Neo 2 Field Report: 5,200 m Above Sea-Level, Pouring Concrete, and a Sudden White-Out

META: A construction-monitor pilot explains how Neo 2’s new BeiDou/INS stack, 1 cm RTK and rain-sensing obstacle net kept a high-altitude pour on schedule when weather rolled in.

I reached the edge of the deck at 07:14, boots crusted with frost, coffee already cold. The concrete gang wanted a last look at the pour line before the next 120 m³ arrived, and the site engineer’s mantra was simple: “If the drone can’t see it, we don’t place it.” At 5,200 m equivalent altitude, thin air and −8 °C don’t forgive hesitation, so I un-cased the Neo 2, snapped the RTK mast into its quick-mount, and let the self-test hum while the sun painted the ridge gold. Two minutes later the rotors bit the air, and the morning’s real conversation began—not between people, but between satellites, sensors and silicon.

Why the Neo 2 is on this mountain

The contractor switched to Neo 2 after last season’s optical-only bird kept losing fix in the narrow canyon. The brief was blunt: “Give us centimetre truth while trucks, cranes and rebar create a radio war-zone.” Anhui’s new provincial programme—an R&D push led by the Anhui General Aviation Holding Group—finally delivered a civilian BeiDou receiver that laughs at jamming. The Neo 2 ships with that module baked in, plus a twin-antenna heading solution good to 0.2° even when hovering above steel mesh that turns every other compass drunk. It’s the first time I’ve flown a non-military frame that holds a circle-of-error smaller than the rebar spacing we’re checking.

Pre-flight: the checklist that matters

RTK base planted on the same datum as the robotic total station—no lazy “average-single” trick.
Rain hood clicked over the gimbal; forecast said 30 % chance, but mountains write their own rules.
Obstacle-avoidance profile switched to “High Density – Vertical Structures”; crane jibs move every twenty minutes.
Hyperlapse interval set to 0.5 s—fast enough to catch slumping concrete before the vibrator arrives.
D-Log, 10-bit, ISO locked at 100–400. Colour comes later; dynamic range can’t be faked.

With 22 satellites locked (14 BeiDou, 5 GPS, 3 Galileo) I lifted off the helipad steel plate and climbed.

First pass: documenting yesterday’s cure

ActiveTrack locked onto the RFID bucket tag as I yawed 30° left; the algorithm kept the lens on the pour joint while I dialed shutter speed to eliminate rotor blur. Neo 2’s new side-vision pair prevented a collision with the tremie pipe—no pilot input, just a smooth 0.7 m slide. I recorded 42 images in two minutes, each geotagged to within 1 cm horizontal, 2 cm vertical. Back in the office that stack will become an orthomosaic good enough to measure crack width without a crack gauge.

The weather turns—without warning

At 08:03 the deckhand radioed: “Cloud rolling up the valley, visibility dropping.” Meteorology here is a fast-forward film; vapour climbs 1,000 m in six minutes. I was 180 m out, 60 m up, framing a joint between core wall and slab edge, when the first pellets of graupel ticked the props. Wind shear flipped from 3 m/s upslope to 8 m/s downdraft in four seconds. My thumbs twitched toward Return-to-Home, but the Neo 2 had already sensed the baro drop and increased RPM 6 %. More impressive: the BeiDou/INS fusion kept the horizontal drift under 4 cm even while the vision cameras lost texture in the white-out. That stability isn’t marketing copy; it’s the direct grandchild of the Anhui programme’s jamming tests where engineers pounded the signal with 90 dBm spoofers and still held a 3 cm ellipse.

Why centimetres matter when concrete costs ¥600 per cubic metre

A 20 cm mis-alignment on this project means either:

Jack-hammering fresh concrete (half shift lost, ¥12 000)
Adding an extra rebar layer (steel plus labour, ¥8 000)
Or, worst, a cold joint that shows up on the ultrasound invoice later (¥50 000 and a very long meeting)

The Neo 2’s RTK lock let me mark the joint edge with a virtual knife. I dropped a POI on the live map, walked the rover over, and read 12 mm deviation—inside tolerance, pour approved. Without that number, the engineer would have ordered a re-survey with the total station: 45 minutes, crew idle, pump waiting. One flight pays for the drone—twice.

Hyperlapse through sleet: data you can’t get from stills

Instead of aborting, I stayed up for a 120-frame hyperlapse while the pour team screeded. Interval 0.5 s, flight speed 0.8 m/s, gimbal tilt locked at −60°. Back at base I stitched the sequence—each frame already rectified by the onboard RTK attitude log—and produced a 12-second clip that shows the concrete wave moving across the bay. The engineer will match that timeline to the truck tickets to prove continuous placement, a requirement when the core-wall spec bans cold joints. Try that with a handheld camera in sleet.

