MONTHS 1–2
Establish the baselineInventory the code and skills. Assign owners. Define interfaces and release criteria. Set up reproducible builds, simulation and logging.
Gate 1A fresh checkout runs a simulated mission and records usable telemetry.
START HERE · SHARED FOUNDATION
Product & mission contextAgree on aircraft type, operating environment, operator needs, interfaces and measurable acceptance targets.
Core engineeringGit, code review, debugging, tests, networking, Linux basics, units, coordinate frames and timestamps.
One reference systemSelect the company's autopilot and supported versions. Establish a reproducible simulation, telemetry connection and shared flight-log format.
A proposed baseline for a mixed UAV software department. These are responsibility areas, not six required hires or six courses for every engineer. Verification and architecture begin with the first sprint.
1. Flight Software & Vehicle Integration
OWNER · Flight software lead
C/C++ · PX4 OR ArduPilot · MAVLink · MAVSDK
- Understand the aircraftFlight modes, sensors, actuators, arming and onboard failsafes.
- Build and run one autopilotPin a supported release; reproduce the build and run software-in-the-loop (SITL).
- Integrate vehicle communicationHeartbeats, commands, acknowledgements, parameters and mission transfer.
- Master timing and framesBody/world frames, NED/ENU, altitude references, timestamps and sensor synchronization.
- Validate interfaces and failure behaviorBench-test payloads; test link loss, invalid inputs and recovery in simulation.
DELIVERABLEA versioned vehicle integration package with repeatable simulation and bench tests.
Exit criterion: Command outcomes are logged; timeout, link-loss and recovery cases pass the agreed test matrix.
Official learning resources
2. Ground Control & Mission Planning
OWNER · Ground systems lead
C#/.NET + WPF OR C++ + Qt · GIS · MAVLink
- Build reliable application foundationsAsync programming, UI state, cancellation, connection handling and unit tests.
- Understand maps and coordinatesWGS84, projections, terrain, offline maps and explicit altitude references.
- Implement mission workflowsCreate, validate, upload, download and compare waypoint and survey missions.
- Make operations understandableFreshness indicators, clear command status, actionable alarms and telemetry replay.
- Test operator journeysExercise reconnects, rejected commands, partial uploads and interrupted sessions.
DELIVERABLEAn operator can plan, validate and execute a simulated survey, then replay its telemetry.
Exit criterion: Mission round-trip preserves required fields; stale data is visible; interrupted transfers recover predictably.
Official learning resources
3. Autonomy, Perception & Edge AI
OWNER · Autonomy / perception lead
Python + C++ · ROS 2 · OpenCV / PyTorch · ONNX / TensorRT
- Establish the robotics foundationLinear algebra, calibration, transforms, state estimation and ROS 2 integration.
- Build a trustworthy datasetRepresentative camera data, labels, provenance and splits by flight or site.
- Implement a bounded capabilityStart with inspection detection or landing-area assessment; record failure cases.
- Connect perception to autonomyAdd localization, planning and control only as the use case requires; define fallback behavior.
- Benchmark on the target computerMeasure end-to-end latency, accuracy, memory, power and sustained thermal behavior.
DELIVERABLEA bounded onboard perception capability with a reproducible dataset and deployment benchmark.
Exit criterion: Agreed accuracy and latency limits pass on held-out flights and target hardware; failures trigger defined fallbacks.
Official learning resources
4. Backend, Data & Fleet Operations
OWNER · Backend / platform lead
REST APIs · SQL · Event messaging · Observability · n8n
- Define the data contractsVehicle identity, units, timestamps, schemas, access roles and retention.
- Build resilient ingestionBuffer offline data; handle retries, duplicates, ordering and reconnects.
- Deliver operational servicesFlight records, fleet status, maintenance history, audit trails and data export.
- Automate supporting workflowsUse n8n for reports, notifications and business integrations after reliable ingestion.
- Operate and recover the serviceMonitor latency and errors; test backups, restores and access revocation.
DELIVERABLEA fleet data service with offline recovery and an automated post-flight report.
Exit criterion: Replay causes no duplicate business records; recovery meets agreed targets; access and audit tests pass.
Official learning resources
5. Simulation, Verification & Release Quality
OWNER · Verification / QA lead
SITL · Hardware-in-the-loop (HITL) · CI · Log replay
- Make requirements testableDefine acceptance criteria, interface contracts and representative mission scenarios.
- Automate checks from the first sprintRun build, unit, protocol and simulation tests for relevant changes.
- Exercise abnormal conditionsInject packet loss, stale sensors, restarts and malformed inputs in controlled tests.
- Validate hardware and field behaviorProgress from simulation to bench/HITL and approved flight tests with recorded evidence.
- Release reproduciblyArchive software, parameters, hardware versions, results and rollback instructions.
DELIVERABLEA repeatable release qualification process linked to requirements and recorded test evidence.
Exit criterion: Required tests pass; blocking defects are closed; responsible engineering and flight-test owners sign off.
Official learning resources
6. Architecture, Security & Team Delivery
OWNER · Software department head / technical lead
Git · Code review · CI/CD · Issue tracking · One AI coding assistant
- Define product scope and ownershipSet operating conditions, product requirements, subsystem boundaries and named owners.
- Create engineering agreementsVersion interfaces; record design decisions; standardize review and release practices.
- Protect the development and product systemsManage secrets, roles, dependencies, update integrity and incident response.
- Use AI with engineering reviewUse one assistant for code, tests and documentation; review outputs and protect company data.
- Build a sustainable teamMaintain a skills matrix, mentoring, onboarding, capacity planning and defect reviews.
DELIVERABLEA department operating model with clear ownership, a shared backlog and measurable release readiness.
Exit criterion: Each subsystem has an accountable owner; builds are reproducible; review and access rules are enforced.
Official learning resources
Suggested 12-month delivery sequence
Planning ranges, not promised dates. Adjust for current capability, staffing, hardware availability and product approval requirements. Advance on evidence, not the calendar alone.
MONTHS 3–4
Integrate the core productConnect mission planning, vehicle commands and backend ingestion. Add protocol, reconnect and mission-transfer tests.
Gate 2Plan → simulate → collect → replay works end to end, including interrupted connections.
MONTHS 5–8
Validate on target hardwareComplete bench/HITL tests and controlled flight validation. If needed, add one bounded AI capability after its dataset and benchmarks are ready.
Gate 3Agreed hardware performance and failure-recovery criteria pass with recorded evidence.
MONTHS 9–12
Prepare repeatable deliveryQualify releases, rehearse rollback and service recovery, complete operator documentation and close field-test defects.
Gate 4A versioned release has traceable evidence, responsible approval and a support handover.