Underground mining and tunnels
GNSS signal does not penetrate rock or concrete. Survey work below ground has always been GNSS-denied by definition, and routes are often long and geometrically repetitive.
GNSS-denied SLAM
SLAM mapping without GNSS
Most outdoor SLAM setups lean on GNSS to bound drift over a long route and to anchor the finished map to real-world coordinates. Underground, indoors, in a tunnel or under heavy canopy, that reference is gone and the system has to hold up on LiDAR and IMU alone.
What actually gets harder
Nothing in the reconstruction stops working when GNSS disappears. What changes is that every error stays in the map. With GNSS present, accumulated drift gets pulled back toward an absolute track that does not drift. Without it, the only corrections available come from the LiDAR data itself, so loop closure and consistent scan matching matter more, not less, than they do outdoors.
Scan · register · accumulate, no external reference
Where this comes up
Indoor robotics
AMRs, warehouse and facility mapping. Indoors GNSS is absent or unreliable no matter what the building is, so indoor stacks are built around LiDAR and IMU from the start.
Search and rescue, confined spaces
The same constraint with time pressure added: no external reference, unfamiliar geometry, and a single pass through the space rather than a careful survey.
Other GNSS-denied research
Dense urban canyon, subsea infrastructure, and defense and security robotics R&D are further examples of fields that work under this constraint. Listed here as categories of GNSS-denied work in general.
What no GNSS means for the output
You still get
A locally consistent map with correct shapes and correct metric scale, plus the estimated sensor trajectory. GLIM handles the reconstruction the same way regardless of whether GNSS was recorded.
You do not get
Georeferencing to WGS84. That step aligns the SLAM trajectory to a GNSS track using FlexCloud from TUM, so GNSS has to be present somewhere in the recording. No GNSS means no WGS84 alignment, full stop. It is a hard requirement, not a quality tradeoff.
What keeps a GNSS-denied map crisp
In general SLAM terms, a map stays sharp when the same physical surface is registered to the same place every time it is observed. Two things drive that. Scan matching has to stay consistent, meaning the registration behaves the same way in a corridor as it does in an open hall, and the IMU keeps the estimate sane through the moments where the geometry says little. And revisited space has to be recognised as revisited, so that the loop can be closed and the accumulated error redistributed along the trajectory instead of smearing the walls into a double image.
Where those two conditions fail, you see it directly in the cloud: duplicated walls, a corridor that bends where it should be straight, or a floor that slowly climbs. Those are the symptoms to look for when reviewing a GNSS-denied run.
Frequently asked questions
Can SLAM work without GPS or GNSS at all?
Yes. SLAM estimates motion by registering each LiDAR scan against the map built so far, with IMU data helping through fast motion. GNSS is an optional external reference, not a requirement. Without it you still get a locally consistent map with correct shapes and correct metric scale, just with an arbitrary origin at the first frame.
What is different about SLAM in tunnels or underground compared to outdoors?
Two things. There is no GNSS to bound drift over a long route, so loop closure and consistent scan matching carry the whole burden. And the geometry is often degenerate: a straight tunnel or a smooth drift looks the same from many positions along its axis, so scans constrain the cross section well and the along-axis motion poorly. An IMU and revisiting known space are what keep those runs usable.
Can I still get a usable map if my rosbag has no GNSS topic?
Yes. Upload a recording with a LiDAR PointCloud2 topic, and an IMU topic if you have one. The reconstruction runs the same way and returns output.pcd, map.ply and the estimated trajectory. Nothing about the pipeline waits on a GNSS topic.
Can LidarFlow georeference a map without GNSS data?
No. Georeferencing to WGS84 aligns the SLAM trajectory to a GNSS track, so the recording has to contain GNSS fixes at some point. No GNSS in the recording means no WGS84 alignment. This is a hard requirement, not a quality tradeoff.
Free while we are testing
Have a recording with no GNSS topic at all? Upload it. That path already works.