Sand Table Technologies
Command autonomy at the speed of speech.
Voice command and control for swarms and autonomous robots. And a digital sand table to train and plan with the people who lead them.
The problem
Autonomy is outrunning command.
- Drone swarms and robot fleets are scaling from one machine to hundreds.
- Control still means one operator per machine: sticks, screens, waypoints.
- A commander leading a swarm under pressure, on a degraded network, needs to state intent and let the units work out the rest.
Swarm control
Command the swarm, not the drone.
One person directs many autonomous units by voice. Each unit carries its own small AI subordinate.
Intent
Orders, not waypoints.
“Screen the ridge.” “Box the objective.” “Fall back when you’re down to ten.” The swarm breaks intent into tasks for each unit.
Edge
Works when the link doesn’t.
Language models run on the device, with no cloud round-trip. Units hold the plan locally and act on a signal, even on a jammed or cut-off network.
Oversight
The human stays in command.
Every order becomes a checkable plan before it runs: who, what, where, how, when. Units regroup, hand off tasks and report back.
Cognitive training & planning
Train the commander. Rehearse the plan.
The same sand table that commands machines trains the people who lead them.
01 · Train
Clear orders under pressure.
Commanders drill giving complete, unambiguous orders against the clock. They see exactly how each order was understood, and where it fell short.
02 · Plan
Build the plan on the table.
Lay out a scheme of manoeuvre on the digital sand table, then run it against an AI opposing force before anyone moves.
03 · Rehearse
Wargame at machine speed.
AI-vs-AI battles test courses of action quickly. Every order and outcome is logged for after-action review.
How it works
Speak. Interpret. Execute.
Command by voice, executed by AI subordinates on the device.
01 · Speak
Orders, the way commanders give them.
Plain speech: units, sectors, formations, a signal. Nothing new to learn.
02 · Interpret
Small models, on the edge.
Language models on the device turn speech into a precise plan the commander can check.
03 · Execute
Subordinates act and report.
Each unit’s AI subordinate carries out its part on its own, then reports back.
The order grammar
- WHOSwarm Two,
- DOmove
- WHEREto K12
- HOWin box,
- WHENon my signal.
Five parts that combine freely: formations, flanking, patrols, passing units between commanders, holding ground, breaching obstacles, conditions such as “when you’re down to ten, fall back”, and synchronised execution on a signal.
The demonstrator
POLEMARCH. The whole loop, playable.
A live proof in the browser: a commander’s voice, AI subordinates, a battlefield that answers.
- Voice orders to on-device AI subordinates
- Many units under one commander, in formation
- Conditions, signals and synchronised execution
- An AI opposing force, and AI-vs-AI simulated battles
- Networked force-on-force play
Applications
One commander. Many machines.
Defence
- Swarm command and control: drones and ground robots under one voice.
- Human-machine teaming at the edge.
- Commander training and decision drills.
- Mission planning, rehearsal and wargaming against an AI opposing force.
Industrial
- Fleet command for autonomous robots by voice.
- Logistics yards, warehouses and ports.
- Mines, construction sites and agriculture.
- Operator training and shift planning on the same table.
Why now
The pieces just arrived.
01
Models fit on the device.
Small language models now run fast enough without a datacentre.
02
Swarms are ahead of command.
Autonomous units are ready faster than the means to direct them.
03
The cloud is not there.
Contested and cut-off environments rule it out. The intelligence has to travel with the unit.
See it work
Give the order.
POLEMARCH runs in the browser. Speak, and watch the units carry it out.