Beskriv uppdraget med ord, utför det.

PilotFlow lets a user create map objects, describe a drone task in ordinary language, review the generated mission, simulate it, and approve or reject it before execution.

Explore features See future work

PilotFlow map interface with 3D buildings

Core functions in the app

Named map objects

Create points, routes, and areas that become command references.

Natural-language planning

Use typed or spoken commands instead of navigating complex mission-planning software.

Mission review

Inspect the interpreted plan before any real execution can start.

3D height simulation

Visualize altitude changes as vertical or diagonal route segments.

Building collision warnings

Use LiDAR-derived building heights to flag possible route conflicts.

Vision concepts

Search for people, estimate crowd sizes, and review visual observations.

Map-first interface with compact controls

The interface keeps the map central, then brings in planning, simulation, and safety feedback only when needed.

Simulation with height path — Simulation

3D buildings and collision warning — 3D warnings

Från kommando till godkänd mission

PilotFlow omvandlar ett skrivet eller talat kommando till en strukturerad mission. Varje steg kan granskas, valideras och simuleras innan användaren godkänner flygningen.

Command

Intent

Mission DSL

Validate

Simulate

Approve

Outdoor field test setup with drone and laptop

Validated in real-world test conditions

PilotFlow has been evaluated through command tests, user feedback, and controlled outdoor flights with a DJI Mini 3. These trials connect the planning experience to real drone operation and guide the next stage of development.

Future work

Focused development priorities that expand PilotFlow's capabilities and operational readiness.

More reliable interpretation

Evaluate stronger models, API-hosted models, and schema-constrained generation.

Local place-name recognition

Extend speech recognition beyond saved object names to understand local place names.

Live weather planning

Integrate SMHI weather data so routes can account for live wind, precipitation, and visibility conditions.

Footprint-aware routing

Generate route alternatives that avoid building footprints instead of only warning after a conflict is found.

Terrain-aware collision checks

Combine LiDAR buildings with terrain elevation, trees, wires, and temporary obstacles.

Robust real-flight layer

Improve DJI execution, return-to-start behavior, logs, telemetry, and manual takeover.