How a Drone Navigates When GPS Is Not Available
Most drones rely on satellite positioning to hold position and follow a mission plan. That works outdoors in open sky, but it degrades or fails in environments where the signal is weak, reflected or deliberately disrupted: under tree canopy, inside structures, urban canyons, and areas where satellite signals are jammed or spoofed. GPS-denied navigation is the set of techniques that lets an aircraft estimate where it is and move safely without trusting a satellite fix.
The core idea is to fuse other sensors — cameras, inertial measurement units, and sometimes laser rangefinders or LiDAR — into a continuous estimate of position and motion. This explainer walks through the main approaches, the sensors and the onboard compute they require. Builders sourcing hardware can start from machine vision cameras, autopilots and lidars.
Why GPS Fails and What It Costs
Satellite positioning needs an unobstructed view of multiple satellites. It is vulnerable to several failure modes:
When the satellite fix is lost, a flight controller that only trusts GNSS cannot hold position. A GPS-denied stack gives the aircraft an independent position estimate so it can continue its mission, hover accurately, or execute a safe recovery even when the satellite link is gone.
Visual-Inertial Odometry: Cameras Plus IMU
The most common GPS-denied technique is visual-inertial odometry, or VIO. It tracks features in consecutive camera frames and combines that visual motion estimate with the aircraft's inertial measurement unit, which measures acceleration and angular rate. The result is a continuous estimate of where the camera has moved, even without any external reference.
VIO works indoors and outdoors, in corridors and under canopy, wherever there is enough visual texture to track. Its limits are motion blur, low light, and feature-poor scenes like uniform walls or open sky, where there are few features to track. A stereo or depth-capable camera, such as the stereo machine vision camera, gives the VIO stack both image and depth information, which improves robustness.
SLAM: Building a Map While Localizing
Where VIO estimates motion along a path, simultaneous localization and mapping, or SLAM, goes further: it builds a map of the environment while keeping track of where the aircraft is within that map. This is valuable for repeatable missions, inspection of known structures, and for returning to a previously visited point.
SLAM comes in two main flavors:
Many systems combine them. LiDAR sensors for drone SLAM are collected in the lidars category, while the cameras that drive visual approaches are in machine vision cameras.
The Companion Computer: Where the Software Runs
VIO and SLAM are not software a typical flight controller can run. They need real computer vision processing — per-frame feature tracking, depth estimation and sensor fusion — which is handled by a companion computer carried alongside the flight controller. The companion computer runs the navigation stack and publishes position estimates to the autopilot over a telemetry link.
Choosing a companion computer is a trade between onboard compute, weight, power draw and thermal budget. It must be powerful enough to run the chosen SLAM or VIO stack at a rate that keeps the position estimate fresh, yet light and cool enough to fly. Smaller depth cameras, such as the compact depth machine vision camera, are designed to pair with this class of onboard compute. When you source compute, confirm it can interface with your flight controller and that you have the power and thermal headroom to run it continuously.
Sensor fusion underneath all of this is what turns raw measurements into a stable estimate. An extended Kalman filter, or equivalent estimator, combines the visual or LiDAR motion data with the IMU and any remaining GPS information, weighting each source by its expected reliability. When satellite signals are healthy, the estimator trusts them; when they degrade, it leans more heavily on the visual and inertial sources. The quality of this tuning, and the time synchronization between the camera frames and the IMU samples, often matters more than the raw horsepower of the companion computer.
How the Pieces Talk: Autopilot and MAVLink
The companion computer does not directly fly the aircraft. It produces a position and velocity estimate and sends it to the autopilot, which closes the actual flight-control loop. The standard bridge between them is a telemetry protocol such as MAVLink, over which the companion computer emits position-vision or odometry messages.
This separation matters for design: the autopilot stays responsible for stabilization and safety, while the companion computer is a removable, upgradable navigation module. If the vision stack fails, the autopilot can fall back to its last-known state or a safe behavior. Pick an autopilot in the autopilots category that accepts external vision or odometry input, and confirm the message interface your software stack supports before you commit to hardware.
Sensor and Compute Selection by Environment
There is no single best stack; the right combination follows the operating environment.
Budget for testing in the actual operating environment, because lab performance rarely transfers cleanly to real canopy, dust or vibration.
Integration Pitfalls to Plan Around
A GPS-denied stack fails most often at integration, not at the core algorithm. Common issues include:
Treat the navigation stack as a system, and test its failure modes as deliberately as its nominal performance. Build test scenarios that intentionally degrade or remove each sensor, and confirm the aircraft behaves predictably rather than drifting or crashing.
Keep a calibrated reference configuration and a set of benchmark test routes so changes to firmware or sensors can be compared against a known baseline rather than judged from impression.
Frequently Asked Questions
What does GPS-denied navigation actually mean for a drone?
It means the drone estimates its position and motion from onboard sensors — cameras, inertial measurement units and often LiDAR — rather than from satellite signals, so it can hold position and fly missions where GNSS is weak, blocked, jammed or unavailable.
What is the difference between VIO and SLAM?
VIO estimates motion by tracking visual features and fusing them with inertial data. SLAM goes further and builds a map of the environment while localizing within it, which supports repeatable missions and return-to-point tasks.
Do I need a companion computer for GPS-denied flight?
Yes. The computer vision and sensor fusion that VIO and SLAM require run on a companion computer separate from the flight controller, which publishes position estimates to the autopilot over a telemetry link.
Can LiDAR replace cameras for indoor drone navigation?
LiDAR SLAM works well in low light and featureless visual scenes, but it is heavier and costs more. Many systems combine LiDAR with vision and IMU fusion for robustness across environments.
How does the vision system connect to the flight controller?
Typically over a telemetry link using a standard protocol such as MAVLink, with the companion computer sending position-vision or odometry messages. The autopilot keeps control authority and can fall back if the vision estimate is lost.


