The Air Force is expanding its experiment with advanced, artificial intelligence-infused pilot training to the pipeline for creating drone pilots and sensor operators.
The program, called RPA Training Next, takes some cues from the Air Force’s Pilot Training Next program. In a June 1 interview, program director Maj. Adam “Boomer” Smith said this new, high-tech system of learning has the potential to teach new remotely piloted aircraft aircrew faster and more efficiently than the old system, and fine-tune their lessons to what they actually need.
“By tailoring the program to the actual requirements of RPA aircrew, both our pilots and sensor operators, in the undergraduate phase, we can get them spun up a lot quicker,” Smith said.
About a decade ago, Smith said, the Air Force adopted its current model of RPA training, which it cobbled together from existing elements of manned undergraduate pilot training. This included almost 40 hours of training in the DA-20, a light propeller-driven aircraft, and then instrument training in a T-6 simulator.
The new RPA Course, or RPAC, will consolidate the RPA Instrument Qualification Course, where students fly the T-6 simulator, and the subsequent RPA Fundamentals course, which focused on academics and mission-focused simulations.
Since 2018, Pilot Training Next has had success teaching aspiring pilots of manned aircraft how to fly through a combination of virtual reality flight simulators, biometrics measuring how students are responding to their challenges in real time, and artificial intelligence.
AI will be one of the most important components of the new RPA Course, Smith said.
Like with Pilot Training Next, the AI will track how a student is doing and tailor the flight simulation to keep challenging him or her at the right level. If a sim is proving too easy and the student is getting bored, for example, the AI can detect that and throw a few curveballs — maybe the weather gets a little choppy, or an enemy emerges. Or if a sim is too hard, the AI can bring it down a little so the pilot doesn’t get excessively frustrated and stop learning.
Eventually, the Air Force hopes to build a database that stores all the AI-related data on each student, Smith said. The service wants to have a comprehensive record of how a student responded throughout his progression through all phases of RPA training, from undergraduate to the formal training unit to the operational unit.
RELATED
This can help the Air Force draw some correlations, if a pilot has problems later on with a particular skill set, Smith said. For example, if a pilot at the formal training unit has a problem when trying to use the drone’s weapons or working with joint terminal attack controllers on the ground, the Air Force can look back in his record to see where in his undergraduate training he started to develop bad habits. And if that was caused by problems with the training curriculum, the Air Force could figure out where those problems are and fix them.
“If we can address those problems earlier, then perhaps we won’t have that attrition later in the program,” Smith said. “But also we’ll be providing a higher quality and caliber aviator down the line.”
Smith’s team is working with the Pilot Training Next team to adapt its AI-operated virtual instructor pilot program for the MQ-9 Reaper. The RPA program is already testing its virtual instructor program in one simulator, Smith said, and is trying to figure out how well it can teach students without having an actual instructor present.
Pilot Training Next’s AI also has a component called Pattern of Life, which allows components of the simulation to behave realistically, Smith said. For example, if a simulation recreated a village full of people, all the virtual villagers would go about their daily routine the way real people would — waking up, going to work, visiting friends in other houses, and going to places of worship are some of the behaviors and patterns Smith said the AI would recreate.
When “they can see a village of people interacting in a realistic manner, it really enhances the realism of our simulator,” Smith said.
The RPA program has tested the first phase of adopting the Pattern of Life program, Smith said. Eventually Smith hopes Pattern of Life will be used not just for undergraduate initial flight training, but also at operational units for the continuation training RPA aircrew need to keep qualified.
The Air Force also added a targeting pod to simulators, to give sensor operators more opportunities to take part in the training.
But not all of the elements of Pilot Training Next will be brought to the RPA program — at least, not for now.
VR won’t be a big factor, Smith said. But Smith said the RPA Course program is looking at how to use VR for more advanced, immersive training on tactics, techniques and procedures for the MQ-9.
The program also isn’t using biometrics at the moment, though Smith said it’s being discussed. Next year, he said the program might adopt eye tracking technology to see how well a student is cross-checking and what he’s really looking at.
In the real world, RPA pilots and sensor operators at a Ground Control Station have multiple screens they frequently need to cross-check, Smith said. So the RPA training program is taking the “sleds” used for Pilot Training Next simulators — which contain a chair, simulated flight controls, and a pair of screens — and adding more screens to bring it closer to what an RPA pilot would encounter.
“We want these students to get exposure [to skills such as cross-check] early on in their training,” Smith said. “If they get that early exposure, when they show up at their [formal training unit, where aircrew receive more advanced training on operating their specific RPA], they’ll be much more prepared to enter training. We can focus on higher-level skills once they get here.”
The new RPA Course, or RPAC, is coming online at the 558th Flying Training Squadron at Joint Base San Antonio-Randolph in Texas, where the majority of undergraduate training takes place. But Smith said the Air Force wants this program to focus on all phases of RPA training, including the formal training unit for MQ-9s at Holloman Air Force Base in New Mexico and the RQ-4 training unit at Beale Air Force Base in California.
The coronavirus pandemic and the travel shutdown delayed the contractor’s rollout of necessary software for its simulators, Smith said. But he hopes the first class will be up and running by late summer, or early fall at the latest.
The first class will likely have somewhere between 12 and 20 students, Smith said, and after it’s done, AETC will take a break and see how well it worked and whether they need to make any changes. The second test class will probably have 20 to 24 students. The Air Force plans to have all 400 students who come through the RPA training pipeline each year use this program.
Stephen Losey is the air warfare reporter for Defense News. He previously covered leadership and personnel issues at Air Force Times, and the Pentagon, special operations and air warfare at Military.com. He has traveled to the Middle East to cover U.S. Air Force operations.
