When GPS drops, a drone has six real options — and they are not interchangeable. Some give velocity only, some give a full pose, one gives an absolute fix that never drifts, and one gives nothing but seconds. This post compares all six, with a full deep dive on each. Start here, then read the one that fits your build.
See also the GPS-denied navigation guide and Vision-Based UAV Navigation.
The six solutions at a glance
| Solution | What it gives you | Drift? | Needs | DIY cost | Best for |
|---|---|---|---|---|---|
| Optical flow + rangefinder | Velocity + altitude | Yes — grows with distance | Texture + light, low altitude | $22–$42 (MicoAir) | Indoor hover on the cheap |
| Visual-inertial odometry | Full 3D pose + velocity | Yes — grows without loop closure | Light, texture, companion computer | $1,400+ (VOXL 2) or a secondhand T265 | Local-frame missions, real autonomy |
| Lidar SLAM / LIO | Full 3D pose + a map | Slow; loop closure corrects it | Geometry to match, compute | ~$99 (2D RPLiDAR) to ~$750+ (3D Mid-360) | Dark warehouses, mines, mapping |
| UWB positioning | Absolute 3D position | No — zero drift | Surveyed anchors installed | $10–$30 per module (DWM3000) | Warehouses, swarms, repeatable indoor flight |
| Terrain / landmark matching | Absolute position fix | No — error stays flat with distance | Preloaded maps, visible features | Research-grade; no turnkey DIY stack | Long-range flight, EW environments |
| Dead reckoning + barometer | Seconds of position, altitude hold | Yes — grows with time² | Nothing (already onboard) | Free | Failsafe, buying time to land |
The deep dives
- Optical Flow + Rangefinder: The $30 Hover Aid — PMW3901 flow + TF-Luna, MicoAir MTF-01 at $32.99, ARK Flow, DJI’s VPS precedent, and why it caps out at low altitude over textured ground.
- Visual-Inertial Odometry: Full 3D Pose, No Infrastructure — OpenVINS and VINS-Fusion on a Pi or VOXL 2, the discontinued T265, Skydio X10, and the calibration burden nobody warns you about.
- Lidar SLAM / Lidar-Inertial Odometry: Mapping in the Dark — RPLiDAR A1 at ~$99, Livox Mid-360 at ~$750, LIO-SAM vs FAST-LIO2, Emesent Hovermap and Flyability Elios 3, and degenerate geometry.
- UWB Positioning: Absolute, Drift-Free, Anchor-Dependent — Nooploop LinkTrack on ArduPilot with no companion computer, $10–$30 Decawave modules, Verity’s warehouse drones, and why Marvelmind is ultrasonic, not UWB.
- Terrain Matching: The Fix That Never Drifts — Twist Robotics OSCAR, Blue Arrow/KEF, UAV Navigation VNS01, NASA Ingenuity’s 72 flights and its flight-72 failure, and why there is no turnkey DIY stack yet.
- Dead Reckoning + Barometer: The Failsafe You Already Own — PX4 EKF2 dead-reckoning mode, the 5-second no-aid timeout, a 28-second SITL reality check, and why stock ArduPilot can’t arm a mission with no GPS at all.
How to choose
Cheapest first flight indoors: optical flow + rangefinder. $30, grams of weight, no companion computer. Hover and slow indoor moves. Don’t expect missions.
Real autonomy with nothing installed: VIO for lit, textured spaces; lidar SLAM for dark or featureless ones. Both need a companion computer and both drift — VIO faster, lidar slower with loop closure.
Repeatable indoor work in one place: UWB. Survey the anchors once, get centimeter absolute position forever, in the dark, over blank floors. The drone is still blind to obstacles without another sensor.
Long-range flight where GPS is jammed: terrain matching — the only approach whose error doesn’t grow with distance — fused with VIO for smooth short-term motion. Every fielded system pairs them.
When everything else fails: dead reckoning is already running. Its job is to buy seconds, hold altitude, and land — not to fly the mission.
The honest architecture for most builds: dead reckoning as the failsafe, optical flow as the cheap velocity aid, VIO or lidar for real navigation, and terrain matching or UWB for the absolute fix when the mission demands it. No single sensor does it all.
Related reading
- GPS-Denied Drone Navigation Guide 2026
- Vision-Based UAV Navigation: Optical Flow, SLAM & Deep Learning
- ArduPilot Optical Flow: Setting Up GPS-Denied Indoor Navigation on Pixhawk
- Shield AI Hivemind: Autonomous Drone Swarms
- Mounting the Sensors for GPS-Denied Flight: Brackets, Vibration, and Known Offsets
- STLTools v1.0.1: Free, Open-Source Parametric STL Generator — print the brackets and cases these sensors mount in