Field Notes
Technical writing on autonomous SAR, drone navigation, fleet orchestration, and the systems that find people.
Field-Ready Maps: Offline Basemaps and Cross-Border No-Fly-Zone Awareness
The map now works on the drone's no-internet WiFi: basemap tiles cache to disk and pre-load around you, an advisory no-fly-zone overlay draws official UAS geozones onto the map (France live; Netherlands, Germany, Denmark, Belgium and the UK as a fast follow), and the planner opens straight on your location. Built for the field, not the desk.
Read article →Maestro Flight Review Now Reads Your Whole Shift
Multi-file drop, per-sortie reports, plus a shift-level cross-sortie pattern analysis that surfaces issues no single .ulg would show — drone-2 GPS issues over the south face twice, drone-3 battery margin tightening across three sorties, the wind window all three drones flew through. Up to 100 files per session, sequential to respect your LLM rate limits, free, BYO LLM.
Read article →Predictive Maintenance for PX4 Fleets — Free, Local, No Cloud
Every flight you analyse in Maestro Flight Review gets recorded in a local fleet-history archive on your machine. After 5+ flights per airframe, Maestro surfaces predictive-maintenance trends (battery capacity decay, vibration creep, GPS antenna health) using linear regression with R² thresholding, plus >2σ anomaly detection against the airframe's own Welford-rolling baseline. Free, local-only, no subscription. The kind of thing manned aviation has done with engine-hours data for decades, applied to drone telemetry.
Read article →Getting an LLM API Key for Maestro Flight Review
Step-by-step for non-developer operators: how to get an Anthropic / Mistral / OpenAI / Google API key, how to pick between providers, what each model costs per flight (€0.001 to €0.05 depending on tier), and how to configure it in Maestro. Five minutes, no programming. Includes the EU-data-residency option (Mistral).
Read article →Introducing Maestro Flight Review: AI-Assisted PX4 Forensics, Free and Local
A free tool inside the Maestro Mac app. Drag a PX4 ULog in; the app parses it locally, plots eight time-series, overlays the flight on a 2D map, runs rules-based health checks, computes predictive-maintenance trends and anomaly detection against your own local fleet history, generates regulator-grade incident reports, and routes the structured summary to the LLM you've configured — Anthropic, Mistral, OpenAI, or Google. BYO LLM, local-first, free. No account, no metering.
Read article →How three Holybro X500 V2 kits in a Belgium workshop turned into A Drone Company. Read in order, or pick the one that matches your question.
Our First Fleet: How A Drone Company Started
Three flat-pack drone kits arrived at a Belgium workshop in early April. By the weekend they were doing manual flights; by the Monday they were crashing on purpose; by the Friday after we were writing what would become Maestro. An €1,780 workshop bench, three X500 V2 quadcopters, and the failures and surprises that shaped the autonomy stack — the internal test fleet Maestro is still validated against today.
Read article →Why We Built Maestro
Maestro is a complete ground station, purpose-built for single-drone SAR and 24/7 multi-drone SAR coverage, not a plugin to QGroundControl or a sidecar to Auterion Mission Control. The gap we saw in existing tooling, the 30-second test that drives every UI decision, why a complete stack was the right answer instead of a plugin, and why software-first was the right starting point.
Read article →Why We Bet on Open MAVLink
The architectural decision behind the company: bet on the open MAVLink protocol (PX4 / ArduPilot), and the autonomy layer — Maestro — becomes the durable product while the airframe stays swappable. The market gap that led here (closed DJI stacks below, €25k+ NDAA platforms above), the EU-sovereignty angle, and why software-first is the bet that survives airframe-replacement cycles.
Read article →Choosing a SAR Drone Platform: What Actually Matters
An honest buyer's guide for search-and-rescue drones. Thermal, weather rating, open SDK access, NDAA / Blue UAS compliance. Freefly Astro Max, Quantum Systems Trinity Pro, Acecore Noa, Skydio X10, and DJI M30T compared — all on airframes Maestro orchestrates.
Read article →Why Software Is the Hard Part of SAR Autonomy
Hardware is commoditising. The hard engineering in SAR autonomy is fleet orchestration, onboard detection, GPS-denied nav, exception alerting. The airframe is the compute substrate; the software is what saves lives.
Read article →Multi-Drone Fleet Relay: How It Actually Works
A deep technical dive. Pre-launch dispatch, waypoint-resume handoff, battery-aware scheduling, relay overlap without backwards teleports, equal-area sector partitioning, four-layer collision avoidance. MAVLink only.
Read article →From Testbench to Operational Deployment
Why we develop SAR on a €1,780 workshop X500 fleet and ship on industrial MAVLink airframes. The engineering discipline behind the test-platform / deployment-platform split.
