AFRL, AFTC collaborate on future technology via weeklong autonomy summit > Edwards Air Force Base > AFMC News
WRIGHT-PATTERSON AIR FORCE BASE, Ohio (AFRL) —
“AFRL and AFTC have been collaborating for a long time — for decades,” Pringle said. “And Gen. Dertien and I established a quarterly battle rhythm where he and I would host a weeklong deep dive into areas of mutual interest.”
Those areas included maturing technology faster and increasing collaborations.
“But most importantly, for teams to come together — meet each other, work on areas of common interest [and] identify barriers,” she said. “So that Gen. Dertien and I could elevate them and work to basically obliterate [barriers] so that our teams can do their jobs better.”
Pringle and Dertien collaborated to focus on an area of rising relevance, autonomy. Pringle said this is a critical area for future warfighters.
Dertien called the summit a great opportunity for interchange and teamwork between the two organizations, promoting further technological advancement.
With a “robust” agenda, Pringle said she looked forward to learning more, including “everything from trust to security to implementation.”
And with the summit covering all domains, Pringle challenged the audience to think outside the box.
Pringle emphasized the need to overcome challenges through collaboration.
“This is a rich opportunity for this team to come together,” she said. “We’re in it for the long haul. We want to see autonomous capabilities make it to the field and assure our technological edge as a U.S. military.”
Tammy Smeeks, AFRL-AFTC Summit coordinator and facilitator, said the event enabled linkages and future collaboration among people and organizations across the Department of Defense.
“The fall autonomy summit was a huge success in the aspects of communication and cross-service sharing of information,” Smeeks said.
Vanguards are high-priority initiatives which represent significant DAF-level commitment for disruptive opportunity or capability. They are led by a partnership between scientists, engineers and program executive offices where these groups join forces and work closely with warfighters to deliver advanced capabilities on aggressive timelines.
“Autonomy technologies are rapidly evolving, and we lack a common environment in which to test them,” Parsons said. “That makes technology readiness level assessment very difficult when you have to bounce between different simulation environments that cater narrowly to a specific technology.”
To address this inefficiency, the DTE will serve as an open environment for custom modular components that can be tailored to support various tests. While the DTE is not available for testing yet, Parsons said the assessment side will ultimately employ data strategies that vet autonomous systems for technical maturity and operational utility.
To deliver DTE to the enterprise, a team from LinQuest Corp, hired by the TCO, is currently expanding the simulation capability and deploying it via the cloud as well as expanding assessment tools and methodologies.
“The demand for simulation outstrips the supply,” Parsons said. “We need to be able to do tens of thousands of simulation runs for every flight that we intend to do. Support of developmental test is absolutely an enterprise imperative.”
Dr. Terry Wilson, director of artificial intelligence, or AI, technology development and transition for AFRL’s Autonomy Capability Team, or ACT3, presented the Autonomous Air Combat Operations, or AACO, program, which has a strong mix of government and contractor support from the Air Force Test Pilot School, the Air Force Strategic Development Planning and Experimentation, or SDPE, office, Air Force Institute of Technology, DARPA and Kessel Run.
AACO focuses on developing AI tactical autopilot behaviors capable of engaging with multiship/multirole; beyond visual range; and intelligence, surveillance and reconnaissance combat operations with proficiency at or greater than weapons school graduates.
“It’s important that we train and show value in these tasks in a digital environment and then take them into live flight,” said Wilson. “Having the ability to have a build up from simulation to live flight enables proof of value and capability in the digital environment before costly live flights. Additionally, this enables growing subject matter expertise in future warfighters while building trust in AI.”
“The bottom line is we are looking to deliver transformational capabilities in Combat Collaborative Aircraft, and we believe that combination of AI approaches is a way to do that,” Wilson said.
Approaches such as reinforcement learning, or RL, specifically deep RL, have seen a resurgence since 2017, according to Wilson, when a British company trained a system to beat world masters at the game of Go. This event sparked interest in deep-RL solving difficult real-world problems with humanlike performance.
Wilson said AACO is currently working with various organizations and teams to deploy a combination of rule-based, RL and deep-RL AI tactical autopilot agents using digital twin models on real aircraft in tactical combat situations.
