Search papers, labs, and topics across Lattice.
This paper introduces MyGO-Splat, a closed-loop Gaussian SLAM framework that addresses the scale ambiguity and geometric self-correction limitations of existing RGB-only SLAM systems. By analytically rasterizing Gaussian primitives into pixel-wise depth and surface normals, the framework allows for active supervision of camera pose optimization, effectively bridging monocular priors with scale consistency. Extensive evaluations demonstrate that MyGO-Splat achieves performance on par with RGB-D methods while relying solely on monocular input, significantly enhancing scale stability and appearance-geometry consistency.
Achieving RGB-D performance with only monocular input, MyGO-Splat revolutionizes SLAM by integrating closed-loop geometric feedback for enhanced scale stability.
Real-time monocular Simultaneous Localization and Mapping (SLAM) fundamentally suffers from scale ambiguity and a lack of geometric self-correction. While 3D Gaussian Splatting (3DGS) enables high-fidelity rendering, existing RGB-only systems remain open-loop because depth priors are injected into mapping but refined geometry cannot effectively regulate tracking drift. We present MyGO-Splat, a closed-loop Gaussian SLAM framework that analytically rasterizes Gaussian primitives into pixel-wise depth and surface normals, allowing the map to actively supervise camera pose optimization. To bridge monocular priors and scale consistency, our framework introduces scale-aware adaptive alignment that projects foundation-model depth estimates into the globally optimized Gaussian space, forming a self-correcting cycle for scale feedback. Extensive evaluations show that this closed-loop design improves scale stability and appearance-geometry consistency, achieving performance comparable to RGB-D methods while using only monocular input.
