Overwatch ships with a full in-browser Simulator that runs the same orchestrator, Overwatch AI, and Safety Gate the live ground station uses — fed by a physics model instead of live MAVLink telemetry. It serves two operational purposes: a prospect showcase in a few minutes, or a real-time dry-run rehearsal for operators preparing a live mission.
One runtime, two jobs
The Simulator compresses or stretches a scenario to fit the operator's intent. Same code path as a live deployment, same decision logic, same state machine — the only variable is how fast the sim clock runs relative to wall-clock time.
Showcase. A prospect loads demo.adrone.company and watches a 24-hour Curacao fleet patrol play out in roughly 14 minutes. Every relay handoff, GPS dropout, weather abort, and detection alert fires as it would in a live deployment, just time-compressed.
Dry run. An operator configures tomorrow's actual patrol — route, fleet size, battery type, schedule — and watches it play out in real time before the drone leaves the ground. The airspace-deconfliction community calls this a dry run; the PX4 ecosystem calls it SITL. The principle is the same: a rehearsal at real pacing, useful for catching misconfigurations that only become visible in motion.
Same Simulator, same runtime — the speed selector decides which job it's doing right now.
Speed controls
The Simulator's top bar carries a five-position speed selector beside the mode toggle:
- 1× — real time, 1:1 with drone flight. An 8-hour patrol takes 8 hours. Dry-run mode: configure your patrol, start it, let the laptop run in the background while you work through a pre-flight checklist and watch relay cycles trigger at the real cadence.
- 10× — fast-forward rehearsal. 8 hours in 48 minutes. Pattern matters more than pacing; you want to see three or four full relay cycles without waiting.
- 50× — mid-length preview. 8 hours in ~10 minutes. Good for tabletop reviews with a team gathered around a laptop.
- 100× — compressed showcase (default). 24 hours in ~14 minutes.
- 200× — extra-fast scan. 24 hours in 7 minutes. Useful for reviewing a multi-day patrol when you just want to see the pattern hold over many cycles.
Speed maps directly to the orchestrator's speedFactor — every 200 ms real-tick advances simulated time by that factor. Detection events, handoff timing, battery drain, and GPS state machines all compress proportionally. At 1× the drone moves across the map at 6 m/s real-world. At 200× it sweeps a 20 km patrol route in about a minute.
How it's wired
When the Simulator is active, the drone uploader switches to a physics-backed stub and the orchestrator tick loop reads position, battery, and GPS state from the simulator instead of a live MAVLink bridge. Everything downstream — the Overwatch AI's 30-second decision cycle, Safety Gate enforcement, detection pipeline, alert engine, patrol scheduler, external notifications — runs untouched. A simulated patrol produces the same decisions, alerts, and audit logs as a real one; only the telemetry source differs.
That's intentional. It's what makes the Simulator useful as a rehearsal tool rather than a toy. If a weather-triggered abort fires in simulation, the same code path would have fired in a real flight. If the Overwatch AI recommends reducing swap threshold from 20% to 18% at 7 m/s wind, that's the recommendation it would have made against live drones. There is no separate "simulator logic" anywhere in the stack.
Try it
No install, no login. When the page loads, a three-drone fleet patrols the Curacao coastline at 100× by default, so the 24-hour scenario plays out in a few minutes. A few things to try:
- Click 1× to watch a relay handoff at real pacing. The standby drone pre-launches and transits to the handoff waypoint in real time — instructive for understanding why the default swap threshold is 20%.
- Change Number of Drones to 1. The UI shifts to Single-Drone SAR Mode; draw a polygon anywhere on the map, click Run Demo, and watch the boustrophedon grid search play out.
- Draw your own patrol route near your home coastline. Set fleet size, pick a battery type, click Run Demo. The Simulator anchors on your drawn shape rather than panning to Curacao.
- Set a 2-hour Patrol Schedule and run at 10× — a 12-minute rehearsal that includes a couple of complete relay cycles.
The Overwatch AI (LOCAL / CLOUD source badge in the top-right of the fleet panel) is on by default, so every recommendation is logged with a full audit trail. In the public browser Simulator, only the local decision engine runs. In the Electron desktop app with API keys configured, the cloud-refinement path (Anthropic → OpenAI → Gemini) activates when connectivity exists.
What's next
Two expansion points on the backlog, both building on the same runtime:
Mission replay. Record-and-replay of real flights, not simulated ones. Same UI, same controls, but drone positions come from a MAVLink log rather than the physics model. Useful for debriefs, training new operators, and documenting incidents.
Training mode. Record a simulator session as a structured scenario — fleet size, route, weather, scripted failures — and replay it for operator drill. A supervisor could script a specific failure (handoff miss, wind exceedance, GPS loss) and have the trainee respond repeatedly until the response is automatic.
For now: one Simulator, two jobs, five speeds. Try it.