The Materials for Physical Compute in Untethered Robotics effort seeks to make autonomous systems more intelligent, while Decentralized Artificial Intelligence through Controlled Emergence aims to help robots form teams and carry out missions. These DARPA projects may feed ideas to the Defense Autonomous Working Group, the lead Pentagon office for drone warfare, whose budget would soar from $226 million this year to $54 billion under the new 2027 spending proposal. 

Much of that huge sum will be wasted if the military spends it before establishing a clear understanding of how operators will buy, train on, use, and maintain autonomous weapons, according to a recent commentary piece by David Petraeus, the retired Army general and former CIA director, and scholar Isaac Flanagan. Writing for The Hill, they argue that the lack of such understanding constrained the use of drones during the past decade of U.S. wars in the Middle East.

“Each Predator combat air patrol of continuous surveillance required nearly 150 personnel,” they write. “As demand for drone coverage surged, the limiting factor was not the number of aircraft but of the trained personnel and the organizational structure to enable them.” 

Until the military fixes this, they write, any new drone is “not a weapons system at all—it is an asset on a spreadsheet.”

The new DARPA efforts aim to help change that.

Materials for Physical Compute in Untethered Robotics seeks to help robots think and reason without relying on connections to vulnerable data centers and without using valuable battery life to upload video and receive commands. Even the most advanced robotics “still require constant internal data processing, with either the end-users or data centers, creating delayed actions through latency and consuming power for data transmission,” the request for information says.

The RFI also urges industry to move beyond the conception of autonomous systems as assemblages of wire, metal frames, and motors. This mindset has been “yielding a robot with small behavior diversity. Therefore, current robot capabilities are limited in ever-changing and contact-rich environments.” It seeks new concepts at the “material, component, and kernel level”—down to chemistry and physics—that can change the very nature of machine intelligence.

DICE aims to enable machines to talk and collaborate with each other, to “dynamically form teams using peer-to-peer coordination to execute complex missions.”

The two DARPA projects are hardly the Pentagon’s only efforts to answer fundamental questions about robots. A contest run by DIU, the Defense Department’s innovation arm, seeks ways to control drones with plain language commands, as one might direct a soldier or a large-language-model tool.

The problem that Petraeus and Flanagan discuss is not as simple as it seems. Technology is moving faster than doctrine. So should doctrine come first? Or the other way around?

U.S. Southern Command is moving to answer that sort of question. Last week, Gen. Frank Donovan announced the SOUTHCOM Autonomous Warfare Command to “maximize the efficient fielding of autonomous systems.”