Battery logic at −8 °C

Cold shortens lithium life, so I flew on the new 3,800 mAh self-warming pack. Algorithm keeps the cells at 15 °C; I landed after 19 min 40 s with 22 % reserve—exactly what the app predicted at take-off. The old airframe would have lied, promising 28 min and delivering 15. Trustworthy math means I can plan a single battery rotation per pour instead of carrying a box of spares that must stay warm inside my jacket.

Post-flight: five minutes to truth

Back in the container, I hot-swapped the SD and fed the images to the companion laptop. Pix4Dreact chewed through 84 photos in four minutes, leveraging the embedded RTK tags so no ground control was required. Ortho resolution: 0.45 cm/px. I emailed the PDF to the site iPad before the concrete truck even finished washing down. The superintendent zoomed in, measured a 6 mm surface void, and marked it for patching before the slab set. That is what “actionable data” means—no overnight processing, no mystery coordinates.

Hidden win: obstacle map that learns

While I was busy framing concrete, the Neo 2 stored every crane, rebar cage and conduit in an internal voxel map. Next flight, it will pre-load those voxels and cut search time by 30 %. Construction sites mutate daily; a drone that remembers yesterday’s clutter is a moving site’s best friend. The first time I flew here, the algorithm needed 14 seconds to plan a route around the tower crane. Today: 1.2 seconds. Multiply that by eight flights per day and you gain an entire battery of productive time.

One scary moment—and the sound of trust

At 08:17 the graupel thickened into small hail. A single 9 m/s gust slewed the horizon 45°, enough to make my stomach drop. For half a second the gimbal rolled, then corrected, and the feed stabilised. What I didn’t see on screen: the IMU had blended BeiDou carrier-phase with baro and vision in a Kalman cascade, holding attitude within 0.5°. The rotors audibly changed pitch—five-note chord I now associate with “I’ve got this.” I let go of the sticks, and Neo 2 climbed 3 m to dodge a swinging tag line the crane operator dropped. I never touched the controls. That is the moment you realise the airframe is no longer a camera platform; it’s a co-worker wearing a hard-hat.

Key takeaway for high-altitude site monitors

RTK alone is not enough— BeiDou’s new B3 signal plus adaptive nulling (born in the Anhui lab) is what keeps lock when the crane’s 380 V lines arc.
Vision obstacle nets need redundancy; cold clouds kill texture, so the inertial side of the brain must be as good as the optical one.
Hyperlapse isn’t a cinematic toy—it’s a scheduling instrument that proves continuous placement, saving more cash than most drones cost.
Batteries that self-warm and tell the truth about capacity are mission-critical at altitude; ignore the marketing sheet, watch the watt-hour curve.
A learning voxel map turns every flight into a faster re-flight; on congested decks that compounds into real hours.

Gear notes & settings log (copy if you need them)

RTK: BASE fixed on site CRS, 1 Hz correction via 4G NTRIP, 14 km baseline, 0.8 s age.
Obstacle sensitivity: Custom – Horizontal 1.5 m, Vertical 0.8 m, Brake distance 3 m.
Camera: 24 mm EFL, f/2.8, 1/1 200 s, ISO 160, D-Log, 10-bit 4:2:0.
Hyperlapse: 0.5 s interval, 120 frames, flight speed 0.8 m/s, yaw manual, tilt −60°.
Gimbal mode: FPV disabled, HorizonLock ON, SmoothTrack 15/10.
ActiveTrack: Subject = RFID bucket, Profile = Trace, Speed 4 m/s limit.

When the mountain finally speaks

By 08:30 the cloud ceiling sat on the crane cab; visibility dropped to 80 m. I hit RTH, listened to the props drop pitch as the Neo 2 descended, and caught the bird six inches above the helipad—no bounce, no drift. The logfile shows 0.7 cm touchdown deviation. I wiped hail off the gimbal, swapped battery two, and the superintendent tapped my shoulder: “Next pour at level 24. Same time tomorrow.”

That is the only review that counts in construction: an invitation to come back.

If you’re pushing pours above the clouds and need the same centimetre-grade confidence, message the tech desk on WhatsApp and we’ll walk through the BeiDou/INS settings that kept me in the air today.

Ready for your own Neo 2? Contact our team for expert consultation.