AI expert addresses military use of AI in synthetic modeling

By Shane Klestinski, Associate Editor

 

Keith Brawner, Ph.D., program manager of the University of Southern California’s Institute for Creative Technologies (ICT) discussed the integration of artificial intelligence (AI) in military training and simulation systems in his session, “Using AI to Improve Synthetic Modeling,” at the Institute for Defense and Government Advancement’s (IGDA) Military Simulation Training Summit in Orlando, Florida, Feb. 25.

 

Brawner earned his bachelor’s in electrical and electronics engineering, as well as his master’s degree and a doctorate in artificial intelligence, from the University or Central Florida, which also presented him with the “UCF Alumni 30 Under 30” award. He worked as an engineer for over four years at the Naval Air Warfare Center Training Systems Division, and he is currently the U.S. Army Combat Capabilities Development Command Soldier Center’s AI lead at the Simulation and Training Technology facility, which are both located in the Research Park area near UCF campus.

 

IGDA’s two-day summit gathered over 300 senior training commanders, procurement officers and acquisitions leaders from across the military and provided a forum for networking and collaboration. This conference focused on training warfighters with simulation and synthetic training, and it explored how militaries (American and allied) are approaching interoperability with new technology and moving toward competency training.

 

Near the beginning of his talk, Brawner described AI as “a bit of a mess, but it’s organized,” as he described applications of AI toward a variety of fields, from engineering, to social, natural and health sciences, which overlap with data science. He noted that most of the problems in the private sector also exist in the military sector as he highlighted the use of generative AI and synthetic environments to enhance training realism, citing One World Terrain’s success.

 

“In an older program, the idea was that you’d have a reconfigurable space and you could reconfigure the walls, each wall would have a projector, and the projectors could be coded to the individual area, which could be recoded, based on the type of training,” Brawner said. “For reasons that are perhaps obvious, we don’t do a lot of that anymore. Instead, we use virtual training that can be reconfigured to whatever area of the world you’d like inside the software.”

 

He went on to discuss technology that employs three-dimensional scanning to take pictures and stitch those pictures into a model that has training applications. In the examples Brawner cited, users had access to synthetic environments with synthetic rendering, as well as real environments with synthetic rendering in outdoor and indoor scenarios. He explained that the technology used a number of techniques related to photo telemetry and LIDAR scanning.

 

“If you’re having difficulty telling what’s real and what’s not – good,” Brawner said.

 

Brawner noted that as new tools develop, they could capture and use metrics about environments to generate more advanced modeling techniques. Having principal AI techniques that generate factors relevant to systems downrange could also address a lack of data in military environments.

 

“A zero-shot classification system that can do things in real time to classify things inside a three-dimensional environment,” Brawner said. “The three-dimensional environment looks a lot like the real environment [and] because you have a three-dimensional model of these things, you can do cool stuff – like blow it up. If you blow it up, you can reform the terrain in real time… then publish that as a terrain update. You can also take good guesses as to what the materials are made of and how they respond to different types of blasts.”

 

As a brief aside, he also mentioned the software acquisition pathway that could bypass the traditional acquisition processes but clarified that the full content and context of that topic involved an entirely different brief.

 

“There is something called the ‘software acquisition pathway,’ by which you can bypass the local acquisition process to quickly fill technologies, as long as it’s not adding anything additional insofar as hardware onto existing systems,” Brawner said. “If the existing hardware can run something, you can actually build a software requirement.”

 

He later explained that using generative AI infill techniques could build a large database of faces for a conventional model, but the practice also had other applications. To illustrate his point, Brawner displayed an image of a bunny rabbit made from various materials.

 

“Instead of using that type of approach of doing faces and face-to face-modeling, you could do that for a variety of different materials, like making models of tanks,” Brawner said.

 

AI-generated content also has relevance in training military officers to manage staff, according to Brawner. In an example, he featured a program’s training visuals that described an area of operation, its historical context, the actions available to learners, and their commander’s orders.

 

“You can also talk to your brigade staff on how to manage these operations,” Brawner said. “That’s one of the tickets we’re using to train the new Field Manual 3.0 [on operations] and how to scale that across the Army. [The training characters] have realistic, generated voices, their opinions are motivated by their individual job descriptions, and their job descriptions are applied to the area of operations.”

 

A brief Q&A period followed Brawner’s remarks, wherein he answered questions about transitions to the Army, such the Oculus Rift virtual reality headset and the counter-IED system that saw deployment during the Global War or Terror. Other questions involved ICT’s recruiting focus, as well as One World Terrain’s science-and-technology side and its Army programmatic side. Brawner also noted the institute’s partnerships and technology transition efforts that aim to maintain innovation and relevance in military training technologies.