How to run the demonstration and read the C2 console.
The demo is a real software-in-the-loop (SITL) simulation of the Visentrix autonomous drone swarm, flying under GPS and RF denial — the conditions the system is built for. You launch a scenario, and a high-fidelity simulation with live camera-based perception runs on our GPU server. You follow the flying drones in real time in the C2 console (the operator's common operating picture). Nothing here touches a real aircraft; it is the same software stack that runs on the drones.
Each scenario highlights one capability. Every one is a real Software-in-the-Loop flight of the actual Visentrix stack — the same code that runs on the aircraft, driven by the same automated regression tests. Here is what each does and what to watch for.
The entire swarm is waiting for you on the ground — this one is a sandbox, not a script, and it starts where a real sortie starts. You define the swarm (participants, formation, spacing) and you give the launch order; the fleet then takes off and stays up for the rest of the session. From there you drive it: draw a recon area on the map, load or build a mission, plan a road-following route, or type an order in plain language. The fleet carries out whatever you ask, as long as you want to keep exploring.
Watch for: the fleet sitting on the ground at the start — that is deliberate. Set the formation, then send the launch order and watch every drone arm, climb and take up its slot. After that, use the tasking panel (draw area / load mission / plan route / formation) or the command bar (plain language) and watch the swarm execute your orders live against the contacts on the map.
The drone works out where it is without any satellite navigation, by matching what its camera sees against known map features.
Watch for: the position estimate tracking the true flight path on the map — a live fix produced from imagery alone, with GNSS switched off.
The drone follows a road purely from what it sees, holding the route without GPS — and independent of altitude.
Watch for: the aircraft staying locked to the road on the map while navigating from vision only.
The swarm is given an area and covers it autonomously in an efficient search pattern, detecting and tracking what it finds.
Watch for: the coverage sweep across the assigned sector and detected objects appearing as tracks on the map.
The system turns what the drones observe into a written situation report, using an onboard AI vision-language model — a plain-language summary rather than raw video.
Watch for: the SITREP panel filling with an AI-generated summary of the scene.
You task the swarm in plain language: type an order such as “search the northern half of the area for vehicles” into the command bar. An onboard AI turns it into a concrete flight plan the swarm then executes. Sovereign — the language model runs on our own hardware, nothing leaves for a cloud service.
Watch for: your sentence being turned into a step-by-step plan, then the drones flying it.
The drone tags a moving target and pursues it with proportional-navigation terminal guidance (the same law used for intercept). The guidance solution is computed live and the aircraft closes the intercept geometry; the effector stays ROE-locked to the human (track, designate and guide only — no engagement).
Watch for: the drone taking off and translating to chase the moving contact, closing the range while the effector stays human-gated.
GPS is denied by electronic warfare. The APNT (Assured PNT) arbiter scores every navigation source and fails the swarm over from satellite to vision/UWB (GeoLocator → VIO → OSM-localisation) so it keeps navigating. Detected jammers and their comms-loss zones are drawn on the map.
Watch for: the amber “APNT ACTIVE — GNSS denied” banner naming the surviving nav source, the magenta jammer markers + interference rings, and the aircraft holding its track without GPS.
Given an area with several targets, the drones split the area between themselves (an autonomous sector auction — no central micro-management), and a per-target auction then assigns each individual target to the best-placed drone — marginal-utility bidding on range, capability and current load, so the closest capable drone wins and the work spreads with no drone overloaded.
Watch for: the per-drone sectors, each drone sweeping its own patch, and the drone→target assignment links — drawn from the real auction result, not a script.
The elected master drone is lost mid-mission (its coordinator is killed outright). The swarm re-elects a new leader and the deputy resumes command — the mission never stops and never needs you to intervene.
Watch for: the COMMAND indicator in the header switching to the new master drone, while the fleet keeps flying.
