TECHNOLOGY / 07
Navigation and autonomy: knowing where you are, and where to go next, beyond GNSS
Whether an inspection drone can capture the same image from the same spot every time depends on positioning and path planning. This chapter covers RTK, visual-inertial odometry, LiDAR SLAM, UWB and markers, obstacle-sensing, levels of autonomy, and how to plan paths that leave no gaps.
On this page
Beyond GNSS: ways to know where you are
In open sky, GNSS (GPS, QZSS, Galileo and others) is the easiest option, but standalone positioning carries errors of metres. To return to “the same position as last time”, RTK (centimetre-level positioning using corrections from a base station) is the norm. The catch is that inspection sites are exactly where the sky is blocked. So positioning by *looking at the surroundings* is added.
| Method | How it works | Strength | Weakness |
|---|---|---|---|
| RTK-GNSS | Corrections from a base station or network bring GNSS to centimetres | High accuracy and absolute coordinates in open sky | Loses fix under cover; vulnerable to multipath |
| VIO (visual-inertial odometry) | Combines image motion from cameras with the IMU | Light, cheap, works indoors and out | Weak in darkness, on featureless surfaces and over water; drifts over time |
| LiDAR SLAM | Measures the shape of the surroundings with lasers and builds a map while localising | Works in the dark, accurate geometry | Heavy, costly; confused where geometry repeats, e.g. long straight tunnels |
| UWB | Measures radio ranges to anchors at known positions | Stable indoors without GNSS | Anchors must be installed and surveyed; limited area |
| Markers (AR tags etc.) | The camera reads markers at known positions | Cheap; handy for landing and reference points | Only where markers are visible |
| Total-station tracking | A ground survey instrument tracks a prism on the aircraft | Very accurate | Needs line of sight, equipment and staff |
Real aircraft fuse these with the EKF described in flight principles. Redundancy — one source holding up when another weakens — is the key to flying safely near structures. Site-specific conditions are covered in GNSS-denied environments, darkness and dust & smoke.
Eyes for not crashing
Knowing your position is not enough; you need to know what is around you. Obstacle sensors each have their strengths and blind spots. The scariest things in inspection are thin objects: wires, cables, handrails, branches.
| Sensor | Strength | Weakness |
|---|---|---|
| Stereo / omnidirectional cameras | Wide coverage at once, shape information | Darkness, glass, blank walls, thin wires |
| LiDAR | Accurate range even in the dark, 3D shape | Weight and cost; glass and black surfaces; rain, fog |
| ToF / infrared ranging | Light, good at short range | Sunlight interference, narrow field |
| Ultrasonic | Cheap, detects glass | Short range only; noise from rotor wash and sound |
| mmWave radar | Works in rain, fog and dust; can sometimes detect wires | Coarse resolution, poor shape detail |
Deciding stand-off yourself starts at the design stage. With checking overlap and separation between plans before flight (Airffic Architect) you can screen whether two aircraft on constant-velocity paths come too close. It is a design-time screen, not a substitute for permission or operational judgement.
Levels of autonomy
“Autonomous flight” covers a lot of ground. For inspection it helps to think in roughly these stages (this is this guide’s own framing, not an industry standard).
| Stage | What the aircraft does | What people do | Example |
|---|---|---|---|
| Manual + stabilisation | Holds attitude and position | Everything about path and capture | Close-up imaging by a skilled pilot |
| Assisted | Collision avoidance, distance keeping, auto capture | Gives rough direction | Sideways travel at fixed distance from a wall |
| Waypoint | Flies a predefined route automatically | Builds the route and monitors | Outdoor grid capture with RTK |
| Autonomous exploration & capture | Builds its own path from live mapping and fills gaps | Sets area and goal, reviews results | Automated scans inside box girders or tanks |
| Remote, resident autonomy | Takes off, returns and charges from a dock on its own | Supervises several aircraft remotely | Scheduled patrols; see operations |
Path planning: capturing without gaps
Inspection path planning differs from ordinary “go from A to B”. The goal is to capture every surface of the asset at a set resolution and overlap, with nothing missed — this is coverage planning. Work through it in this order:
- Set stand-off from the required resolution (millimetres per pixel; see imaging and sensors).
- Set shot spacing and line spacing from the image overlap in both directions.
- Decide a camera orientation square to each surface. Use spirals or rings for curved surfaces and cylinders.
- Set how close you can go from minimum stand-off, wind and ceiling effect (flight principles).
- Split the route by battery capacity, with swap and return points.
- Check the blind spots on a 3D model and add capture to fill them.
| Asset shape | Path pattern | Examples |
|---|---|---|
| Flat surfaces | Lawnmower grid | Solar PV, facades, dam faces |
| Cylinders and towers | Spirals or rings by height | Chimneys, storage tanks, turbine towers |
| Linear | Back-and-forth along the line, or parallel passes on both sides | Power lines, railways, pipelines |
| Complex 3D | Select viewpoints from a 3D model and link them (view planning) | Bridge soffits, plant piping |
| Enclosed interiors | Autonomous exploration, then gap-filling | Tank interiors, boilers, sewers |
Inspection paths are drawn to leave no gaps; delivery routes start by discarding what cannot be flown. The contrast is clear next to building delivery routes by removing infeasible legs first (Airffic Logistics).
Repeatable missions: the same photo again
The value of inspection lies in comparison with last time. Whether a crack has grown or rust has spread is hard to judge unless images share the same position, orientation and lighting. Repeatable missions need:
- Common coordinates: an RTK frame, or reference points and markers fixed to the structure. In GNSS-denied spaces, re-localise against the previous 3D map.
- Stored routes: waypoints and camera poses saved and linked to member IDs.
- Recorded capture conditions: camera settings, lighting, time, weather.
- Tolerances: decide in advance how many centimetres and degrees still count as “the same photo”.
Routes should be kept not as a tool for one day’s flight but as part of the asset register. If routes and camera poses are tied to members, you can retake the photo from ten years ago even after the aircraft or vendor changes.
What is still hard
- Featureless spaces: long straight tunnels, white walls, water. VIO and SLAM get lost when everything looks the same.
- Thin obstacles: wires and cables are easy for most sensors to miss.
- Moving things: workers, vehicles, cranes, swaying branches break the assumptions of mapping.
- Long-term drift: the longer you fly without GNSS, the more position error accumulates; you need ways to re-anchor to reference points.
- Certification and accountability: how to demonstrate and approve autonomous decisions to third parties. Regulation is still catching up.
Check the evidence
Rules, figures and cases cite the sources below. Future scenarios and design proposals are the editors’ own view.
- Cabinet Office, Japan: QZSS (Michibiki) Centimeter Level Augmentation Service (CLAS)Cabinet Office, Government of Japan · Full text read
- PX4 Autopilot User Guide: Visual Inertial Odometry (VIO)Dronecode Foundation / PX4 · Full text read
Sources checked: 2026-10-22