Top: Sensor setup (left) and checkpoints overlaid on the SLAM map of the Stadtgarten scene (right). Orange: open-sky; Cyan: GNSS-obstructed. Bottom: Absolute 3D error per checkpoint for Stadtgarten Seq. 1 using FAST-LIO-SAM. Standalone RTK errors grow to tens of meters in GNSS-degraded zones, while offline RTK-SLAM remains mostly below 10 cm.
 

Abstract

RTK-SLAM systems integrate simultaneous localization and mapping (SLAM) with real-time kinematic (RTK) GNSS positioning, promising both relative consistency and globally referenced coordinates for efficient georeferenced surveying. A critical and underappreciated issue is that the standard evaluation metric, Absolute Trajectory Error (ATE), first fits an optimal rigid-body transformation between the estimated trajectory and reference before computing errors. This so-called SE(3) alignment absorbs global drift and systematic errors, making trajectories appear more accurate than they are in practice, and is unsuitable for evaluating the global accuracy of RTK-SLAM. We present a geodetically referenced dataset and evaluation methodology that expose this gap. A key design principle is that the RTK receiver is used solely as a system input, while ground truth is established independently via a geodetic total station — a separation absent from all existing benchmarks. The dataset covers two outdoor-to-indoor scenes with synchronized LiDAR, camera, IMU, and RTK inputs. We evaluate LiDAR-inertial, visual-inertial, and LiDAR-visual-inertial RTK-SLAM systems alongside standalone RTK, reporting direct global accuracy and SE(3)-aligned relative accuracy to make the gap explicit. Results show that SE(3) alignment can underestimate absolute positioning error by up to 76%. RTK-SLAM achieves centimeter-level absolute accuracy in open-sky conditions and maintains decimeter-level global accuracy indoors, where standalone RTK degrades to tens of meters.

Dataset

Data Walkthrough

The videos below show recorded sensor data replayed at 4× speed with online FAST-LIO-SAM processing. Top row: 3D local map view with accumulated point cloud and trajectory (left); IMU readings (right). Bottom row: Global map view (left) — ★ stars: surveyed checkpoints, ● red points: GNSS measurements, ● blue trajectory: RTK-SLAM online estimate; GNSS status with RTK fix quality indicators (center); camera image (right).

• Stadtgarten Seq. 1
• Stadtgarten Seq. 2
• Construction Hall Seq. 1
• Construction Hall Seq. 2

Sensor Platform

The handheld RTK-SLAM device integrates:

• Livox MID360 LiDAR with integrated IMU — 360° horizontal, 59° vertical FOV, 10 Hz, non-repetitive scan pattern
• 2 MP global shutter camera — hardware-triggered at ~10 Hz
• UM980 GNSS receiver — RTK corrections provided by the German SAPOS service, centimeter-level accuracy under open-sky

The LiDAR and its built-in IMU are hardware-synchronized to GNSS time via a 1 PPS signal. All extrinsics are carefully calibrated; the GNSS antenna phase center and device base center are referenced via CAD model offsets, enabling direct comparison of estimated positions against surveyed control points.

Scenes

 
Stadtgarten

Public park in Stuttgart. Two sequences covering 1.04 km and 0.46 km. Three distinct GNSS zones: open sky, partial obstruction (buildings/trees), and a fully GNSS-denied 30 m underpass tunnel. 55 control points total.
 
OutdoorGNSS-degradedGNSS-denied

 
Construction Hall

IntCDC construction site, University of Stuttgart. Two sequences (clockwise & counter-clockwise) covering 0.48 km and 0.39 km. Each sequence begins and ends outdoors with RTK fix and traverses the interior where GNSS signals are severely degraded (>400 s, ~150 m). 32 control points total.
 
Outdoor-to-indoorGNSS-severely degraded

Geodetic Ground Truth

Ground truth is established via a two-stage procedure entirely independent of the RTK receiver used as system input. First, open-sky anchor points are surveyed by static GNSS observations (<5 mm std). A Leica TS16 total station is oriented to these anchors and measures all remaining control points, including those under GNSS obstruction and inside GNSS-denied areas, propagating the global reference frame via a traverse. The final ground truth accuracy is better than 1 cm for all control points.

Critical design principle: the RTK receiver is exclusively a system input. Ground truth is established by the total station independently. This separation — absent from all existing benchmarks — is what enables meaningful evaluation of absolute global accuracy.

Evaluation

Key Insight

Methods

We benchmark three RTK-SLAM configurations:

• FAST-LIO-SAM — LiDAR-inertial-GNSS system combining FAST-LIO2 front-end with factor graph backend. Both online and offline (batch pose graph optimization) results reported.
• OKVIS2-X(vig) — Keyframe-based visual-inertial-GNSS system with tight GNSS integration via 4-DoF frame alignment.
• OKVIS2-X(lvig) — Same system extended with LiDAR (LiDAR-visual-inertial-GNSS). Compared against the vig configuration to quantify LiDAR's contribution.
• Standalone RTK — Direct RTK positioning without any odometry integration.

Quantitative Results

Absolute ATE (no alignment) vs. SE(3)-aligned ATE. The Gap column shows how much the alignment hides. A gap of 76% means the standard metric underestimates the true absolute error by a factor of more than 4.

Trajectory Comparisons

Trajectories for all four sequences overlaid on satellite imagery, with GNSS-denied and GNSS-degraded zones annotated.

Drift Under GNSS Outage

Positioning error (log scale) as a function of time elapsed and distance traveled to the nearest RTK fix, aggregated over all checkpoint measurements. Because offline optimization propagates corrections both forward and backward, the x-axis is distance to the nearest RTK fix rather than only since the last fix. LiDAR-aided methods show low drift rates: 9.2 cm/min (0.25% of path) for FAST-LIO-SAM and 8.0 cm/min (0.22%) for OKVIS2-X(lvig). Standalone RTK degrades rapidly once signal quality deteriorates.

How to cite