MyGO-Splat: Multi-Objective Closed-Loop Geometric Feedback for RGB-Only Gaussian SLAM
2026-06-29 • Robotics
Robotics
AI summaryⓘ
The authors address a common problem in real-time 3D mapping from a single camera, where it is hard to know the true size of objects and fix errors in tracking. They introduce MyGO-Splat, a system that uses special 3D shapes called Gaussians to create a map that helps correct the camera’s position continuously. Their method also uses depth estimates from a foundation model to keep the size of the map accurate. Tests show their approach improves stability and looks as good as some methods that rely on extra depth sensors, despite using only one camera.
Monocular SLAMScale ambiguity3D Gaussian SplattingCamera pose optimizationDepth estimationSurface normalsClosed-loop systemRGB-D methodsTracking driftFoundation models
Authors
Fan Zhu, Ziyu Chen, Zhenjun Zhao, Zhisong Xu, Hui Zhu, Mingrui Li, Chunmao Jiang, Javier Civera
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.