XGRIDS Lixel L3 SLAM test: 270 m tunnel scan under the Tyne

We took the new XGRIDS Lixel L3 handheld LiDAR scanner through the Tyne Pedestrian Tunnel: 270 m long, dipping 12 m below the River Tyne, with no GNSS once underground and a long, repetitive stretch with very few features for the software to lock onto.

These are the conditions where long handheld scans usually start to struggle. This blog explains how we scanned the tunnel, why it is a tough test for SLAM, how the L3 handled it, and what the result means for tunnel and infrastructure surveys.

Tip: if the video doesn't play in high resolution automatically, click the settings cog and select the highest quality available.

Key takeaways

• A tough test for SLAM: A 270 m GNSS-denied tunnel with repetitive tiling and few distinct features is a classic environment for drift and layering.

• Fast capture: The 32-channel L3 scanned the tunnel chest-mounted, one pass each way, in 45 minutes of scanning.

• Anchored at both ends: RTK fixes at each end of the tunnel anchored the scan in real-world coordinates, with SLAM bridging the gap in between.

• A clean result: LiDAR SLAM, multi-sensor fusion and LixelStudio's high-precision optimisation produced a true-colour point cloud with no visible layering.

• A capability demo, not an accuracy test: This project shows how the L3 handles a difficult environment; it was not a controlled accuracy survey.

The site: the Tyne Pedestrian Tunnel

The Tyne Pedestrian Tunnel runs 270 m beneath the River Tyne, close to heliguy™'s North Shields base. At its deepest point, it lies 12 m (40 ft) below the river bed.

The tunnel descends from one bank, levels out under the river and climbs back up to the other side. Its walls are lined with near-identical tiling, broken only by a coloured band, a run of ceiling services and the occasional sign.

RTK was available at the surface at each end. Underground, the L3 relied on SLAM and its onboard sensors.

Why tunnels are hard for SLAM

Handheld scanners like the L3 use SLAM (simultaneous localisation and mapping) to work out where they are as they move. They do this by matching what the LiDAR sees from one moment to the next. A long tunnel makes that difficult in three ways.

• No GNSS: Once underground, there is no satellite positioning to correct the scanner's estimate of where it is.

• Repetitive geometry: Every section of tunnel looks much like the last, so the software struggles to tell how far along it has travelled.

• Few features: Smooth, tiled walls give the LiDAR very little distinctive detail to lock onto.

Small errors in the scanner's position can then build up over the length of the scan. This is known as drift.

Drift often shows up as layering: when the scanner passes the same spot twice, it records the same wall in slightly different places. Instead of one clean surface, you get two or more offset copies, which makes the data unreliable for measurement.

When the passes line up, each surface appears once. When they drift, it appears twice, slightly offset.

How we scanned the tunnel with the Lixel L3

We used the 32-channel version of the Lixel L3, chest-mounted, and walked the tunnel in one pass each way. Scanning took 45 minutes.

The RTK fixes at each end matter. They tie both ends of the scan to known real-world positions, leaving SLAM to bridge the GNSS-denied stretch in between.

How the Lixel L3 handled SLAM drift

Three things worked together to keep the scan on track.

• LiDAR SLAM and multi-sensor fusion: The L3 combines its LiDAR with a high-accuracy IMU and real-time loop-closure optimisation, which kept the two passes aligned through the tunnel, with no visible layering.

• RTK anchoring: The fixes at each end anchored the scan in real-world coordinates, giving the processing known positions to correct against.

• LixelStudio optimisation: XGRIDS' processing software applied high-precision optimisation to the dataset after capture.

XGRIDS designed the L3 for this kind of site. It is optimised for tunnels, repetitive structures and other low-feature environments, as well as areas with GNSS obstruction.

The results: a clean, true-colour point cloud

As the flythrough above shows, the L3 produced a clean, true-colour point cloud of the whole route, from street level to street level, with no visible layering.

Looking down the tunnel, the tiled walls, coloured band, ceiling services and floor joints stay crisp along the length of the scan. The repetition that makes the tunnel hard for SLAM is clear to see.

Up close, individual tile courses are still visible and the coloured band is a single, sharp line. If the two passes had drifted apart, these edges would appear doubled or blurred.

Because the scan was anchored with RTK at each end, the point cloud is also georeferenced: it sits in real-world coordinates, ready to combine with other site data.

From street level to the tunnel, in one continuous scan

The scan didn't start in the tunnel. It began at street level, where the L3 had an RTK fix, and continued through the entrance building, down the escalators and into the tunnel itself, without a break.

The entrance building at Jarrow, captured from the street, with brickwork, signage and the shutters clearly defined.

Inside the entrance hall: tiled walls, windows and floor finishes, with the street visible through the doorway.

The escalators down to the tunnel, with the handrails, step housings and fire exit signage all captured.

At the bottom, the junction between the cyclists' and pedestrians' tunnels, where the signage is legible in the point cloud.

This is where the RTK fixes came from: open sky at each end gave the scan known positions, before SLAM took over underground.

This was a capability demonstration rather than a controlled accuracy test, so we have not published accuracy figures for this scan. The results shown are what you can see in the data. For controlled accuracy testing, see our XGRIDS Lixel L3 accuracy validation.

What this means for tunnel scanning and infrastructure surveys

Tunnels, culverts, underpasses and long corridors share the same problems: no GNSS, repetitive geometry and few features. A handheld LiDAR scanner that copes with these conditions makes these sites quicker and simpler to capture.

• Faster capture: A 270 m tunnel took 45 minutes of scanning, walked on foot, with no static scanner setups.

• Fewer return visits: The L3's built-in display shows the point cloud, trajectory and coverage during capture, so gaps can be spotted before leaving site.

• Ready for existing workflows: Point clouds can feed measurement, sections and Scan-to-BIM, while georeferencing ties the data to other site records.

• More than a point cloud: The same L3 dataset can also produce a textured mesh and a photorealistic 3D Gaussian Splatting scene.

About the XGRIDS Lixel L3

The Lixel L3 is XGRIDS' flagship handheld spatial scanner, built for high-accuracy surveying, engineering and reality capture.

heliguy™'s in-house survey division can support you with XGRIDS Lixel L3, workflow integration, and data processing.

Want to see what the L3 can do in your space? Find out more about the XGRIDS Lixel L3 or get in touch to arrange a demo.