While the commercial world tiptoes toward the notion of a self-driving car, the military is charging forward with efforts to make autonomy a defining characteristic of the battlefield.

Guided by artificial intelligence, the next-generation combat vehicle now in development will have a range of autonomous capabilities. Researchers at Army’s Communications-Electronics Research, Development and Engineering Center (CERDEC) foresee these capabilities as a driving force in future combat.

“Because it is autonomous, it can be out in front to find and engage the enemy while the soldiers remain safely in the rear,” said Osie David, chief engineer for CERDEC’s mission command capabilities division. “It can draw fire and shoot back while allowing soldiers to increase their standoff distance.”

Slated to come online in 2026, the next-gen combat vehicle won’t be entirely self-driving. Rather, it will likely include a combination of autonomous and human-operated systems. To realize this vision, though, researchers will have to overcome a number of technical hurdles.

Getting to autonomy

An autonomous system would need to have reliable access to an information network in order to receive commands and relay intel to human operators. CERDEC’s present work includes an effort to ensure such connections.

“We need resilient comms in really radical environments — urban, desert, trees and forests. All those require new and different types of signal technologies and communications protocols,” David said.

Developers also are thinking about the navigation. How would autonomous vehicles find their way in a combat environment in which adversaries could deny or degrade GPS signals?

“Our role in this is to provide assured localization,” said Dr. Adam Schofield, integration systems branch chief for the positioning, navigation and timing (PNT) division.

In order for autonomous systems to navigate successfully, they’ve got to know where they are. If they rely solely on GPS, and that signal gets compromised, “that can severely degrade the mission and the operational effectiveness,” he said.

CERDEC, therefore, is developing ways to ensure that autonomous systems can find their way, using LIDAR, visual cues and a range of other detection mechanisms to supplement GPS. “We want to use all the sensors that are on there to support PNT,” Schofield said.

In one scenario, for example, the combat vehicle might turn to an unmanned air asset for ISR data in order to keep itself oriented. “As that UAV goes ahead, maybe it can get a better position fix in support of that autonomous vehicle,” he said.

Even as researchers work out the details around comms and navigation, they also are looking to advances in artificial intelligence, or AI, to further empower autonomy.

The AI edge

AI will likely be a critical component in any self-directed combat vehicle. While such vehicles will ultimately be under human control, they will also have some capacity to make decisions on their own, with AI as the software engine driving those decisions.

“AI is a critical enabler of autonomy,” said CERDEC AI expert Dr. Peter Schwartz. “If autonomy is the delegation of decision-making authority, in that case to a robotic system, you need some confidence that it is going to make the right decision, that it will behave in a way that you expect.”

AI can help systems to reach that level of certainty, but there’s still work to be done on this front. While the basics of machine learning are well-understood, the technology still requires further adaptation in order to fulfill a military-specific mission, the CERDEC experts said.

“AI isn’t always good at detecting military things,” David said. “It may be great at recognizing cats, because people post millions of pictures of cats on the internet, but there isn’t an equally large data set of images of adversaries hiding in bushes.”

As AI strategies evolve, military planners will be looking for techniques that enable the computer to differentiate objects and actions in a military-specific context. “We need special techniques and new data sets in order to train the AI to recognize these things in all different environments,” he said. “How do you identify an enemy tank and not confuse that with an ordinary tractor trailer? There has to be some refinement in that.”

Despite such technical hurdles, the CERDEC team expressed confidence that autonomy will in fact be a central feature of tomorrow’s ISR capability. They say the aim is create autonomous systems that can generate tactical information in support of war-fighter needs.

“As we are creating new paradigms of autonomy, we want to keep it soldier-centric,” David said. “There is filtering and analyzing involved so you don’t overwhelm the user with information, so you are just providing them with the critical information they need to make a decision.”