Global SLAM Using 5G ToA Integration: Performance Analysis with Unknown Base Stations and Loop Closure Alternatives

TL;DR

This work addresses the lack of global localization in indoor SLAM by integrating 5G Time of Arrival (ToA) measurements into ORB-SLAM3. It introduces an SE3 transformation node $\mathbf{Tr}$ and base-station bias terms to fuse ToA with visual-inertial data, enabling globally referenced trajectories and monocular scale recovery. The ToA factor links camera pose, local/global transforms, base-station positions, and scale, with uncertainty propagated through Hessian-based information matrices, and it supports operation with unknown base stations and ToA-driven loop-closure replacement. Evaluations on Aerolab and EuRoC datasets using simulated 5G ToA at 28 GHz and 78 GHz show that 78 GHz ToA provides superior global accuracy and that ToA can maintain trajectory consistency even when loop closure is unreliable or unavailable. The results offer practical guidance on base-station deployment and demonstrate the potential of 5G-ToA-enabled global SLAM for robust indoor drone navigation, while outlining future work in real-world 5G hardware validation and multi-agent extensions.

Abstract

This paper presents a novel approach that integrates 5G Time of Arrival (ToA) measurements into ORB-SLAM3 to enable global localization and enhance mapping capabilities for indoor drone navigation. We extend ORB-SLAM3's optimization pipeline to jointly process ToA data from 5G base stations alongside visual and inertial measurements while estimating system biases. This integration transforms the inherently local SLAM estimates into globally referenced trajectories and effectively resolves scale ambiguity in monocular configurations. Our method is evaluated using both Aerolab indoor datasets with RGB-D cameras and the EuRoC MAV benchmark, complemented by simulated 5G ToA measurements at 28 GHz and 78 GHz frequencies using MATLAB and QuaDRiGa. Extensive experiments across multiple SLAM configurations demonstrate that ToA integration enables consistent global positioning across all modes while maintaining local accuracy. For monocular configurations, ToA integration successfully resolves scale ambiguity and improves consistency. We further investigate scenarios with unknown base station positions and demonstrate that ToA measurements can effectively serve as an alternative to loop closure for drift correction. We also analyze how different geometric arrangements of base stations impact SLAM performance. Comparative analysis with state-of-the-art methods, including UWB-VO, confirms our approach's robustness even with lower measurement frequencies and sequential base station operation. The results validate that 5G ToA integration provides substantial benefits for global SLAM applications, particularly in challenging indoor environments where accurate positioning is critical.

Figures (14)

• Figure 1: Illustration of an indoor warehouse environment where a drone navigates while two 5G base stations provide ToA measurements. The base stations assist in refining the drone's positional accuracy, supporting real-time localization and mapping (The warehouse figure was generated using OpenAI's ChatGPT).
• Figure 2: Diagram illustrating the relationship between the local and global frames in SLAM. The transformation $\mathbf{Tr}$ aligns the local frame ($x', y'$) with the global coordinate frame ($x, y$). The single map, shown in both frames, is represented in red for the local frame and teal for the global frame. The initial pose of the drone is faded to indicate its starting position, while the current pose is shown in the local frame. A base station in the global frame provides ToA measurements, aiding in the alignment and refinement of global positioning. The dashed arrow represents the influence of ToA measurements, linking the local and global frames via $\mathbf{Tr}$.
• Figure 3: Structure of the ToA factor, illustrating the key components: camera pose node ($\mathrm{T}$), scale factor ($s$), local-to-global transformation node ($\mathrm{Tr}$), base station position node ($\mathrm{L}$), and bias node ($\tau$). Double-bordered nodes indicate fixed parameters during optimization.
• Figure 4: Diagram of ORB-SLAM3 with ToA integration, illustrating the pipeline across Tracking, Local Mapping, and Loop Closing threads. New components are shown in green, including ToA-based global map refinement, while modified components are in yellow, such as Local Bundle Adjustment.
• Figure 5: Structures of optimization graphs for various components in the SLAM process: (a) Local Inertial BA: Optimization that integrates visual, inertial, and ToA measurements to refine keyframe poses and map points while addressing significant IMU noise. (b) Local Bundle Adjustment: Local refinement of keyframe poses, map points, ToA biases, and local-to-global transformations using visual and ToA constraints within a local optimization window. (c) Tracking Pose Optimization: Real-time optimization of the camera pose for the incoming frame by minimizing reprojection errors and incorporating ToA distance constraints. (d) Global Map Refinement: Global map refinement by leveraging odometry, co-visibility, loop closure, and ToA edges to improve global map consistency and keyframe accuracy. (f) Transformation Refinement: Periodic optimization focusing solely on refining the local-to-global transformation with fixed keyframe poses, enhancing robustness against IMU noise and drift. (e) Scale Refinement: A dedicated process for monocular SLAM systems to resolve scale ambiguity by optimizing the global scale factor, ensuring consistent keyframe poses and map points. Note: Double-bordered nodes in the graphs indicate fixed nodes during optimization.