Target drones made by Michigan-based Swarm Defense Technologies were on display on November 21, 2025, during the Project Flytrap 4.5 counter-drone exercise at the military training center at Putlos, Germany.

Target drones made by Michigan-based Swarm Defense Technologies were on display on November 21, 2025, during the Project Flytrap 4.5 counter-drone exercise at the military training center at Putlos, Germany. Morris MacMatzen/Getty Images

The Pentagon leans into drone swarms with a $100M challenge

An Ender’s Game challenge illustrates the U.S. military's evolving approach to AI.

The U.S. military wants to make it easy for troops to direct swarms of drones—on land, at sea, or in the air—and is offering vendors up to $100 million to do so.

On Tuesday, the Pentagon’s Defense Innovation Unit announced the Orchestrator Prize Challenge, the latest sign that the military is looking beyond simply buying warbots in bulk toward the larger challenge of giving commanders a way to command groups of different drones from different makers.

“We want orchestrator technologies that allow humans to work the way they already command–through plain language that expresses desired effects, constraints, timing, and priorities—not by clicking through menus or programming behaviors,” said Lt. Gen. Frank Donovan, who leads the Defense Autonomous Warfare Group. The DAWG, essentially a re-branding of the Replicator initiative, is running the challenge along with DIU and the Navy.

Replicator sought to procure thousands of small, highly autonomous drones quickly but missed a key August goal for delivery. (The bar is set by Ukraine, which last year sent more than a million drones to its frontline units.)

The Pentagon has been hacking away at bureaucratic obstacles that prevent the fast acquisition of drones, but the larger problem is how to make them actually useful to commanders who can’t spare troops to operate one drone at a time, as Ukrainian units do.

Hence the Ender’s Game-like “Orchestrator” challenge, to find a “robust, scalable and vehicle-agnostic capability for understanding, tasking and coordinating autonomous systems at the fleet level.”

The concept of swarm robotics goes back more than 30 years, to this 1995 paper. The Pentagon has been holding multiple-robot challenges, like the Navy’s SWARM and LOCUST events, for more than a decade. But SWARM used about a dozen autonomous boats in very simple naval maneuvers, while LOCUST employed choreographed maneuvers.

The challenge now is to integrate multiple drones capable of autonomous decision-making and to subordinate them to a human, or at least to human judgment (which is a closer reading of the actual Defense Department policy on lethal autonomous weapons).

The end result may be the real test of how well humans and robotic swarms really understand each other.