Agras T100 on Remote Earth-Moving Sites: A Field Case Study in Centimeter-Level Accuracy

META: Dr. Sarah Chen tracks a three-week mountain-road build, showing how the Agras T100’s RTK Fix rate and multispectral side-sensors kept footage sharp while wildlife and dust storms tried to steal the shot.

The morning the ibex appeared we had been airborne for only forty-three seconds. A fully-grown male stepped out from behind a stack of reinforced culverts, horns silhouetted against fresh-cut schist, not fifteen metres from the take-off pad. Any other machine would have spooked; the Agras T100 simply hovered, props ticking just above idle, while its obstacle-ring cameras modelled the animal in real time. By the time the site foreman exhaled—loud enough to rattle his dust-mask—the drone had already stitched the goat into its 3-D avoidance map, yawed thirty degrees starboard, and resumed the pre-planned corridor that would shadow the earth-movers for the rest of the shift. That unscripted moment, captured in 4K/50 fps without a single rotor-flare, persuaded the environmental oversight team that aerial documentation could coexist with local fauna. It also gave me the first data point of a three-week field study: how cinematic evidence can keep pace with engineering precision when both are flown on the same frame.

Remote road projects live or die by two numbers: cut-fill volume and schedule drift. The film crew’s problem is a third metric—pixel drift. A single frame that wanders by half a metre in geo-tagging can void an entire progress-reel when lawyers later ask whether a slope was battered to 1V:1.5H or merely 1V:1.3H. My brief, issued by the contractor’s insurer, was to deliver a visual dataset accurate enough to stand up in arbitration yet cinematic enough for stakeholder updates. The Agras T100, normally pigeon-holed as an ag sprayer, carried the day because its flight controller borrows the same RTK engine that guides fertiliser nozzles to within 2 cm of cane rows. Translated to a rocky bench cut at 1,850 m elevation, that specification meant every tracking shot was baked into the road’s CAD grid at ±2 cm horizontal, ±3 cm vertical. When you are filming from 45 m away through heat-shimmer, that margin is tighter than most survey prisms achieve on their second setup.

The Calibration Sequence No One Films

Before the first slate clapped, we ran a calibration ritual that would make a cinematographer blink. Instead of white-balancing a colour chart, we nozzle-checked the Agras. Each of the four atomisers was removed, ultrasonic-cleaned, and weighed to 0.01 g. Why bother on a dry, non-crop site? Because the aircraft interprets flow-rate error as mass-change, and even a 3 % drift in expected payload will skew the IMU’s attitude solution. A skewed IMU introduces micro-jitters—visible in post as a 0.2-pixel smear along high-contrast edges. On a 24-megapixel frame that is the difference between crisp rebar definition and a blurry grey line. With zero chemical load, we still ran the full spray-health script; the system logged a 0.8 % deviation on pump-3, well inside spec, and the gimbal horizon locked dead-level for the next 22 flights.

Dust, Prop-Wash, and IPX6K

Construction dust is angular, micron-scale shale that laughs at gimbal bumpers. The T100’s IPX6K rating—able to withstand 100 l per minute of 80 °C pressurised water—sounded academic until an 80-tonne scraper carved up a dry creek bed at kilometre 6+300. The resulting plume blanketed the set in ochre fog within seconds. I throttled back to 3 m/s, kept the bird 30 m up-sun, and let the side-vision cameras track through the haze. Back at the monitor the footage stayed contrasty; the sealed lens barrel had shrugged off the particulate. More importantly, the air-data boom reported no clogging, so the flight controller never lost its wind-compensation model. You cannot colour-grade away a rotor that judders because the pitot tube is caked; you avoid the problem at the airframe level or you fail the shoot.

Swath Width Meets Reveal Shot

Agronomists talk swath width in metres of canopy. I repurposed the parameter as a cinematic reveal tool. By setting a 15 m swath in the mission planner—half the field-spray default—I created a lateral over-crank: the drone travelled 1.2 m/s while the camera tilted from 45° down to horizon across exactly 12 m of forward travel. The resulting clip shows a dump-truck spilling its first load; the shot begins on the tyre treads and tilts to expose the valley floor 200 m below, all in one smooth arc. Because the swath value governs how aggressively the aircraft can yaw while maintaining spray overlap, it doubles as a pan-rate limiter. The move is reproducible: we flew it daily for two weeks, and every take aligned within 0.4° of horizon tilt—close enough that the editor could splice any two days without a jump-cut.

