Sending a human into a storage tank, a boiler, a sewer, or a mine drift is one of the most regulated and dangerous tasks in industry. A confined space entry under OSHA 29 CFR 1910.146 requires a permit, atmospheric testing, an attendant, a rescue plan, and often scaffolding costing more than the inspection. What takes a crew three days of setup can take a drone forty minutes.
That business case is strong. The engineering case is harder, because almost everything an outdoor drone relies on is missing inside a steel vessel.
No GPS, and Not Much Else Either
Inside a tank there is no satellite signal, no magnetometer worth trusting — you are flying inside a large ferrous object — and often too little texture or light for visual odometry. The navigation stack has to be rebuilt from what is left.
- Visual-inertial odometry from stereo or multi-camera rigs fused with the IMU. Good where the surface has texture and you control the lighting; it fails on clean painted steel, in steam, and in heavy dust.
- LiDAR odometry and SLAM. A lightweight scanning or solid-state LiDAR gives geometry independent of lighting, which is why it dominates in tunnels and shafts. It struggles in featureless straight pipes where every scan looks identical along the axis.
- Optical flow plus rangefinders as a floor: cheap, robust, enough to hold position when the higher-level estimator degrades.
- Ultra-wideband beacons placed at the entry hatch, where you control the geometry.
Serious confined-space aircraft fuse several of these and manage graceful degradation between them — the estimator must fail into a stable, pilot-flyable hover rather than a drift toward a wall. The general problem is covered in depth in our article on GPS-denied drone navigation.
Contact Is Not a Failure Mode, It Is Normal
Outdoor drones treat touching something as a crash. Confined-space drones cannot. Passages are narrow, ventilation drafts are unpredictable, and the whole point is to get close to the surface being inspected.
The two design answers are a full protective cage or shrouded ducted rotors. A gimballed spherical cage — where the aircraft rotates freely inside an outer frame — lets the drone roll along a wall or ceiling without transferring the disturbance into the flight controller, which is why the concept keeps reappearing commercially. A fixed cage is lighter and cheaper but couples every bump into attitude. Either way you pay: a cage adds 20 to 40 percent to aircraft mass and disturbs propeller inflow, costing thrust and flight time.
Cages also change the obstacle-avoidance philosophy. Outdoor systems stop before contact; indoor systems cannot see well enough or stop fast enough at close range, so they are designed to tolerate contact and recover. Our overview of obstacle avoidance sensors and algorithms explains why time-of-flight and radar behave differently indoors than the marketing suggests.
You Have to Bring the Light
Inside a tank there is zero ambient light, and the inspector needs imagery good enough to judge weld condition, coating breakdown, and pitting. Lighting is a primary payload, not an accessory.
Practical figures: 5,000 to 20,000 lumens of onboard LED lighting drawing 50 to 200 W, a meaningful fraction of a small aircraft's power budget. Beam design matters as much as raw output. Direct on-axis light on a shiny or wet surface produces glare that hides the defect you are hunting, so offset arrays, diffusers, cross-polarized filters, and switchable panels for oblique raking light all earn their weight — raking light at a shallow angle is how surface deformation becomes visible. Add heat management: high-output LEDs in an enclosed steel space with no airflow will thermally throttle.
Dust, Vapor, and Explosive Atmospheres
Many confined spaces are hazardous locations. A fuel tank, a grain silo, a paint booth, a coal mine, a wastewater digester — each can hold a flammable atmosphere where a spark or hot surface is an ignition source. Brushless motors, ESCs, LiPo packs, and high-power LEDs are all textbook ignition sources.
There are three honest approaches. Purge and gas-test the space first, which is common industrial practice and lets you fly a standard aircraft. Build an intrinsically safe or explosion-protected aircraft to ATEX or IECEx and the NEC Class/Division scheme — a long, expensive path that costs power and mass. Or restrict the product to non-hazardous spaces and say so in the documentation. Pretending the question does not exist is how a supplier ends up in litigation.
Dust brings its own problems even where it is not explosive: it blinds optical sensors within minutes, coats lenses, packs into motor bells, and kills bearings. Sealed motors, sacrificial lens covers, and a realistic cleaning interval belong in the design.
Comms Through Steel
A 2.4 or 5.8 GHz link does not survive a steel vessel wall. Once the aircraft is inside and around a bend, there is essentially no signal. The options:
- A thin tether carrying fiber or copper for video and control, and sometimes power. It solves comms and endurance in one move at the price of snag risk and careful tension management — the power side is covered in tethered drone systems.
- A repeater at the hatch, relaying through a node lowered into the space.
- Fly autonomously and recover the data on landing, with only degraded live telemetry. This demands genuine onboard autonomy and reliable return-to-hatch behavior.
Most fielded systems use a tether or a hatch repeater, because an inspector who cannot see live video cannot direct the inspection.
The Weight Budget Decides Everything
Every requirement above costs mass, and all compete for one small budget on an aircraft that must fit through a 20-inch manway. A realistic split on a 2 to 4 kg aircraft:
| Subsystem | Share of all-up mass |
|---|---|
| Airframe and cage | 25–35% |
| Propulsion and ESCs | 15–20% |
| Battery | 20–30% |
| Navigation sensors and compute | 8–15% |
| Lighting | 5–12% |
| Inspection payload | 8–15% |
Flight times of 8 to 15 minutes are normal and fine — the mission is short and the aircraft lands at the hatch. Resist adding endurance at the cost of size: fitting through the opening is the hard constraint, and an aircraft that cannot enter the space has zero endurance. Payloads are typically a high-resolution visual camera, sometimes an ultrasonic thickness probe on a contact arm, and often a thermal imaging payload for refractory and insulation work. The wider inspection context, including how this fits alongside external asset work, is in drone infrastructure inspection.
Confined-space aircraft are a clear case where a purpose-built design beats an adapted commercial platform. Projects House develops unmanned systems for US clients from requirements through flight test. If you have an inspection problem that currently needs a permit and scaffolding, describe it through our contact form.