Read article →The Hidden Procurement Cost of Proprietary Drone Stacks
The TCO math of closed vs open drone stacks. Skydio Dock, DJI ecosystem, MAVLink-based alternatives. What agencies on 5–10 year procurement horizons should weigh beyond unit price.
Read article →Altitude-Banded Separation: How Maestro Keeps a Fleet Apart
How Maestro separates a multi-drone fleet sharing one base — pre-allocated altitude bands by drone and sector, staged takeoffs, and a runtime proximity safety net. The mechanism behind one-operator multi-drone patrol. Validated in simulation, in field validation.
Read article →Sector Patrol: Covering Large AOIs with Parallel Sub-Fleets
Auto-recommended sectors based on fleet size, AOI size, and battery cycle math, prefers fewer sectors with continuous coverage over more sectors with gaps. Equal-area partitioning via binary-search bisection. Scales with your fleet.
Read article →The SAR Simulator
The full mission planner runs in a browser, fed by a physics model instead of live telemetry, same orchestrator, Maestro AI, and Safety Gate. Speed controls (1× / 10× / 50× / 100× / 200×) let it be a prospect showcase or an operator dry-run.
Read article →Air-Gapped Drone Operations
Maestro ships with a single-toggle air-gapped mode for classified, EU-data-residency, and contested-environment deployments. No external calls, persistent AIR-GAPPED badge, auditable offline operation.
Read article →Remote ID, Built In
EU and US Remote ID rules apply to every SAR mission. Operator ID and per-drone UAS serials are first-class settings in Maestro, surfaced as a pre-flight check, blocking launch until configured.
Read article →Configurable Detection Targets and GPS-Resilient Patrol
Maestro AI supports two SAR-focused detection targets, Person Search (visible) and Fire / Heat Source (thermal), with FOV-based detection simulation. Plus GPS dropout/recovery handling that keeps the relay running.
Read article →Meet Maestro AI
The decision layer inside Maestro: pre-flight health review, in-flight relay optimisation, post-flight debrief. Local physics engine first (offline, instant), optional cloud refinement, always gated by a Safety Gate.
Read article →Maestro AI: Intelligent Relay Orchestration
How Maestro AI uses a physics-based decision engine to dynamically optimise drone relay parameters, swap timing, patrol speed, fleet dispatch, in real time, with no internet required. MAVLink only.
Read article →24/7 Coastal Patrol with Drone Fleets
How a fleet of drones maintains continuous coastal coverage. Accurate fleet-size math for both continuous 24/7 and shift-based operation, relay mechanics, coverage gap calculation, and cost comparison to crewed patrol.
Read article →Fleet Patrol Mode in Maestro
Set Fleet Size to 2+ and Maestro activates its ground-station fleet layer: battery-aware handoff, waypoint resume, continuous patrol logic, and exception-only alerting across a multi-drone fleet. MAVLink-only. Validated in simulation, in field validation.
Read article →When Autonomous Missions Fail
Handoff failures, comms loss, detection misses, weather degradation. How the safety system handles every failure mode across single-drone SAR and fleet patrol modes.
Read article →Three Layers of Safety: How SAR Stops a Drone
Every SAR mission has three independent stops: the ground-station UI kill button, the PX4 onboard failsafes, and the RC transmitter kill switch. None depends on the others.
Read article →The Pre-Flight Health Panel
A persistent visible panel above the Launch button with eight automated health checks per drone plus an AI-assisted go/no-go recommendation. Ships in both modes, single-drone SAR and fleet patrol.
Read article →The 7-State Flight Supervisor
Inside the finite state machine that governs every SAR flight. Seven states from idle to emergency, with battery thresholds, geofence enforcement, VIO drift monitoring, and wind limits.
Read article →Building Onboard Human Detection for SAR Drones
Running edge inference on the drone's onboard companion computer. Why Maestro detection must happen onboard, how the pipeline works, and what confidence threshold actually means in the field.
Read article →Boustrophedon Search Patterns: The Math Behind Systematic Grid Search
How to compute sweep spacing from camera FOV and altitude, align sweep lines to polygon geometry, and convert grid points into relative flight vectors.
Read article →GPS-Denied Navigation: Why Maestro Drones Must Fly Without Satellites
Visual-inertial odometry, drift compensation, and hybrid navigation architecture. How to build a flight plan that survives GPS dropout in exactly the environments SAR operates.
Read article →Why Maestro Teams Need Autonomous Drones, Not Better Pilots
Manual flight creates coverage gaps. Video feed monitoring causes fatigue-induced misses. The case for autonomous grid search with onboard detection over skilled piloting.
Read article →