On a frigid February day, there is no sanctuary from the biting winds at Wright-Patterson AFB in Ohio. Yet even as the Midwestern gusts whip across vast fields and over immense hangars, the US Air Force’s colossal C-17 Globemasters soar over the base with ease. As the gargantuan cargo aircraft circle overhead, it is difficult to imagine that 100 years ago, and just a few miles southwest at McCook field, researchers at the original incarnation of what is today the Air Force Research Laboratory were concocting plans to weaponise a wooden flying machine.
The Air Force Research Laboratory’s name and its base have changed since 1917, but its mission to spearhead defence aviation technology has remained the same. Today, the service’s research arm is at the forefront of developing the autonomous systems and UAV concepts that form the foundation of the Department of Defense’s third offset initiative. At the same time, it continues to plug away at technologies that, for decades, have been five years from fruition.
As evidence that persistence can pay dividends, hypersonics are in favour today on the back of success with programmes such as the Boeing X-51 scramjet demonstrator. Indeed, even as the USAF has scrambled for dollars over the last eight years, the DoD has heavily protected funding for hypersonics research. Morley Stone, AFRL’s chief technology officer, told FlightGlobal during a February visit to Wright-Patterson that part of that financial support came through a partnership with the Defense Advanced Research Projects Agency, which shared both funding and knowledge with the USAF.
The ARFL's work includes testing an automated collision avoidance system
Carla Thomas/NASA
Now, the AFRL is turning its focus to another technology with great challenges but huge promise: autonomous systems. The crown jewel of the laboratory’s autonomous flight research is its Loyal Wingman programme, a concept that aims to multiply a manned fighter’s capabilities by teaming it with an autonomous jet. Loyal Wingman promises to transform unmanned aircraft by jumping from routine automation to a platform capable of decision making. Though not part of its short-term demonstration plan, the advanced reasoning ambitions central to the Loyal Wingman concept will be among the laboratory’s greatest challenges.
“The litmus test of an autonomous system is the ability to handle the unplanned event,” Stone says. “I would perceive that as the biggest gap that needs to be filled… making sure we’re seeing the appropriate response to that unplanned event.”
Compared to Loyal Wingman, Boeing’s unmanned research scramjet represented a simple demonstration, according to Stone. Whereas X-51 flew as an unmanned asset, Loyal Wingman will pair a manned fighter with a UAV. When a human is brought into the testing environment, the risk calculus and level of complexity increases along with the demonstration cost, he says.
The laboratory is working on a demonstration with another government agency, though Stone declined to identify which entity. By fiscal year 2019, the laboratory will begin testing Loyal Wingman payloads on small-scale, 2.4m (8ft) UAVs, says Kris Kearns, autonomy lead at the AFRL. The laboratory will develop algorithms, test them in simulations and then test components on small vehicles before refining and integrating additional capabilities, she adds. A capstone flight activity demonstrating a manned-unmanned teaming strike mission in a contested environment is scheduled for fiscal year 2022, although this could be brought forward.
Stone notes that DoD budget signals coming from Congress are encouraging, so the AFRL is looking at ways to accelerate the programme should more money be made available.
The AFRL is developing Loyal Wingman’s software, not the aircraft, but eventually the laboratory will move the programme’s software from a small UAV to a larger platform with a tactical profile. Based on the results of a broad agency announcement, the USAF could integrate the software on a platform offered by industry or an existing aircraft within the service’s own inventory.
The group is also studying the F-35 programme to understand what tasks could be shifted from a manned platform to the autonomous wingman, Kearns says. By measuring how well pilots perform during different scenarios and the difficulty of juggling various tasks throughout a flight, it should be possible to define baseline expectations for Loyal Wingman, so programme managers can develop technologies to better support the pilot.
“There will always be a pilot in the decision of target identification and any use of deadly force, so that will always remain on the pilot’s task list,” she says. “But there are lots of decision support aids that we can create to help that pilot understand and make that decision on whether he thinks that target is what it is and how to execute it.”
While the AFRL will not create new communications or sensors for Loyal Wingman, the laboratory is tracking the commercial sector’s sensor development. In its search for cutting-edge sensor technology, the air force is surveying innovation on the ground rather than in the air. The automotive industry is investing more in autonomous technology than any other sector, Stone says. That industry’s focus is contributing to a drop in sensor costs and speeding up algorithm development, but it is also sucking up talent that would otherwise be available to work on the AFRL’s projects.
