Astra Robotics is the crop-intelligence layer for the working farm. Our autonomous drones read what a crop is saying in near-infrared — days before it looks sick to the eye — and turn it into a same-day prescription. Not pretty maps. Dollars saved per acre.
Scouting that produces a pretty map is a toy. Astra closes the loop: the same flight that finds the problem generates the variable-rate prescription that fixes it — and verifies the result. Six steps, one system, zero blind spraying.
Autonomous multispectral flight covers every acre on RTK GPS.
On-board AI reads NIR & red-edge stress invisible to the eye.
Models classify disease, pest, and nutrient zones with confidence.
A variable-rate map: treat the 12 acres that need it, skip the rest.
The execution drone applies the prescription, only where mapped.
A follow-up flight confirms the outbreak stopped. Loop closed.
Plants broadcast stress, disease, and nutrient deficiency through signals humans can't see — most of them in the light a leaf reflects. Our sensors read the near-infrared and red-edge bands where those signals live, surfacing trouble days before a visible symptom appears.
A human scout walks a fraction of a field. We image every square meter, every flight.
On-board inference means insight in the field — not a week later from the cloud.
Where satellites are coarse and cloud-blocked, we fly under the clouds at leaf resolution.
We don't bolt AI onto a drone. Our perception stack is grounded in the frontier of geospatial machine learning — self-supervised foundation models, vision transformers, and on-device inference — specialized for the exact crops and diseases of the Northern Plains.
We adapt self-supervised geospatial foundation models — the Prithvi-EO and masked-autoencoder lineage — to our crops, so accurate perception needs a fraction of the labeled data a from-scratch model demands.
NMS-free, quantization-friendly detectors in the YOLO-26 generation resolve fine lesion detail and long-range field context together — the architecture behind today's highest-accuracy UAV crop-disease results.
Distilled, INT8-quantized networks run on an on-board NVIDIA Jetson Orin — perception happens in flight, in fields with no connectivity, with explainable disease-activation maps for the agronomist.
RGB, near-infrared, and red-edge bands feed NDVI and NDRE indices — quantifying chlorophyll, biomass, and stress per square meter, geolocated to sub-inch precision by RTK.
Our roadmap applies reinforcement learning — the founder's published domain — to adaptive scouting: aircraft that re-plan their own paths based on what they're seeing in real time.
Region-specific imagery accumulates into a proprietary dataset that compounds — a moat no reseller can copy and no satellite can match. The more we fly, the better we see.
We're building the full software pipeline in-house — owning every layer from the sensor to the spray map. This is the compounding asset: the airframe is a component, but the intelligence is ours, and it gets sharper with every acre we fly.
Fine-tuned geospatial foundation models and transformer detectors turn multispectral frames into classified disease, pest, and nutrient zones with confidence and severity.
PyTorch → ONNX → TensorRT, INT8/FP16 quantized and distilled, compiled to a Jetson Orin engine — 4–8× smaller, real-time, fully offline in the field.
Indices and model outputs become variable-rate prescriptions, exported as shapefile / GeoJSON that load straight into the sprayer and the grower's equipment.
Every operation's orthomosaics, analytics, scouting reports, and prescriptions in one cloud workspace — the sticky, recurring layer growers log into.
Mission planning, fleet health, and — on the roadmap — reinforcement-learning adaptive scouting, where aircraft re-plan their own paths from what they see.
Our fleet is built on the DJI Agras T100 — a world-class flight platform we'd be foolish to reinvent. So we don't. Using DJI's open Payload SDK, we strip one bird of its spray kit and rebuild its payload bay into our own: a gimballed multispectral pod, Jetson Orin compute, and our autonomy software — without ever touching DJI's locked, best-in-class flight controller and safety stack.
The official, MIT-licensed path to mount custom payloads. Through the SkyPort / E-Port adapters, our pod and compute speak directly to the aircraft.
We subscribe to the aircraft's LiDAR and millimeter-wave point cloud and push real-time AI recognition overlays back to the ground station.
Spray module removed; a radiometrically-calibrated RGB + NIR + red-edge array, Jetson Orin, and Starlink/LTE comms fabricated into a single gimballed pod.
A ground-up PX4 / ROS 2 platform with our compute and sensing integrated — regulation-proof and supply-chain-secure against foreign-drone restrictions.
The traditional path is scout by foot, then blanket-spray everything. We scout every acre for under five dollars and our map means a grower treats only where there's a problem. Same outcome, lower spend, less chemical in the water.
~$4.75/acre to scout delivers $15–$25/acre of value through chemical savings and yield protection — a missed Cercospora or blight outbreak can cost $100–$300/acre.
