Autonomous Drone Swarm for Precision Farming: Fleet Control, Satellite Analytics & Field Ops

Client Request

Our own R&D product for agriculture: a swarm of scouting and spraying drones that works a farm as one system. Fields are analysed from Sentinel-2 satellite imagery and drone flights, problem zones get a diagnosis and a treatment plan, and the fleet picks up the resulting tasks on its own - a task starts as soon as an aircraft frees up, without a dispatcher. Everything lives on one 3D map: seven aircraft in the air, their onboard cameras, the operators cleared to fly them and the work queue.

Problem Solution

We built a browser-based ground control station on deck.gl and MapLibre over real satellite imagery: a live fleet map with 3D drone models and serpentine flight paths, a per-aircraft drawer with an onboard camera in RGB, NDVI and NDRE false colour, twelve data layers from Sentinel-2 and soil-lab samples rendered as extruded 3D zones, per-zone reports with fertiliser recommendations and savings, crop-suitability maps, a polygon tool to add new plots, a self-running task board and an operator directory with editable aircraft clearance. Ukrainian and English.

Final Result

The prototype turned a stack of agronomy formulas into something a farm manager can understand in a minute: which zones are stressed, what to apply and where, what it saves against blanket application, and which drone is already on its way. The aircraft link is simulated for now - telemetry, flights and the task queue are generated - while satellite and weather data are real. It is the base we use to talk to agri-tech clients about swarm control, field analytics and variable-rate application.

Project overview

The idea behind the product is simple: a farm should not need a person to route every drone. Fields tell you where the problems are, agronomy tells you what to do about them, and a fleet of aircraft should take it from there. We built the prototype in-house to prove the whole loop on one screen, from a satellite image to a drone picking up a task.

The field side starts with real data. For every plot the app pulls Sentinel-2 indices - NDVI, NDMI, NDRE, NDWI, EVI2, MSAVI, SBI - through the Copernicus Data Space and paints them onto 85-metre zones extruded in 3D over Esri satellite tiles. A drone fly-over reveals zones under the aircraft's footprint as it flies its serpentine path; a lab run adds pH, nitrogen, phosphorus, potassium and humus per zone. Click a zone and you get a report: index values with status, a diagnosis such as nitrogen deficit, the recommended dose for the crop, and the estimated saving against blanket application. A crop-suitability model scores winter wheat, corn, sunflower, soybean and rapeseed for every zone. New plots are added by outlining a polygon on the map.

The fleet side is the swarm. Seven aircraft of three types - sprayers, heavy sprayers and NDVI scouts - fly over three fields at once. The fleet map shows each aircraft as a 3D model on its route with the zones it has already covered; the drawer for any tail number adds the rotating model, an onboard camera with RGB, NDVI and NDRE modes, a top-down scan map and telemetry: battery, altitude, speed, hectares covered, time airborne and remaining. Unavailable aircraft show why - no link, maintenance, charging.

The decision layer is the task board and the crew directory. Recommendations become tasks - NDVI survey, crop-protection spraying, foliar feeding, scouting, a check flight after rain - with a priority and an area. The board runs itself: a queued task starts when an aircraft with the right clearance frees up and moves to Done on completion, and the shift total in hectares updates live. Operators are listed with licences, flight hours and an editable clearance matrix, so the swarm never assigns an aircraft to someone who is not allowed to fly it. The interface is in Ukrainian and English, and the whole thing runs in the browser with a small Node proxy for the satellite API.

Client: In-house R&D product, Lytvynov Production
Industry: Agri-tech, precision farming, drone operations
Scope: Ground control station in the browser - fleet map, onboard camera, satellite and lab analytics, recommendations, task automation, operator clearance
Stack: React, Vite, zustand, deck.gl, MapLibre GL, glTF, Node.js, Copernicus Sentinel-2 API, Open-Meteo, Esri World Imagery
Data: Real satellite indices and weather; drone telemetry and task flow simulated in the prototype
Languages: Ukrainian and English

Key features we implemented

Swarm on One Map

Seven aircraft over three fields as 3D models on their routes, with covered zones lighting up under each drone in real time.

Onboard Camera & Telemetry

Per-aircraft drawer with the rotating 3D model, live camera in RGB, NDVI and NDRE, scan map, battery, altitude, speed and time remaining.

Sentinel-2 Field Analytics in 3D

Seven satellite indices pulled from Copernicus and painted onto extruded 85 m zones over real imagery; drone and lab layers on top.

Zone Report & Soil Lab

Click a zone: indices with status, pH, N, P, K and humus, a diagnosis, the dose for the crop and the saving versus blanket application.

Where to Sow

Crop-suitability model scores wheat, corn, sunflower, soybean and rapeseed per zone by pH, nitrogen, moisture and vigour.

Add a Plot by Drawing

Outline a polygon on the satellite map, save it, and the new plot is zoned, analysed and added to the farm project.

Self-Running Task Board

Surveys, spraying, feeding and check flights queue by priority; a task starts the moment a cleared aircraft frees up and moves to Done by itself.

Operators & Aircraft Clearance

Pilots, scouts and technicians with licences and flight hours; an editable clearance matrix decides who may fly which aircraft.

Services and guides behind this project

Services AI proof of concept A short, fixed-price project that answers one question with evidence: can AI do this job on your data, accurately enough and cheaply enough to be worth building? View service Services AI agent development What a business AI agent really is, which tasks it handles well, how we design the tools, permissions and approval steps around it, and what building one costs. View service Services AI consulting A short assessment that tells you which AI use cases are worth building, what they will cost to build and run, and what to do first, written by the engineers who would then build it. View service Services AI product development The stages an AI product goes through between an idea and a product people pay for, what each stage must prove, and how we run them with our clients and for our own products. View service Services Node.js development We build Node.js back ends in TypeScript for products that are JavaScript end to end, need real-time updates, or stream AI answers to the browser, together with the React and React Native apps on top. View service Buyer's checklist How to choose an AI development company: a checklist for buyers How to choose an AI development company in 2026: a scorecard, questions on RAG, evaluation, data privacy and costs, and red flags to walk away from. Read guide

Telegram Book a call

Keep scrolling

Next project:
Flower Subscription Service: Telegram Ordering, Uklon Delivery Dispatch & CRM