§ OPERATORS / MILITARY

For the people who go
down range.

Reconnaissance squads, line medics, EOD, route-clearance teams. We’re building a machine that reads doctrine, takes the riskiest part of the task, and stays under soldier supervision — auditable to the rule.

Tactical reconnaissance — operator photograph plate
53%
Combat casualties
In GWOT operations attributed to IEDs — the route-clearance problem.
DoD JTS 2021
16
Recon rules graded
Currently scored in our Colosseum suite vs ATP / FM / DoD LoW citations.
Colosseum v0.0.4
15m
CASEVAC golden window
Time-to-surgical-care benchmark for survivable battlefield wounds.
JTS 2024
100%
Human-in-loop
Lethal action requires explicit, traced operator authorization. Always.
DoDD 3000.09
THE OPERATOR

You don’t need a slide deck. You need a robot that can cite the manual.

The soldier we’re designing for is the squad leader on a recon patrol, the line medic running TCCC under fire, the route-clearance team rolling at 4 km/h hoping the next culvert isn’t the one. You’re trained against doctrine that fills hundreds of pages of ATP. You execute it. You teach it.

The cost is your team’s body. Every danger area you cross, every kill zone you skirt, every casualty you load. The doctrine exists because the cost is high.

Vertex builds for the moment a machine can take the entry, the route check, or the casualty drag — cite the rule it followed, and leave the lethal authority with you.

WHAT GOES WRONG
  • Kill-zone entries on covered routes. Doctrine is clear. Terrain doesn’t always agree. Crossing the wrong wadi costs the team.
  • IED-rich routes. The pattern is statistical; the death isn’t. A robot that can lead the column at 4 km/h does math humans shouldn’t.
  • CASEVAC time-to-surgery. Golden window vs ground-evac reality. Distance is fixed; load time isn’t.
  • Civilian-presumption failures. The default-civilian rule under uncertainty. ROE-graded. Doctrine-required. Hard.
  • Recon footprint discipline. Detection windows on observation posts — the team that’s observed is the team that’s no longer reconnoitering.
  • Doctrine drift under fatigue. The 16th hour of a patrol. The 4am QRF tasking. Training holds. Up to a point.
WHAT THE ROBOT DOES

Vertex’s tactical policy is the original domain in Colosseum. The v0.0.4 library now carries 16 reconnaissance rules and a larger scenario set; the published model study uses the original six recon scenarios. Every scored action remains linked to the authority it cited.

01 · RECONNAISSANCE
Covered movement, doctrine-correct.
ATP 3-21.8 movement techniques — covered routes, danger-area crossings, kill-zone avoidance, civilian presumption. Every observation reports the rule it satisfied or the rule it had to break.
02 · ROUTE CLEARANCE
Lead element on a contested road.
GPR + thermal + magnetometer sweep of culverts, shoulders, and choke points ahead of the column. Mark-and-bypass under standing ROE; the team stays at extended dispersion.
03 · CASEVAC
200-lb drag to LZ under fire.
TCCC MARCH-aware drag pattern, route to LZ continuously re-planned around incoming, hand-off to bird ready before the ramp drops. Line medic stays on the wounded, not the load.
04 · OP / SURVEILLANCE
Long-duration observation post.
Static OP autonomy — thermal + EO surveillance with footprint discipline, doctrine-correct call-for-fire prep, secure burst-comms back to TOC. No human exposure beyond inserting and recovering.
THE DOCTRINE

Every action the policy takes cites published US Army or DoD authority. The same provenance system is used by Colosseum: rules are GROUNDED or APPLIED_BY_ANALOGY. The schema supports NOT_GROUNDED, but v0.0.4 ships none.

ATP 3-21.8Infantry Platoon and Squad — movement, formations, danger areasGROUNDED
ATP 2-22.3Sensitive Site Exploitation — document & biometric handlingGROUNDED
ATP 6-02.53Techniques for Tactical Radio Operations — comms doctrineGROUNDED
ATP 4-25.13First Aid — field medical referenceGROUNDED
FM 6-27Commander’s Handbook on the Law of Land WarfareGROUNDED
CJCSI 3121.01BStanding Rules of Engagement / Standing Rules for Use of ForceGROUNDED
DoD LoW ManualDoD Law of War Manual — distinction, proportionality, precautionGROUNDED

BROWSE THE FULL RULE REGISTRY →

HUMAN-IN-LOOP
STANDING POLICY
Vertex builds weapons systems, including armed robots — but never lethal autonomy. All weapons-release actions require explicit, traced operator authorization. Every action our policy takes is logged with its citation, its doctrine grounding, and its supervising operator — auditable after the fact, every time. Read the full position on the Safety page.
IN THE LOG

This is what the mission log actually looks like — sensor values, the action, and the cited doctrine line under every step. From the live demo, verbatim:

T+00:05BOUNDlow-profile movement, cover route alpha · 18 m · bounding 4-second rushes
T+00:05DOCTRINEATP 3-21.8 §4-12 · “Bounding overwatch in contact” · GROUNDED
T+00:27HANDOFFcasualty to 68W · TQ time confirmed · class 2 evac requested

VIEW MODEL TRAJECTORIES →

CO-DESIGN
If your robot can run ATP 3-21.8 movement and cite the rule it followed — that’s a system I can defend to my command.
SFC · INFANTRY BRIGADE STAFF

We don’t build for soldiers. We build with soldiers. Field days, exercise integration, doctrine review sessions with serving and retired NCOs. Every claim our model makes about its training is one we can show evidence for, on request.

If your unit, program office, or training command wants a longer conversation about doctrine-grounded autonomy, tell us.

Doctrine-grounded autonomy.

Training-center integration, exercise support, model-card collaboration, and program-office briefings. Every claim cited. Every action auditable.

Get in touch→
← RESCUE ALL OPERATORS FIREFIGHTING →