MyGO-Splat: Multi-Objective Closed-Loop Geometric Feedback for RGB-Only Gaussian SLAM

2026-06-28

AI Summary

Monocular SLAM suffers from scale ambiguity and limited geometric self-correction capability, and existing RGB-based 3D Gaussian splatting methods are typically open-loop systems that cannot effectively correct pose drift. This work proposes the first closed-loop Gaussian SLAM framework that relies solely on monocular RGB input. It leverages analytical rasterization to generate per-pixel depth and normals, enabling geometric supervision of camera poses by the map. Furthermore, a scale-aware adaptive alignment mechanism is introduced to project depth priors from foundation models into the globally optimized Gaussian space, establishing a scale self-correction loop. The proposed method significantly enhances scale stability and geometric-appearance consistency, achieving performance comparable to RGB-D approaches.

Abstract

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.

Problem

• monocular SLAM
• scale ambiguity
• geometric self-correction
• closed-loop feedback
• 3D Gaussian Splatting

Innovation

• Gaussian Splatting
• Closed-loop SLAM
• Monocular Depth
• Scale Consistency
• Geometric Feedback