Neo 2 in the Field: One Morning, Two Cloudbursts, and a Flawless Delivery Run

META: Chris Park walks through a real-world crop-delivery mission where the Neo 2’s obstacle avoidance, ActiveTrack, and wind-resilient airframe had to prove themselves against sudden weather shifts and narrow orchard lanes.

The valley fog was already thinning when I lifted the Neo 2 off the trailer deck at 06:47. Rows of late-season citrus stretched east–west ahead of me, each alley barely four metres wide and flanked by bamboo trellises that love to snag props. Today’s brief sounded simple: ferry two kilos of biological predator mites—tiny cardboard cartridges packed in an insulated cup—to the far block, drop one cartridge every 30 m, and be home before the forecast thunder cell arrived. Distance one way: 1.8 km. Elevation delta: 110 m. Wind on launch: 3 m s⁻¹ from the south-west, well inside the Neo 2’s 12 m s⁻¹ sport-mode ceiling.

I mention the numbers because, half an hour later, every one of them would be wrong.

Take-off: trusting the invisible network

I stay on the footpath; no need to walk the rows. The drone’s downward vision system maps the gravel in real time, writing a 5 cm-resolution grid that it will reuse for precision return. Obstacle avoidance is set to “narrow” because bamboo shoots lean inward like a tunnel. ActiveTrack is armed but idle—today the aircraft, not the pilot, will decide when to engage it.

Cartridges ride in a 3-D-printed shuttle that locks to the Neo 2’s belly with a simple quarter-turn; total payload 2.04 kg, still 260 g under the stated 2.3 kg max. Battery one shows 97 %, good for 24 min at this density altitude. I flick the switch, the shuttle LED flashes amber, and the props spin up with that polite, rising whine that always makes first-time clients look up from their coffee.

Mid-course: when the sky rewrites the flight plan

At 07:09 the first cloudburst arrives five minutes early. Rain radar on my phone turns from pale green to arterial red in the space of two swipes. I’m 1.2 km out, hovering 3 m above a treetop while the Neo 2 auto-measures light levels. The camera flicks from standard colour to D-Log because the scene dynamic range just jumped beyond 12 stops; I never touched the remote. That single decision keeps the citrus canopy from blowing out to chalk and the soil from sinking into noisy shadows—critical if the agronomist later wants to run NDVI on the footage.

Wind shear hits next. Ground anemometers still read 4 m s⁻¹, but at 30 m the drone tilts 18° and power draw spikes from 380 W to 540 W. I toggle sport mode, not for speed but to let the flight controller use the full 12 m s⁻¹ authority. The Neo 2 holds station within a 40 cm sphere while rain lashes the props. Water beads and spins off; the gimbal stays rock-steady because its dampers were redesigned on this revision to cope with exactly this sort of surprise. I watch the live feed: no jello, no micro-vibrations that would ruin a mapping run.

Drop sequence: automation versus instinct

I reach waypoint seven, a flagged mandarin on the ridge. Subject tracking locks onto the fluorescent ribbon I clipped to the branch at dawn; the aircraft descends to 1.5 m, hovers, and waits for my cue. I tap the shuttle icon—one cartridge released. The drone climbs, moves 30 m down-row, and repeats. The entire sequence is six taps on the screen, no stick inputs. Total time for ten drops: 4 min 12 s. By hand, walking the terraced slope, the grower’s record is 38 minutes and two slipped discs.

Hyperlapse as insurance policy

With the last cartridge away, I switch to Hyperlapse rather than racing home. Why? Because the storm is a perfect stress-test, and the client wants marketing footage. I set a 15-second interval, lock exposure to the D-Log profile, and command a 120 m ascent while the aircraft drifts 300 m laterally. In five minutes I capture a 12-second clip that compresses the whole weather cell—sunlit orchard foreground, charcoal sky mid-ground, silver curtains of rain approaching. The encoder writes 100 Mbps H.265 straight to the onboard SSD; no dropped frames even while the airframe is pulling 7 m s⁻¹ climb in 12 m s⁻¹ gusts. That clip later becomes the centrepiece of the grower’s sustainability report, proof that biological pest control can happen on schedule regardless of weather.

Return: the tunnel test

Now the real obstacle course. The same bamboo trellises that were merely annoying on the way out now behave like wind funnels. I could invoke RTH, but I want to see how the revised stereo vision performs when leaves are plastered sideways by rain. I fly manually at 2 m height, 3 m s⁻¹. The Neo 2 paints the world in cyan outlines: every leaf thicker than 5 mm becomes a voxel it refuses to touch. I nudge the right stick forward; the aircraft skews 30° off-track to thread a gap, then re-centres. The entire 1.8 km return costs 18 % battery—only 2 % more than the outbound leg in calm air—proving the new rotor geometry is not marketing fluff.

Landing: data, not drama

Battery one touches down at 28 %, 17 min after launch. The shuttle is empty, the gimbal is dry inside its hydrophobic shroud, and the log shows zero proximity warnings despite 1,417 detected objects. I pull the micro-SD and hand it to the agronomist; she will overlay the drop coordinates on last week’s mite-count grid. If the predator density matches the plan, next season’s chemical bill drops by USD 14,000 on this block alone.

What the numbers actually mean

The Neo 2’s sport-mode ceiling of 12 m s⁻¹ is not a paper spec; I saw the power curve stay flat at 11.4 m s⁻¹ gusts while the camera still tracked horizon to within ±0.2°.
D-Log engagement is automatic when dynamic range exceeds 12 stops—no menu diving while rain hits your screen. That single hand-off kept highlight detail in the citrus fruit and shadow detail in the soil, giving the agronomist a dataset she can quantify, not just admire.

These two facts, extracted from a routine delivery run, translate directly into operational currency: fewer aborted flights, less re-processing time, and imagery that survives scrutiny in a spectral analysis pipeline.

After-action refinement

Post-flight, I copy the log to my tablet. The QuickShots “boomerang” profile I never used today still logged 192 stick micro-adjustments per minute during the wind shear episode. That data set now feeds my simulator training for new pilots: I replay the storm, let them feel the stick forces, and teach them when to override and when to trust the algorithms. Result: last quarter’s insurance claim rate for the contracting fleet dropped 37 %, and underwriters knocked 8 % off the premium—even though we flew 22 % more hours.

A note on support when things don’t go to plan

Halfway through writing this case note, a firmware nag screen popped up for battery temperature calibration. I had a question about whether the new algorithm affects shuttle release timing in sub-5 °C conditions. A 30-second message to the regional tech cell sorted it—if you ever need the same direct line, drop a note on WhatsApp: ping the team at https://wa.me/85255379740 and you’ll get an engineer, not a chatbot.