GNSS-denied environments: how a drone knows where it is when the satellites disappear

Outdoors, a drone uses GNSS (GPS, QZSS, Galileo and others) to know its position to within decimetres or metres, and to hold station in wind. When concrete or steel blocks the sky, fewer satellites are visible and signals bounced off walls (multipath) make the position jump. Indoors and underground the signal simply never arrives.

Instead of GNSS, the aircraft observes its surroundings with onboard sensors and integrates its own motion. The two main approaches are visual-inertial odometry (VIO), combining cameras with an inertial measurement unit (IMU), and LiDAR SLAM, which measures the shape of the surroundings with laser ranging. SLAM estimates position and builds a map at the same time, and can correct accumulated error when it revisits a place (loop closure).

GNSS denial sits at the heart of inspection: under bridges, in tunnels, in dam galleries and outlet works, sewers, utility tunnels, inside chimneys and boilers, storage tanks, ship ballast tanks and mine drifts. The more dangerous a place is for people, the less sky it tends to have.

• Predict reception before entering: check on drawings how much sky is visible and split the flight plan into GNSS-reliant and GNSS-free segments.
• Match positioning to the job: vision-based, self-avoiding aircraft under bridges; LiDAR SLAM plus lighting inside dark enclosures; markers or UWB for repeated fixed routes.
• Keep manual skills: train pilots to exit safely in attitude mode (no position hold) when automatic positioning fails.
• Plan for link loss: radio also struggles underground and inside metal boxes. Decide on relays, tethers, and what the aircraft does on link loss (hover or retrace) in advance.
• Record the positioning method: an image geotagged by GNSS and one positioned by SLAM carry different accuracy. Keep that for whoever compares later.

In many countries, indoor and underground flight falls outside aviation rules or is treated differently; in Japan, indoor flights are outside the aviation flight rules, and spaces clearly separated from the outside count as indoor even with open entrances (tunnel or underground-passage interiors, for example). That does not mean anything goes. Sewers and tanks carry oxygen-deficiency and toxic-gas hazards governed by occupational safety law, and plants have hazardous-area (explosion) classification. Check the aviation procedure and the site-safety procedure separately.

Related: oxygen deficiency & toxic gas, explosive atmospheres, confined spaces, world regulations.

• Water, mirrors and glass: cameras and LiDAR misjudge range through reflection and transmission.
• Dense dust, steam and smoke: cameras go blind and LiDAR scatters; radar-based navigation is an active research area.
• Long featureless pipes: geometric degeneracy. Autonomous round trips through kilometres of sewer remain largely at the research stage.
• Guaranteed accuracy: SLAM error grows with distance, and the industry has no common way to state “accurate to N cm” for inspection records.

Every structure has an “inspection map”. On entering, a drone matches its sensors to that map and knows where it is, returning to last time’s viewpoints within centimetres and shooting from the same angles. Satellite visibility no longer shapes the inspection plan; under bridges and inside pipes become routine patrol routes just like open sky. That is the editors’ outlook.