Mission Templates and Event-Driven Handover for Persistent Multi-Drone Inspection and Mapping

Persistent drone operations require more than isolated autonomous flights. Inspection, mapping, and emergency-response missions must remain active as batteries deplete, vehicles become unavailable, tasks are interrupted, and priorities change. This work extends Swarm-Steward, a natural-language multi-robot coordination system using an LLM-driven multi-agent architecture, toward persistent autonomy with limited human intervention. Operators specify goals, while the system grounds requests in geospatial context, telemetry, and aerial perception, then executes them through deterministic tools and priority-based task queues.

The main contribution combines reusable mission templates, event monitoring, and queue-aware handover. Mission templates define workflows for industrial asset inspection, aerial mapping, stockpile monitoring, hotspot assessment, search and rescue, and follow-up inspection of detected events. These workflows combine take-off, lawn-mower coverage, nadir and oblique camera angles, 3D feature orbits, adaptive gimbal control, and safety-validated trajectories. Mission targets can be known map objects, manually defined locations or areas, or dynamic features produced from drone video streams through detection, geolocation, and cross-drone fusion. This allows the same instruction to address planned, operator-defined, and unexpected targets.

Event-driven handover enables Swarm-Steward to monitor progress, telemetry, queued actions, and priorities, and to react when a drone reaches a battery threshold, becomes unavailable, is interrupted, or a higher-priority event appears. Unfinished actions and pending queues can be inherited by another drone or replacement group while preserving target references, task order, priorities, and execution history. The system has been tested with real drones and integrated with an Unreal Engine-based simulator for digital-twin-style visualization, mission preview, and operator feedback. It is being developed for biomass power-plant scenarios, where imagery of large fuel stockpiles supports mapping, volume estimation, and hotspot inspection. Humans remain responsible for goals, confirmation, feedback, and supervision, while mission templates and event-driven handover allow missions to continue beyond individual drone endurance.