Human-Machine Formations: The Constraint Is the Operator

The number that matters in the human-machine formations discussion is not the vehicle count. It is the number of people required per vehicle, and what happens to that number when the link drops.

Walk the floor at the Association of the United States Army's annual meeting and the hardware is impressive. Tracked chassis with remote weapon stations, quadrupeds with sensor masts, convoy trucks that follow a leader. None of that is the hard part anymore.

The hard part is that a formation is a manpower document. The Army can buy robots faster than it can write the positions, the training pipeline, and the doctrine that tell a sergeant what to do with four of them at 2am in a jamming environment.

The formation changed, not just the inventory

Human-Machine Integrated Formations are not a procurement line. They are a restructuring argument. A company that owns robotic systems as organic elements has a different table of organization, a different maintenance concept, and a different casualty calculus than a company that requests them.

That is why the concept is more consequential than any single vehicle program. Adding a robot to a platoon is a fielding decision. Making the robot part of the platoon's structure is a doctrine decision, and doctrine decisions outlive administrations.

The vehicle programs get the attention because they have pictures. The structure question is where the schedule actually lives, and it is being answered by people who do not give conference talks.

The operator math does not work yet

Today's ground robots require close supervision. A control station, a dedicated operator, and a crew that treats the machine as a task rather than a teammate. That model does not scale past a handful of vehicles in a formation, because the Army does not have spare soldiers to sit in trucks watching a screen.

The target is more vehicles per operator. That target is a claim about autonomy, and it is also a claim about trust. Somebody has to be willing to let a machine move across open ground while they are doing something else.

Two things have to be true before that happens. The autonomy has to be reliable in the boring cases, not the demo cases. And the operator has to be trained to a standard that does not exist yet.

That second item is the one programs underfund. You can accelerate a vehicle program with money. You cannot accelerate the institutional knowledge of how to employ a new kind of unit. It took more than a decade to build real expertise in unmanned aircraft employment, and those operate in a medium with fewer surprises than a muddy field with civilians on it.

The Army can buy a robot in years. It takes a generation to learn how to use one.

The network is the second constraint

Every autonomy concept assumes connectivity, and every tactical network assumes it will be contested.

The bandwidth problem is not subtle. Current tactical networks strain under video from a few unmanned aircraft. A formation of ground robots generates position, telemetry, sensor, and command traffic at rates that were not part of the design assumptions for the radios carrying them.

Then add electronic attack. The question that matters is not what happens when the link is good. It is what the vehicle does when it loses the link in contact. Stop and wait is safe and useless. Continue the mission is useful and requires a machine that can be trusted with the decision. Return home is a plan until the route home is the compromised one.

Doctrine has to answer that before the procurement does. Otherwise the answer is improvised by a 22-year-old under fire, which is how a lot of doctrine gets written and not how anyone should plan it.

Authority is settled on paper

The Department of Defense has a published position. Directive 3000.09, updated in 2023, requires appropriate levels of human judgment over the use of force and sets up a review process for autonomous weapon systems. That is the policy.

In practice, the boundary is already blurry at the defensive end. Shipboard systems have engaged incoming threats at machine speed for decades under human supervision, because a human cannot react in the time available. Nobody calls that autonomous weapons employment, and the distinction is doing real work.

Ground formations will push on that boundary from the other direction. A robot that can identify a target and hold it for a human decision is a different system from one that can act on the identification. The first is a sensor with legs. The second is a policy question wearing a chassis.

I do not think the policy will collapse. I think it will get specific, slowly, in the way that acquisition programs force specificity. The uncomfortable part is that specificity tends to arrive after the first operational surprise, not before.

What this means if you build in this space

The differentiator is not the chassis. Anyone can integrate a remote weapon station onto a tracked platform, and several companies already have.

The value is in four places. Autonomy that degrades gracefully rather than confidently. A control interface one soldier can run for four vehicles without a second screen. A sustainment model that survives a deployment with no depot nearby. And a simulation environment realistic enough that units can train before the hardware arrives, because the hardware will never arrive in the quantities the training plan wants.

That last one is the most neglected and the most likely to be bought. The Army cannot field enough robots to train every unit on real metal. It can field software that behaves like the metal, badly enough to teach the failure modes.

Human-machine formations are coming, and the services are serious about them. But the thing that will decide the timeline is not the autonomy stack. It is whether a sergeant with four vehicles and a degraded radio can be given a doctrine that says what to do next.

The programs that answer that question will be fielded. The programs that answer the vehicle question and leave the human question open will produce excellent demonstrations and a very long requirements document.

The robot is the easy purchase. The formation is the hard one.