Enabling Military Decision-Making at Operational Tempo

Tue, 10/03/2023 - 12:04
Ajai Raj

The military can be conceived as a vast, intricate machine for making and executing plans — and as it turns out, that’s an incredibly useful way to improve how it operates. By applying algorithmic approaches used to help robots navigate terrain and perform tasks, scientists at the Johns Hopkins Applied Physics Laboratory (APL) in Laurel, Maryland, are enabling the military to automate key facets of tactical decision-making, ensuring the U.S. can maintain its competitive edge well into the future.

Complexity Versus Time

“You can think of the military as an incredibly complex ‘internet of things’: people, assets, ships, planes, weapons, sensors and so on,” said Greg Hicks, a mathematician in APL’s Air and Missile Defense Sector (AMDS), who, along with colleague Sawyer Elliott, is leading the internally funded effort. “The Army handles this complexity by using a decision-making process that walks commanders through the steps they need to take to ultimately arrive at a course of action — to make a plan to achieve a given objective. And the process is quite good.

“But our military likes to keep the initiative, to go on offense and keep adversaries on the defensive, so there’s not always time to work through the process in its entirety,” he continued “And the more information-rich the situation, the more detailed the plan, the more time everything takes.”

This problem — successfully managing the fundamental tension between complexity and time inherent in mission planning — is known in military parlance as “decision-making at tempo.” How can you make plans quickly without compromising the quality of those plans?

Same Process, New Efficiencies

As it happens, the problems the military needs to solve to execute missions are essentially no different than the problems that roboticists need to solve to enable robots to successfully navigate difficult terrain and carry out collaborative tasks. In both cases, the goal is autonomous behavior, achieved through processes or algorithms, that can be trusted.

“We have to be able to tell robots: ‘go to position A, do something there and come back’ — and the military has problems just like that: ‘Military unit, go to this enemy camp, clear them off and come back,’” Hicks said. “Commanders don’t micromanage each unit — they give them set objectives and expect them to execute.”

Hicks emphasizes that introducing formal algorithms into the process does not fundamentally change the military’s processes, which are already mathematical in nature. The difference is analogous to an accountant doing all the math with pencil and paper or using software.

“If we do it right, we can take the parts of military planning that are time-consuming for humans — like breaking maps into grids — and hand those off to computers, which can perform the task much more efficiently,” he said. “We’re only looking for places where we can do what the military is already doing manually, but with faster algorithms, so as to alleviate the time tension.”

Quantified Risk for Deliberate Tactical Decision-Making

Outsourcing calculation-driven planning tasks to algorithms provides another decisive advantage besides speed: the ability to quantify risk.

“It often happens in the military that you have a limited time window to make a plan, and then you have to move on. You take the hit in the uncertainty that comes from having to make certain assumptions and using crude abstractions,” said Hicks. “The difference is that when we use an algorithm, we can quantify the risk that’s involved. We can consciously say, ‘OK, given that we have an hour to make this plan, we can give you an answer that’s accurate within 100 meters.’

“Computers aren’t get-out-of-jail-free cards; if you hand a computer a big problem, it will take a long time. But what you get is a precise notion of the trade-off between complexity and time,” he explained. “Commanders are always acting at risk — what this approach gives them is the ability to choose exactly how much risk they’re willing to bear to maintain the operational tempo they’re after.”

Abstracting Up the Hierarchy

This effort — known internally as Wolf Howl — is now in its third year of internal funding. Each year has involved applying the algorithmic approach at increasingly higher levels of abstraction in a hierarchy: the first year focused on the individual unit level, the second focused on the tactical level and this year focused on the mission level.

“In the first year, we worked out and verified an algorithm to get a multi-agent system to do something: hold a certain formation, for example. The military has a whole library of things they need small units to do,” said Hicks.

“Then, once we had that locked in, we asked, when unit-level orders are combined to get a team of teams to achieve a tactical objective — a course of action or COA — how do we verify it meets performance specifications? If the enemy has a policy to reach and avoid and our policy is to seek and capture, and we begin with our units in these positions and the enemy in those positions, then this is how it will play out.

“And now in our third year, having verified our tactical-level work, we’re working on algorithms to synthesize mission plans,” Hicks continued. “Or, in algorithmic terms: If X is the beginning state, and Y is the desired end state, then here is the sequence of intermediate states that are required to get from X to Y with minimal specification violation.”

The ability to reliably abstract across the entire hierarchy is where the value of Wolf Howl really begins to show itself. “We’ll give commanders the ability to ‘wargame’ different strategies from mission to unit within a given time frame or risk tolerance,” he said. “That way, humans and computing machines can focus on aspects of planning they are each currently better suited to, and you really try to get the best of both worlds.”

Testing on Real Robots

To verify that the algorithms are effective in real-world, dynamic physical systems, APL is collaborating with the University of Maryland, using the university’s Collaborative Controls and Robotics Lab. “We’re actually going to run our algorithms on real robots to verify that they can coordinate the actions of autonomous agents to do what we’ve tasked them to do,” said Hicks.

Robert Patterson, a technical leader in AMDS who is overseeing this work, said that the work has exceeded his expectations.

“When I first heard the Wolf Howl proposal, the idea of mathematizing military planning seemed incredible, yet very compelling,” said Patterson. “As the mission-level results emerge, it’s clear that the concept is both credible and compelling. I look forward to engaging our forward-looking sponsors with the final autonomous agent results, seeking transition sponsors and ultimately seeing this capability employed by the warfighter.”