracking performance and robustness in long sequence prediction, alongside efficient, high-quality novel view synthesis across diverse synthetic and real-world scenes.