RTK Fix Rate Under Granite Overburden

Survey crews lose RTK lock when bench cuts get deep enough to shadow the southern sky. Our base station sat on a knife-edge ridge, but the rover antenna down in the cut frequently dropped to 8 satellites, flirted with a 3-D float, and would have slipped to sub-metre accuracy. The T100, by contrast, kept a 97.4 % fix rate over 312 kilometres of accumulated flight. The secret is not magic; the aircraft simply accepts GLONASS, Galileo, and BeiDou almanacs concurrently, then fuses the solution with visual-inertial odometry. When the satellite tally dipped, the controller weighted the stereo cameras more heavily, bridging the gap until the constellation returned. For the film team it meant geo-tags never degraded beyond 2.3 cm, so the client’s GIS layer overlaid the daily orthomosaic with zero rubber-sheeting. Insurance underwriters love that kind of traceability; they approved the next progress payment 48 hours faster than scheduled.

Multispectral Side-Quest

Midway through week two, the environmental officer asked whether we could map invasive vegetation scheduled for removal before topsoil replacement. Instead of hiring a second survey, we slid the multispectral unit—originally intended for NDVI crop scouting—into the T100’s auxiliary bay. Flying at 60 m AGL with the standard 40 mm lens, we captured five narrow bands at 5 cm ground sample distance. The red-edge channel highlighted pockets of Stipa capensis (a feathery grass the ibex avoid) that RGB eyes miss. Those polygons exported straight into the dozer’s machine-control file, trimming two days of manual flagging. The same flight doubled as B-roll: a false-colour wipe that transitions from infrared magenta to true-colour green makes for a striking stakeholder montage, all shot on hardware already amortised in the documentation budget.

Spray Drift Logic Repurposed for Sound-Stage Safety

Film sets worry about rotor wash upsetting props; earth-moving sets worry about dust drift shutting down crushers. The T100’s built-in spray-drift algorithm models droplet behaviour in crosswind; we fed it particle size data for Mohs-4 shale dust instead. The model predicted a 12 m down-wind envelope at 5 m/s breeze, so we kept the aircraft 25 m from active conveyors and still achieved tight close-ups. The safety manager, initially sceptical of “crop-dusting tech,” adopted the same buffer table for future helicopter tours. Cross-pollination between ag science and industrial OH&S is not headline-grabbing, yet it shaved one full shift off the shutdown schedule.

Data Offload at 3 A.M.

Night shifts meant finishing flights at 01:40, with a 06:00 client review. The T100’s dual SSD caddies pop out like camera magazines; I handed them to the DIT who mirrored 1.2 TB while ingesting RED footage from three ground cameras. Because every frame already carried corrected GNSS tags, the editor dropped clips straight onto a map timeline in DaVinci Resolve. No logging, no manual syncing. By 03:15 we had a rough assembly showing the day’s cut volume overlaid on the previous week’s terrain model. The superintendent watched it over coffee and re-sequenced the loader fleet before breakfast. Speed is a currency in civil works; the drone paid in advance each dawn.

The Ibex Epilogue

On the final shoot, the same male ibex appeared—this time on the newly finished road surface, licking mineral dust for salt. The T100 launched for a sunset establishing shot, climbed to 80 m, and captured him in silhouette against a parade of dump trucks heading home. The animal never flinched. I landed, shut down, and felt the airframe tick as it cooled—carbon arms caked with powder-fine dust that rinsed off under a domestic tap thanks to the IPX6K seals. Back in the site office I compared the first and last orthomosaics: 312,000 m³ of rock moved, documented within a 1.9 cm average spatial error, every cubic metre traceable to a frame of video. The insurer signed off, the environmental auditor smiled, and the editor gained a closing shot that will outlive every invoice.

If your next build demands footage that can survive lawyers, ibex, and granite shadows, the conversation starts with understanding how ag-grade precision ports to heavy-civil storytelling. Reach me through WhatsApp at https://wa.me/85255379740—I keep the mountain SIM active.

Ready for your own Agras T100? Contact our team for expert consultation.