One of the major hurdles is the integration of modeling and simulation capable of producing a complex environment with multiple reliable aircraft types and sensors into the training system, Wilson said. The team selected the Advanced Framework for Simulation, Integration, and Modeling, or AFSIM, as the current simulation environment and is pivoting to other simulation environments during live flights.
Maintaining flexibility in simulation is key to support various mission sets, Wilson added.
While this approach requires millions of hours of experience in high-fidelity simulated environments using high-performance computing, the result is an AI tactical autopilot agent that can learn optimal tactics and react with machine precision during an engagement.
Our modeling and simulation, AI training and evaluation frameworks are 100% government-owned, Wilson said.
The AACO team has employed numerous techniques to help bridge the performance gap of simulation-trained AI tactical autopilot behaviors when deployed to target platforms. The team deployed techniques across multiple flight test events to compare agent performance in real flight against simulations across mission sets.
AACO’s live, virtual and constructive, or LVC, architecture allows AACO to deploy AI tactical autopilot behaviors on a live airplane and to human-flown simulators on the ground. This approach allows AACO to bring human pilots in and have them engage with AI agents, Wilson said.
“It allows us to go fast with a fly-fix-fly approach to develop new AI agents and get those installed onto the airplane and then flown,” he said. “The envelope protection monitor will make sure that no command can put the aircraft in an unsafe state. Control of the aircraft will switch to the safety pilot.”
Additionally, the safety pilot can always assume control of the aircraft.
“Because of this, we can go very fast with developing, integrating and flight testing a range of AI capabilities,” Wilson said.
Live demonstrations with virtual capabilities
The test was a project under the Golden Horde Colosseum, a DAF Vanguard program that leverages challenges in virtual spaces to launch new solutions for warfighters.
As part of the demo, a team of scientists and engineers integrated multiple unmanned aerial systems in an LVC scenario
“The test approach is core to development and testing and ultimately fielding complex autonomous systems,” Newsted said.
The team used TACE, or testing autonomy in controlled environments, software to integrate the aircraft with the LVC scenario. Algorithms were hosted on hardware, and the team ultimately delivered test results for three different autonomies.
“The ability to rapidly deploy the performer agents in the field was game changing,” he said.
Experimentation and flight testing
Lt. Col. Robert “Wikid” Waller from the 40th Flight Test Squadron, or 40 FLTS, Eglin Air Force Base, Florida, highlighted lines of effort related to autonomy technology development in the air domain.
“The idea here is that the experimentation phase gives us the opportunity to quickly on-ramp and off-ramp software capabilities inside of these platforms to really see what the art of the possible is, as well as start working into actual tech development for systems that show promise,” Waller said.
The Eglin Test and Training Complex, where the 40 FLTS operates, covers 134,000 square miles and reaches 99,999 feet in altitude. It has significant telemetry coverage across the panhandle of Florida for data link and weapons testing related to autonomous combat platform efforts.
That partnership is testing two XQ-58s out of a newly constructed facility on Florida’s Gulf Coast for launch and land-based recovery at bombing ranges. Verification and validation phases will continue in 2022 through 2023, and the team will use the XQ-58 as a surrogate platform for future autonomy efforts.
Waller also explained the Test Flag Enterprise based on the Air Force flag construct or Red Flag. The Test Flag Enterprise is a safety-defined paradigm for rapid experimentation or technology development.
If a technology is under a test effort and fits within the safety framework at Eglin, it can be ramped up into the recurring Emerald Flag test series, he said.
Orange Flag refers to events that take place at Edwards Air Force Base, California, or the Western Ranges. Emerald Flag events happen three times a year in Florida’s Gulf region. Black Flag events occur at Nellis Air Force Base, Nevada, or at the Eglin range.
Early in the presentation, Waller addressed some of the challenges in communication that can result in a highly technical field.
“Lines of communication can break down when people are using terminology that can mean different things to different people depending on their backgrounds and experience,” Waller said.
People can use the same words and mean completely different things, he added. This can be especially prevalent in a high-technology environment; therefore, it is important to define terminology when using highly technical jargon.
Waller also discussed the difference between traditional flight testing versus testing with autonomous platforms.
“A lot of time and effort is spent navigating the safety frameworks that are predicated on the assumption that the vehicle itself is what is being tested and the aircrew must be kept safe,” Waller said. “Under the autonomy testing paradigm; however, the vehicle itself is a means to do the testing, not the object of the testing itself, and more emphasis is on the autonomy systems.”
