Satellite positioning is the thing most drones depend on most, and it disappears exactly in the environments that are commercially interesting: inside buildings, in warehouses and tanks, in tunnels and mines, under bridges, among tall buildings where reflections corrupt the fix, and in areas where the signal is deliberately jammed. A drone that loses position typically drops to manual control or initiates a landing, and either way the mission is over. Flying without satellites means replacing that one absolute reference with a combination of onboard sensing — optical flow, visual-inertial odometry, LiDAR odometry, or external anchors — fused so that the aircraft always has a position estimate it can act on.
Why the IMU Alone Is Not the Answer
Every drone already carries an inertial measurement unit, and it is tempting to think accelerometers and gyros can dead-reckon through a satellite outage. They cannot for long. Integrating acceleration twice to get position compounds bias and noise quadratically, so a MEMS IMU drifts meters within seconds. The IMU's real job is to bridge short gaps and to provide high-rate attitude, while some other sensor supplies the position or velocity correction. Two related helpers also fail indoors: magnetometers are corrupted by steel structure and motor currents, and barometric altitude drifts with HVAC pressure changes, so altitude usually moves to a downward rangefinder.
Four Families of Solution
Optical flow
A downward-facing camera measures how much the image shifts between frames and, combined with a height measurement, produces a horizontal velocity estimate. Cheap, light, computationally modest, and effective at holding a stable hover over a textured surface. Limits: it fails over uniform floors, water, and in darkness without illumination, and because it measures velocity rather than position it drifts steadily over any long traverse.
Visual-inertial odometry and SLAM
One or more cameras identify landmarks in the environment, track them frame to frame, and simultaneously build a map while locating the aircraft within it. This is the strongest general-purpose approach and the one that makes autonomous flight through an unmapped building possible. It also gives loop closure: recognizing a previously visited place and correcting accumulated drift. The costs are real — meaningful compute load, sensitivity to lighting changes, poor performance against blank walls and reflective surfaces, and fragility in scenes that change between passes.
LiDAR odometry
Laser ranging does not care about ambient light at all, which makes it the honest answer for dark tanks, sewers, and mine drifts. It is accurate and robust in geometrically featureless-looking but structurally rich environments where cameras struggle. You pay in mass, power draw, and unit cost — and in long uniform corridors, where every scan looks the same, LiDAR has its own degeneracy problem along the tunnel axis. The payload tradeoffs are covered in LiDAR drone mapping.
External anchors and fiducials
Ultra-wideband beacons at surveyed positions, or printed visual markers the aircraft recognizes, deliver accurate absolute position at modest cost. The catch is installation: anchors have to be placed and surveyed in advance. That makes them excellent for a warehouse, a factory, a fixed inspection bay, or a docking station approach corridor, and useless in unprepared terrain.
The Real Principle: Fuse, Do Not Choose
A serious system does not pick one method. It runs several concurrently and shifts weight between them according to how trustworthy each measurement is at that instant: satellites outdoors, vision indoors, optical flow for hover hold, LiDAR in the dark, and the IMU tying it together at high rate. This is normally implemented as an extended Kalman filter or a factor-graph estimator with per-sensor covariances and health monitoring.
The hardest engineering problem is not any single sensor — it is the handoff. When a source drops out or comes back, the position estimate must not jump. A discontinuity in the estimate is fed straight into the controller and comes out as an abrupt, sometimes violent, maneuver. Practical mitigations include maintaining a local drift-free odometry frame that the controller uses, and applying global corrections slowly as a smooth offset. Any GPS-denied system needs an explicit, tested behavior for degraded state: hold position on the remaining sensors, alert the operator, and land in a controlled way if the estimate quality falls below a threshold. This is closely related to how obstacle avoidance is layered — the last line of defense has to work locally, without any link.
The Compute, Power, and Thermal Budget
Vision-based navigation needs processing power, and on an aircraft every watt is battery mass and every watt is heat. Three levers reduce the burden: run at lower resolution and process only regions of interest; move inference onto a dedicated accelerator instead of a general-purpose CPU, in the spirit of running models on constrained hardware; and offload part of the pipeline to a ground computer when — and only when — the link is reliable enough to depend on. Heat rejection at altitude and in enclosed spaces is a genuine design constraint, which is why thermal management ends up on the critical path of a navigation payload.
How to Test It Honestly
Bench and simulation testing catch software bugs, not estimator failures. Build a test progression: replay recorded sensor logs offline, fly in a netted indoor space with an external tracking reference to measure actual position error, then run repeatable field trials with deliberate sensor dropouts — cover the camera, kill the satellite feed, fly at a blank wall, fly into darkness. Record the estimator's own confidence output alongside ground truth, because the system's ability to know it is lost matters as much as its accuracy when it is not.
Where This Is Worth Money
Three markets are clear. Confined-space inspection — tanks, boilers, stacks, penstocks, and tunnels where there is no signal and no safe place for a person. Warehouse and yard inventory — cycle counting indoors, where anchors or fiducials can be installed and the ROI is measured in labor hours. Defense and public safety — operations where satellite navigation may be jammed or spoofed, and where alternative navigation is a threshold requirement rather than a feature, in the same family of concerns as counter-drone systems. In all three, GPS-denied capability is not an upgrade; it is the entry ticket. If satellite-based positioning is available, incidentally, the accuracy chain looks completely different — see drone photogrammetry.
If you are building an aircraft that has to operate where satellites do not reach, Projects House develops navigation payloads, sensor fusion, and the mechanical and thermal design around them. Describe your environment through our contact form and we will tell you which sensor mix the mission actually needs.