A paper on arXiv by Junrui Zhang presents DGV-DLIVO, a method that combines lidar, inertial and camera data to locate inspection robots in long underground tunnels. In hydropower spillway tunnels and drainage galleries, repeating structures give lidar too little to work with along the tunnel, and weak texture and changing light make cameras unreliable.
DGV-DLIVO uses the direction in which lidar fails as a guide. It picks useful image regions, ties them to lidar points and adds camera information only for the motion lidar cannot measure, while keeping the reliable lidar constraints.
In real tests in hydropower tunnels, with the sensors on a cart and carried by hand, leading lidar-inertial, lidar-camera and camera-inertial methods stalled, oscillated, fell short, drifted or diverged. DGV-DLIVO kept estimating motion along the tunnel and achieved a 91.4% progress ratio over the 40 m sequence.