Collaboration with industry
“The end goal is to increase autonomy capabilities in the future force,” Meagher said. “Prime makes this happen by tapping into commercial markets for Department of Defense needs.”
AFWERX recently collected responses via a request for information to assess how autonomous technologies being developed could support military needs.
“We want to leverage the work that companies are already doing,” Meagher said.
Autonomy Prime will build off the lessons learned in the Agility Prime program, an ongoing partnership with companies developing electric vertical take-off and landing, or eVTOL, aircraft. Agility Prime has advanced the eVTOL market by facilitating agreements with the Federal Aviation Administration and by connecting firms with test facilities.
“There’s a lot of commercial investment that’s going on in this space,” Meagher said. “However, some of the companies may not have all the access to the test resources and the expertise that the DOD has. So, part of our program is to assist and offer any kind of testing through our partnerships.”
Just as AFWERX did in the Agility Prime program, Meagher said Autonomy Prime will support companies’ testing needs using a purpose-built fleet that facilitates iterative testing. AFWERX will extend proving grounds to partnering firms for testing autonomous technologies with resources including people, places, processes and data.
“By offering proving grounds, we promote rapid autonomy development and iteration capability as well as increased flight development and capability transitions,” Meagher said.
Lt. Col. Larry Ware, the AFRL Autonomous Collaborative Enabling Technology, or ACET, portfolio lead, discussed how AFRL is aligning its tech portfolio to support the Autonomous Collaborative Platforms, or ACP, mission space. Ware spoke on the need to deliberately integrate the science and technology community, unify warfighters and acquirers and work toward a common vision with stakeholders and partners.
“[W]e are trying to look at the data capability of all these mission sets and all of the aspects that are required to support all of them and how they play on each other as we are building out the technologies,” Ware said.
Ware highlighted the commander’s intent and its emphasis on constructing capability road maps within the lab. He said AFRL is ensuring the various directorates move in the same direction and align technology development plans. The goal of this unification is to deliver capability faster and more affordably.
An integrated AFRL ACET program will influence the lab’s priorities, including reducing platform production costs and time to market. Impacts would also be felt within sensor capability and affordability, autonomous control and collaboration, onboard processing, communications, weapons capabilities and integration, human-machine interfaces and more. Every area of the lab would feel the effects, according to Ware.
“[T]here’s a lot of things going on in the lab, and we are really trying to make sure that we are appropriately aligned, and we have some good solid objectives out there … in order to get things out the door quickly,” Ware said.
Trenton White, flight demonstration program manager, Off Board Sensor Station, or OBSS, discussed how this program leverages best practices from the automotive industry. As part of an effort called Low Cost Attritable Aircraft Platform Sharing, or LCAAPS, OBSS is testing the feasibility of producing several aircraft variants from a single chassis or framework.
“This represents an alternative acquisition approach for autonomous collaborative platform aircraft,” White said. “Using this product line philosophy, every few years you can evolve, develop, build and fly a new vehicle variant based on this platform sharing approach.”
White said platform sharing offers faster development, lower costs and greater opportunities for frequent technology refresh.
“This is really about being smart about flight tests rather than applying the one-size-fits-all approach,” Mahoney said.
Blackburn discussed the importance of running full system tests as quickly as possible. Testing early saves money and time, he said, noting how early tests enable teams to learn from issues and achieve success in the end.
“It’s not so much about changing risk management,” he said. “It’s just changing the scope of your risk management where now maybe you are looking at the risk of time.”
Blackburn said testing early does not mean a team is irresponsible or impatient.
“They are just trying to learn stuff earlier,” he said. “If you include schedule as a major risk item, that changes how you would test.”
Blackburn noted he and Mahoney are partnering with the Air Force Institute of Technology to apply this approach to pilot programs and analyze its application. The goal is to improve future recommendations.
The Air Force Research Laboratory, or AFRL, is the primary scientific research and development center for the Department of the Air Force. AFRL plays an integral role in leading the discovery, development and integration of affordable warfighting technologies for our air, space and cyberspace force. With a workforce of more than 11,500 across nine technology areas and 40 other operations across the globe, AFRL provides a diverse portfolio of science and technology ranging from fundamental to advanced research and technology development. For more information, visit www.afresearchlab.com.