FM-01Tested in simulation
GPS jammed
They keep going on what they can still trust.Each vehicle works out where it probably is, and how sure it is, and plans around that.
01Threat picture
Picture a dozen drones watching a border at night.
Then someone jams GPS. The radio link gets patchy. The operator back at base loses contact. Each drone can usually keep itself in the air. What falls apart is the team: who watches which stretch, who follows what was spotted, and who fills the gap when one goes down.
Condition boardContested
GNSSJammed
Radio linkPatchy, then gone
Link to baseLost
Each droneStill flying
The teamFalls apart
This is what current conflicts look like. It is not rare any more.
02Product / TurtleShield
TurtleShieldAI
It puts the whole team plan on every vehicle. No vehicle waits for base, and no vehicle is the boss. If the group gets split up, each part carries on by itself.
Every drone has an autopilot that flies it very well. It has no idea what the other drones are doing. So today the teamwork usually runs from a laptop or server at base, and when that link is jammed, every vehicle is left waiting for orders. We moved the teamwork onto the vehicles.
Where it sits
PeopleSet the mission, approve, pause, call everyone home
TurtleShieldRuns on every vehicle and keeps the team working together
AutopilotFlies each vehicle: balance, steering, motors
VehiclePlanes, ground robots, boats
SpecificationTS-01
- Runs on
- Every vehicle, on a companion computer next to the autopilot
- Built for
- NVIDIA Jetson Orin with a Cube Orange flight controller
- Autopilots
- PX4 in all flying tests. ArduPilot tested standing still, in simulation.
- Domains
- Fixed-wing aircraft, ground robots, surface boats
- Coordination
- No central brain. Every vehicle carries the full plan.
- Keeps working through
- GPS jamming, radio loss, losing a vehicle
- Authority
- Finds, follows and suggests. A person decides. No weapon code.
- Status
- Tested in simulation. First real flight planned for month 3.
FM-02Tested in simulation
Radio cut
Messages find another way through.Vehicles pass messages along for each other and hold them until a path opens up again.
FM-03Tested in simulation
A vehicle is lost
Someone else picks up its job.The rest of the group notices the gap and shares out the work again, without being told.
StatusAll of this has been tested in simulation. None of it has flown on a real aircraft yet.
03Core principle
Know what you don't know.
Most systems store a position as a dot on a map and trust it. Ours stores the dot and a circle around it that says how sure the vehicle is. The worse the GPS, the bigger the circle, and the vehicle plans around the circle.
GPS good
Dashed box: the spot it takes. Orange circle: how sure it is of its own position.
changes to our core software when we added a second autopilot brand.
Armies rarely buy all their drones from one maker, and they change suppliers. So TurtleShield sits on top of the autopilot. Adding a new kind of vehicle means writing a small adapter. The part that does the thinking stays exactly the same, and customers keep the drones they already own.
The second brand, ArduPilot, has only been tested standing still in simulation so far. All our flying tests use the first, PX4.
04Product / Turtle Eyes
Turtle Eyes
What the operator sees. Three modes, one click each. Orders go in as plain English and nothing moves until a person holds the button down.
Built for a rugged laptop in the field, with its own maps and no internet needed. It tells vehicles where to go and what to watch. Nothing more.
TURTLE EYESPERSISTENT OVERWATCH
TARGETING: find something of interest and keep eyes on it.▲AIR■GROUND◆SURFACE◇TRACK□HOME
VEHICLES 9MESH PEERS 8REJECTED/S 0ALERTS 1
Drawn for this page. The real Turtle Eyes runs on sample data today and says MOCK in the corner, same as this one.
SpecificationTE-01
- Modes
- Targeting, surveillance, simulation. One click each.
- Orders
- Plain English, shown on the map, one hold-to-confirm
- Network
- Works with no internet and carries its own maps
- Hardware
- Built to run on a rugged laptop in the field
- Access
- Operators see only what they need. Engineering controls sit behind a separate login.
- Control
- Pause, stop, or call every vehicle home at any time
- Scope
- Tells vehicles where to go and what to watch. Nothing more.
- Status
- Running on sample data today
05Evidence / as of 15 Sep 2026
What we've proven. And what we haven't.
We keep every failure, because that is where the lessons are. Everything on the left happened in simulation, with the same autopilot software that flies real aircraft.
Demonstrated in simulationMeasured
742
test missions flown in simulation
531 passed, 141 failed, 70 were skipped. Every failure is kept on record.
2,000+
orders from our software, accepted by the autopilot
Our software commanding simulated aircraft after the base station handed over, counted across recorded runs.
8
aircraft in tight formation, kept safely apart
6 out of 6 runs passed. Closest approach 33.25 m against a 30 m limit, measured from the simulator's true positions.
90%+
less position error without GPS, after one day of fixes
Average error went from 396–750 m down to 25.5–39.0 m. Still not accurate enough to rely on.
103
problems written down, with their causes
47 of them are still open.
Not yet demonstratedRead this first
0flights on a real aircraft.
- Everything so far is simulation, or a real flight controller wired to a simulator.
- The biggest group flown together: 8 aircraft in simulation, 11 with a real flight controller in the loop.
- Navigation without GPS has improved fast, and is still not accurate enough to rely on.
- Each vehicle uses about 33 times the radio capacity we budget for.
- Some backup behaviours are written and tested, and not yet switched on in the flying software.
06Human authority
The software finds and follows. A person decides.
01 →Spot something
02 →Keep track of it
03 →Suggest a way to intercept
04A person decides
TurtleShield contains no weapon code. It stops at a suggestion, and a person takes it from there.
An operator can pause the mission, stop it, or call every vehicle home at any time.
07Programme / next 12 months
Out of the simulator, into the air, with someone watching.
MONTH 3
First real flight with our software in charge
A safety pilot on standby, every bench safety check done first.
MONTH 6
Three aircraft in formation, in real wind
Radio use brought down to budget. Real numbers replace simulated ones.
MONTH 9
Flying without GPS, against a jamming rig
Backup behaviours switched on. Our core idea, tested for real.
MONTH 10
A five-aircraft field trial, run by an outsider
Operated through Turtle Eyes, with jamming and radio trouble, and signed off by someone who watched.
08Contact
Work
with us.
with us.
Homingvector is raising a pre-seed round to take TurtleShield from simulation to a witnessed field trial. Founded by Nidhip Sharma, who wrote TurtleShield and the testing system behind it. If you fly drones, build them, or fund the people who do, write to us.
© 2026 Homingvector / India
Performance figures come from simulation unless stated otherwise
No cookies