Sub-tier suppliers provide the smarts to car manufacturers, but some of those suppliers have also partnered with the air force on machine learning technology that transcends automotive and aerospace platforms, Stone says. Whether it is a Mercedes Benz or a Lockheed Martin F-16, both platforms have a human operator who will suffer from fatigue. Today’s cars are capable of monitoring eyes, seat shifting and steering wheel movement to identify a tiring driver. As the air force and DoD talk about human-machine teaming, it is important to have a model that understands human fatigue and its consequence on operator performance.
“The system needs to understand, for example, what are the weaknesses in this case and what does that look like in its human counterpart so it can correspondingly compensate or suggest courses of action that can help ameliorate that weakness,” he says.
COLLISION AVOIDANCE
Much of the automotive industry’s work on unplanned events has focused on mapping roads, but technology is beginning to explore abnormal operating conditions, Kearns says. As UAVs approach autonomous operation in the same airspace as manned platforms, the AFRL must focus on air collision avoidance. The USAF has already fielded the automated ground collision avoidance system (auto GCAS) on the Lockheed F-16 and has tested an integrated air and ground collision avoidance system (ICAS) at Edwards AFB in California for the past year. Air and ground recovery involve two different manoeuvres, says Amy Burns, automatic collision avoidance technology programme manager at the AFRL. Auto ACAS (air collision avoidance system) was set up specifically to work in a training environment where the USAF finds the majority of mishaps occur, Burns says.
Auto GCAS employs one recovery manoeuvre to save the pilot’s life from an imminent ground collision: roll to wing level with a 5g pull. But GCAS was built on the premise of a man inside the cockpit, so the USAF must develop an algorithm for an autonomous system that can continue flying the aircraft following the recovery manoeuvre, Kearns says: “You need to make sure that the algorithms are set such that you don’t get into another downpull into the ground.”
The USAF began its first phase of flight tests for ICAS this past year and is planning a second phase of flight tests in April at Edwards. ICAS has been tested using air combat manoeuvring instrumentation pods, which provide the datalink to connect co-operative aircraft and allow both aircraft to perform a recovery manoeuvre. The AFRL has developed a co-operative solution where both aircraft can manoeuvre, but the laboratory also created a non-co-operative solution where a pilot could avoid other aircraft in the area if the pilot can locate them.
“If you can get a track file on them, then you can develop a recovery manoeuvre away from them,” Burns says.
The system is constantly ingesting datalinked information on position, air speed and velocity, which generates a track file. The aircraft can then predict where another aircraft will be at a specific point in time and decide whether it will collide.
“It can take in 19 different track files, sort them and pare it down to your highest threat that you would collide with closest to you and that’s the one you would recover away from,” Burns says. The aircraft would use radar to identify non-co-operative aircraft, which do not have access to the datalink.
Between phases 1 and 2, the AFRL tweaked the algorithm for the air collision avoidance piece in formation flight.
“We want to allow the other aircraft to rejoin and fly formation with the lead, but we don’t want them to collide with the lead,” Burns says. “We’re working to define a region that’s acceptable to pilots…. If they’re in a certain closing rate a certain distance away then the system goes to standby and lets them rejoin with the lead.”
When the AFRL tweaked the system in phase 1, the laboratory also made changes to the fielded GCAS. The researches found that when pilots performed a split S manoeuvre, which typically pulls more than 5g, the GCAS pulled the aircraft out because the system was predicting a 5g recovery. The AFRL went back to the design and changed the split S routine so the system pulls 7g to give pilots more room before the system takes over.
Avoiding moving aircraft in the air adds a great deal of complexity to ICAS, Burns says. Where GCAS employs one 5g pull manoeuvre, auto ACAS employs nine recovery methods to navigate around other moving aircraft. That includes seven different roll and pull moves, a manoeuvre that pushes the nose of the aircraft under the other aircraft and a maintain manoeuvre that locks pilots out of their current activity until they have avoided the interfering aircraft.
Researchers have also found that the aircraft may prematurely pull up if it believes it is approaching the ground. The system assesses digital elevation data by taking a circular scan of the ground and then chooses the highest point as an altitude reference for recovery. The AFRL is working to change the system so it no longer uses a high point as the floor, Burns says. The system would still assess a peak in the distance, but would not use that point as the baseline.
“We’ll just cover the part that’s close to that mountain with that altitude and then we’ll do another sampling that’s closer in for the highest point in that region, almost like an inverted wedding cake,” she says. “They would do a sampling in the middle and then another so those would make your terrain profile so it wouldn’t constantly pull the aircraft out at altitudes the pilot did not want it to be pulled out at.”
Source: FlightGlobal.com