Targeted, variable-rate treatment cuts fungicide and pesticide volume on flagged fields versus a full blanket pass — protecting margin and the Red River watershed alike.
$39,553 in revenue across 7,500 acres — profitable in the very first field month. Disciplined growth, reinvested straight back into the fleet.
Next month's flights are scouted, scoped, and signed — roughly four times our first season's footprint, anchored by a 12,000-acre potato program and spanning eight crops and eight distinct service lines. These are committed routes we're now staffing and scheduling: the work is mapped, not yet flown.
| Operation & crop | Acres | Service we were hired for | The agronomic problem |
|---|---|---|---|
| R.D. Offutt FarmsPOTATOES · IRRIGATED | 12,000 | Late-blight surveillance, variable-rate nitrogen & vine-desiccation timing | Late blight (Phytophthora infestans) can take a field in days; early multispectral detection drives a targeted fungicide pass before it spreads, and our maps time the pre-harvest vine kill. |
| Minn-Dak Beet GrowersSUGAR BEETS | 4,500 | Cercospora leaf-spot early detection & stand counts | Cercospora is the costliest beet disease in the Valley; catching it before canopy-close cuts the number of fungicide passes a grower has to run all season. |
| Bremer Grain Co.FIELD CORN | 4,000 | Variable-rate nitrogen + tar-spot disease scouting | Tar spot is spreading fast across the upper Midwest; combined with zone-based N prescriptions, we protect both yield and the nitrogen budget. |
| Red River Soy PartnersSOYBEANS | 3,500 | Iron-deficiency-chlorosis mapping, white-mold risk & waterhemp weed mapping | High-pH Valley soils trigger IDC in patches; herbicide-resistant waterhemp demands targeted, mapped control instead of a blanket spray. |
| Northern Durum CollectiveSPRING & DURUM WHEAT | 3,000 | Fusarium head-blight (scab) fungicide timing & lodging assessment | Scab slashes grain quality and raises DON; the fungicide window is roughly three days at flowering, so same-day intelligence is the whole job. |
| Prairie Oilseed GroupCANOLA | 2,000 | Sclerotinia bloom-stage scouting & flea-beetle pressure | The sclerotinia spray decision hinges on bloom stage and canopy moisture; we quantify both, and flag flea-beetle pressure during establishment. |
| Sunrise Sunflower Co.SUNFLOWERS | 1,500 | Stand establishment + rust & downy-mildew scouting | Rust can defoliate late in the season; early stand and disease maps guide a single, well-placed protective pass. |
| Dahl Bros. FarmingDRY EDIBLE BEANS · RETURNING | 1,500 | White-mold scouting + desiccation timing | A returning client expanding scope from scouting into harvest-timing — white mold thrives in dense canopies, and we map the desiccation window. |
| Eight operations · eight crops | 32,000 | Scouted & scheduled for the coming cycle — execution to follow | |
Eight crops and eight problems, one method. Each program runs the same closed loop — tuned to its specific threat and its specific clock.
A multispectral baseline flight maps each field before any treatment decision — the same scout-to-prescription loop, run eight different ways across the portfolio.
We load crop-specific perception — late blight vs. Cercospora vs. Fusarium vs. iron chlorosis — so detection is tuned to each field's real threat, never a generic average.
Late blight, sclerotinia and head-blight carry roughly three-day decision windows. Our same-day turnaround from flight to prescription is precisely why these growers signed.
Added aircraft, battery sets, and a seasonal pilot-agronomist — plus the BVLOS-readiness to cover more ground per sortie — sized to the schedule before the season opens.
These 32,000 acres are scouted, scoped, and scheduled commitments for the coming cycle — not yet flown. Crop programs, service lines, and the agronomic risks named above are the planned scope; actual coverage, findings, and outcomes will be reported after execution. Anchor potato acreage reflects a single large irrigated operation.
Land with low-friction scouting, prove the dollars, then own the whole field — expanding geography, crops, software, and capability as the regulation and the fleet scale.
Seven operations, 7,500 acres, six crops across Minnesota and North Dakota — the revenue engine and the data flywheel, live today.
AstraSense, AstraEdge, AstraScript, and AstraView shipping in parallel with variable-rate execution under FAA Part 137 and state applicator licensing.
The FAA's Part 108 BVLOS framework — agriculture a defined category — lets one operation cover a whole county. We're building detect-and-avoid readiness now to be first through the door.
A ground-up PX4 / ROS 2 platform with our compute and sensing integrated — regulation-proof, supply-chain-secure, and entirely ours.
Tell us your fields and your crop. We'll scout a block, find what your current tools miss, and hand you the number — the dollars-saved-per-acre that change how you farm.
