Two Different Questions Called Altitude

Ask a flight controller how high it is and you get an answer to a question you may not have asked. There are two distinct measurements, and confusing them is behind a large share of hard landings.

Altitude above mean sea level comes from the barometer and GNSS. It is what airspace rules and mission planners use, it is stable over long distances, and it tells you nothing about the ground. Fly a barometric hold at 200 feet over rising terrain and you will meet the terrain.

Height above ground level comes from a downward-looking range sensor. It is what matters for terrain following, for low-altitude spraying, and for the final 30 feet of every landing. It is also noisy, surface-dependent, and limited in range.

A serious aircraft carries both and fuses them, with clear rules about which one owns the altitude estimate in each flight phase. The handoff between them, typically somewhere between 15 and 50 feet AGL, is where you should expect to spend real tuning time.

What Each Altitude Sensor Actually Does

  • Barometer. Cheap, always present, and drifts. A passing weather front moves indicated altitude by 30 to 100 feet over a few hours, and the pressure field around a hovering multirotor is disturbed by its own downwash, so the port needs foam over it and a sheltered position. Good for relative altitude over minutes, useless as an absolute reference.
  • GNSS vertical. Roughly two to three times worse than horizontal accuracy, so 10 to 30 feet of error with plain GPS. RTK brings that to an inch or two but needs a base station or a correction service and a clear sky.
  • Lidar rangefinder. A single-point time-of-flight laser. Accuracy of an inch or better, range from 130 to 650 feet depending on unit and target reflectivity. Struggles over water, fresh asphalt, and in bright sun. The default choice for landing and terrain following.
  • Ultrasonic. Cheap, works to about 15 feet, and gets confused by rotor wash, tall grass, and echoes off nearby walls. Fine indoors, marginal outdoors.
  • Radar altimeter. Millimeter-wave units see through dust, fog, snow, and vegetation that stop a laser cold. Heavier and pricier, but for agricultural and firefighting work it is often the only sensor that keeps working.
  • Optical flow with a downward camera. Gives horizontal velocity, not altitude, but it is what holds position when GNSS drops out, and it needs a scale reference from one of the sensors above. Its failure modes overlap with those in GPS-denied drone navigation.

Why Landing Is Harder Than Hovering

Three effects converge in the last few feet.

Ground effect is the big one. Below roughly one rotor diameter of height, the recirculating flow under the disc increases thrust for the same throttle. The aircraft gets buoyant right when the controller is commanding descent, so a naive altitude loop overshoots, balloons, and then chops thrust. The fix is a descent profile that reduces gain and commands a fixed low descent rate below a set height rather than closing a position loop.

Sensor noise rises as you close in. Rotor wash kicks up dust and grass, the rangefinder starts reading the debris, and optical flow loses texture on a plain concrete pad.

Touchdown detection is the third. The controller has to know the gear is loaded before it disarms, or it keeps commanding thrust against the ground and tips over. Most implementations combine a low commanded descent rate, a throttle drop, and a vertical acceleration spike. Tune all three, and test on a slope.

Landing on a Target, Not Just on the Ground

Plain GNSS puts you within 3 to 10 feet of a commanded point. That is fine for a field and useless for a charging dock with contacts, or a rooftop pad, or a moving vehicle. Two approaches get you to inch level.

Visual fiducial tracking. A downward camera detects a printed marker such as an AprilTag or a nested pattern, computes the full relative pose, and feeds a position loop. Accuracy of 1 to 2 inches at touchdown is routine. Use a nested marker set, a large one visible from 100 feet and a small one still resolvable from 12 inches, because a single-size tag falls out of the frame exactly when you need it most. Lighting is the enemy: add IR illumination and an IR-pass filter if the pad has to work at night or in glare.

Active beacon. An infrared beacon on the pad and a matched sensor on the aircraft. Less flexible than vision, immune to most lighting problems, and common on commercial docking systems.

Either way, the mechanical design of the pad does half the work. A shallow funnel, tapered guides, or a V-groove converts a 3-inch landing error into a perfectly seated aircraft. That is much cheaper than chasing the last inch in software, and it is standard practice on the docks described in drone-in-a-box docking stations. If the aircraft has to detect its pad among obstacles on the way down, the sensing overlaps with drone obstacle avoidance.

Emergency Landing Behavior

Decide in advance what happens on low battery, lost link, GNSS loss, or a motor fault. The options are return to launch, land immediately in place, fly to the nearest surveyed alternate, or deploy a chute. Each needs a trigger threshold and a priority order, and they conflict: return to launch on low battery is wrong if the battery cannot reach launch. Model the energy honestly and default to the conservative branch. Above a certain mass, the right answer to a structural or propulsion failure is not a landing at all but the recovery systems covered in drone parachute recovery. Whichever logic you pick lives in the flight stack, so confirm your controller supports it before selecting hardware, per how to choose a drone flight controller.

Landing Gear Is Part of the Sensing Problem

Gear geometry sets how much lateral error and slope you can absorb. A wide stance with a low center of gravity tolerates a 10-degree slope; a tall narrow stance tips at 6. Add compliance, either sprung legs or an energy-absorbing foam core, so a 3 ft per second arrival does not transmit a shock spike into the airframe and the payload. Keep the rangefinder's field of view clear of the legs, and never mount it where a leg shadow crosses the beam during descent. Consider a switch or load cell in the gear for unambiguous touchdown detection, and design the whole assembly to survive the drop cases described in designing for drop test.

Get the Descent Right the First Time

Projects House works on altitude sensor selection and fusion, descent controller tuning, precision landing vision pipelines, and pad and gear mechanics for teams building autonomous aircraft. Send your platform, pad type, and accuracy target through our contact form.