The operator link is deliberately cut mid-mission. Instead of aborting or turning back, the swarm keeps carrying out the mission on its own — the core Visentrix differentiator for contested, jammed environments.
Watch for: the console signalling the loss of the command link while the drones continue their task, then re-synchronising when contact returns.
Three drones are sent to separate areas beyond mutual radio range. Out of contact with each other, each flies its own reconnaissance and builds its own picture on its own on-board blackboard — no swarm link, no central coordinator in the loop. When done they reconvene at an agreed rendezvous, where the shared blackboard merges all three pictures into one common operating picture — deduplicated, with every contact each drone found. Losing the link never turns the drones back; it only defers the picture-sharing to the rendezvous.
Watch for: three separate recon-area rings (amber while the drones are comms-independent, turning green as each delivers), the convergence lines drawn to the rendezvous point as they reconvene, and the banner switching to “shared blackboard MERGED — N contacts fused”.
The AI-forward heart of the console: describe a whole mission in plain language in the command bar at the top — e.g. “fly to the northern edge of the area and scan the road for vehicles”. A sovereign on-device model (running on our own hardware — nothing leaves for a cloud service) turns it into a concrete, reviewable sequence of command verbs — take off, form up, follow the road, track a vehicle class, report — which you review and confirm; the master then loads the plan and the swarm’s behaviour tree executes it step by step. Named places are resolved to real coordinates automatically, so “fly to <town>” just works.
Watch for: your sentence becoming a numbered, step-by-step plan in the Context · Tasking panel — each step naming the drone, the command verb and its parameters — and, on confirm, the master loading the plan and the swarm beginning to fly it.
The swarm is commanded to start a mission and drives itself from a standing start through take-off to mission execution, entirely autonomously: the elected master runs the full launch lifecycle — arm, climb past the airborne gate, form up, execute — with no manual arming or take-off. It is the swarm configuring and launching itself on a single command.
Watch for: the mission phase advancing to EXECUTE and both drones climbing to altitude, rendered airborne on the map. (A short autonomous flight — the fleet lands itself at the end.)
The console follows one idea: one picture. The map is the single operating picture, and everything else annotates it.
| Element | What it shows |
|---|---|
| Map | The live operating picture. Drones and detected objects are drawn as standard military symbols (APP-6 / milsymbol), so the scene reads at a glance. |
| Roster | The fleet: each drone, its role and its status. One drone is the elected MASTER (the swarm's single gateway to you); another stands by as DEPUTY and takes over if the master drops out. |
| Tracks / entities | Objects the swarm has detected and is following, shown as symbols on the map with a heading/vector. |
| SITREP | The AI-generated situation summary (most prominent in the SITREP scenario). |
| Jammers & comms-loss zones | In the GNSS-denied / EW scenario, detected jammers are magenta markers and their estimated interference (comms-loss) areas are shaded magenta rings — the electronic-warfare picture on the map. |
| APNT banner | An amber “APNT ACTIVE — GNSS denied” banner appears when a drone has failed over from satellite navigation to vision/UWB, and names the source it is now navigating on. |
| Assignment / pursuit links | A cyan dashed line from a drone to a target shows which drone is covering or pursuing it (multi-target and pursuit scenarios). |
| Rendezvous overlay | In the comms-out rendezvous scenario: per-drone recon-area rings (amber while comms-independent, green once delivered), convergence lines to the rendezvous point, and a banner that flips to “shared blackboard MERGED” at reconvene. |
| COMMAND (header) | Which drone is the elected master right now. In the master-loss scenario you see it switch to the new leader live. |
| Autonomy & attention | The current autonomy level and anything the system wants you to look at — so you supervise by exception rather than micro-manage. |
Depending on the scenario, some panels are more active than others — e.g. the SITREP panel in the SITREP scenario, the jammer/APNT overlays in the GNSS-denied scenario, the assignment links in the multi-target scenario, or the command indicator in the master-loss / C2-loss scenarios.