Early this year DARPA’s Squad X technology development program held a week-long series of tests at Twentynine Palms, Calif. While Marines demonstrated the powerful ISR potential of emerging autonomous technologies, they also proved a less-obvious point about the need for pre-planning at the intersection of autonomy and combat.
“We demonstrated that the artificial intelligence starts in mission planning and rehearsal. That’s key,” said Army Lt. Col. Phil Root, program manager for the agency’s tactical technology office.
The experiment had soldiers working alongside multiple unmanned ground and aerial vehicles equipped with a range of advanced targeting and sensing gear, including radio frequency and multispectral sensors. Testers sought to integrate the new abilities into the battle scenario not just as dumb tools, but rather as collaborative partners fighting alongside their human peers.
“We are asking a young Marine to wrestle with how to trust AI,” Root said. “As we included the AI in the rehearsal, we could place the blue forces into the decision engine and see what actions the AI would take.”
The hope, he said, is that human fighters would have a greater sense of confidence in their autonomous partners, as well as a better understanding of how the autonomous systems would likely act on the battlefield.
“That’s normally what you do: Before your mission you look at all your Marine and confirm that they have the plan in their head. But with autonomy we have never really done that. You had to just trust the autonomy. So this was something new.”
Understanding windows and doors
An ongoing research project, Squad X has brought to the forefront a number of emerging ISR capabilities. In earlier experiments, the researchers have pushed the boundaries of air-to-ground sensing. In one instance, they offered Marines the ability to draw an area on a map and to instruct drones to conduct reconnaissance. The aircraft mapped the zone well enough, but commanders wanted more: They wanted the drones to identify and highlight individuals fleeing the scene.
“They wanted the [unmanned aircraft systems] to serve as an airborne squad member, without a Marine having to take out a controller and telling it to change its behaviors,” Root said. “It was trivial to make those changes in software and the next day we could see the Marines using the technology in a way that we had never seen before. That’s a tremendous change from the current operations concept in which the Marine has to tell the UAS everything it needs to do.”
This approach could free up the Marine to be a more effective combatant. “Instead of taking the Marine out of the fight in order to control the UAS, we have returned them to the fight and given this augmented capability,” he said.
Squad X also has been developing enhanced ground-to-ground detection capabilities.
Robots on the ground can do more than just see and report back. They can, in theory, be trained to recognize and prioritize what they see – a potentially valuable ISR function.
“On the ground you might have little time to detect a threat, so we trained it to prioritize doors and windows, to understand what doors and windows were. Then the system could find individuals as soon as they came out of doors and windows,” Root said.
Infrared sensing further enhances the machines’ capabilities by allowing them to “see” into shadows where the human eye can’t readily discern objects or individuals.
The evolving systems layer on multiple points of proof including IR scans and motion-detection data, in order to help soldiers quickly confirm a threat assessment. The robot uses simple graphic interface to inform the operator: “Here’s the motion I saw, here’s a still image of what I saw, here’s the weapon I saw – all this adds up to my belief that this is a threat,” Root said. “You click on a map and it reports all this information.”
Looking ahead, the Squad X team is thinking about how to squeeze more work out of autonomous systems. Typically when a unit is en route, its autonomous systems are shelved. But the team is thinking about a way to keep those robots in more continuous use.
“How do we have AI with a bias toward action? The robots should always be on the move in some way that soldiers are on the move,” Root said. “AI on the battlefield shouldn’t be waiting. It should always be looking at how to improve the local situation.”
Dr. Hans C. Mumm