Demos Show Rapid Advance In Autonomous Rotorcraft
By Graham Warwick
Source: Aviation Week & Space Technology
April 21, 2014
Credit: U.S. Navy
Among flying machines, helicopters would seem to require more piloting than most, as they operate at low altitude, close to terrain, obstacles and other hazards. But research into applying autonomy to rotorcraft is making rapid progress, with several key unmanned-helicopter demonstrations in recent weeks.
On March 11, Sikorsky and the U.S. Army showed that a minimally trained ground operator could command an optionally piloted UH-60 Black Hawk to pick up, move and set down an external load. Over two weeks in February, Aurora Flight Sciences, Lockheed Martin and the Office of Naval Research (ONR) showed that unmanned helicopters could respond to a request for resupply, avoiding obstacles and selecting safe landing sites autonomously.
Additionally, the Sikorsky Autonomous Research Aircraft (SARA) has received its optionally piloted vehicle (OPV) experimental certificate from the FAA, allowing the modified S-76 to begin unmanned flight tests under a company-funded program to develop autonomy technology for vertical-takeoff-and-landing aircraft that can be FAA-certified.
The rapid technical progress in helicopter autonomy is being fueled by the promise of increased mission capability, safety and reliability, and reduced pilot workload. “Rotorcraft operate at low altitude in an obstacle-rich environment. It is very challenging, but if we can bring autonomy and optionally piloted capability to helicopters, we will get the highest returns,” says Mark Miller, vice president of research and engineering at Sikorsky.
Under the Manned/Unmanned Resupply Aerial Lifter (Mural) program with the Army, Sikorsky flew its optionally piloted Black Hawk demonstrator on a cargo logistics mission at its test site in West Palm Beach, Fla. A flight engineer with minimal training, using a backpack ground control station, was able to run the mission from start to finish, says Igor Cherepinsky, chief engineer for autonomy.
“The person running the mission did not care that it was a helicopter,” he says. The ground operator maneuvered the Black Hawk to pick up the 5,000-lb. load with point-and-click commands and a “coarse beeper” joystick similar to that used by a crane operator. With the load attached, he then entered a simple flight plan and pressed a button to send the Black Hawk to its destination, where another operator took over, repositioned the helicopter and disconnected the load.
A UH-60MU fly-by-wire (FBW) test aircraft was modified with the data link and software required for optionally piloted operation. A downward-looking belly camera was added to give the ground operator a view of the load. “We designed the interface without the pilot in mind,” says Chris Van Buiten, vice president of technology and innovation at Sikorsky. “The operator was just doing the cargo crane part. It was transparent to him how the aircraft flew.”
The next step is a 2015 demonstration of manned-unmanned teaming between the optionally piloted Black Hawk and a large unmanned ground vehicle (UGV) developed by Carnegie Mellon University. The two machines will collaborate to deploy, operate and retrieve the UGV, says Van Buiten.
ONR's Autonomous Aerial Cargo Utility System (Aacus) program aims to take autonomy all the way to landing. Under Phase 1, which ends this month, Aurora and Lockheed Martin demonstrated that a minimally trained field operator could request a resupply flight via handheld tablet, and an unmanned helicopter equipped with the Aacus package would respond, navigating to the requested location, avoiding no-fly zones and obstacles, and selecting a safe landing site autonomously.
In demos at Quantico, Va., Aurora flew its Aacus package on Boeing's H-6U Unmanned Little Bird, while Lockheed used the Kaman K-Max already performing unmanned cargo missions in Afghanistan. Both helicopters were fitted with electro-optical/infrared (EO/IR) and three-dimensional lidar sensors, autonomy processors performing onboard path and trajectory planning, and interfaces to the flight controls.
Where other demonstrations of autonomous landing-zone selection have allowed the aircraft to overfly and survey the site before landing, Aacus is required to evaluate the requested location and, if necessary, select safer alternates as the aircraft flies in. Aacus is intended to work with any helicopter, regardless of size or performance, but the ultimate goal of being able to land within 2 min. from 5 nm away means the system must be capable of selecting a safe landing site while approaching at over 200 kt., says Jon McMillan, Lockheed vice president for K-Max business development.
“Time is a function of the helicopter, not the Aacus system,” says Max Snell, ONR program manager. “But we constrained the response time so they would not hover and stare at the landing site. We did not want to give an inordinate amount of time to the perception system. We wanted them to proceed and approach as would a piloted aircraft.”
In the demo scenario, the field operator used a pull-down menu on a tablet to send a request for a resupply flight to the main operating base. There, the ground control station uploaded the mission plan to the aircraft, which took off and switched to autonomous mode. At the start of final approach, the helicopter autonomously contacted the field operator and requested permission to land. Given the go-ahead, Aacus began assessing the landing site using the lidar.
“If there were obstacles, the system would offer up alternatives,” says Jon Wissler, Aurora director of aerospace systems. The alternatives were displayed on a map on the tablet and the operator picked one. “The operator does not have to know what the helicopter is capable of,” he says. “They can pick what's good for them, and the helicopter can say it's not interested; here are some alternatives.”
The ability to select a landing site from incomplete data is key to Aacus. “The system has to make a decision without having perfect data; that's what makes this relatively unique,” says McMillan. “It's about how you put all that intelligence on board so it can land in the presence of uncertainty.”
The system's world model is continuously refined as it approaches the landing zone and sensor data improves. “At long range, it uses lower-resolution two-dimensional EO/IR to get an idea. As it gets closer, it builds a higher-resolution model using 3-D lidar to get a detailed map,” he says. But the result is a probabilistic view of the world. “The onboard autonomy has to deal with the uncertainty and reduce it as the aircraft comes in to get to a point where it can decide to land.”
At Quantico, the helicopters navigated autonomously around no-fly zones, both those entered before launch and those sent by the field operator while the aircraft was inbound, indicating pop-up threats and requiring the aircraft to replan its approach. “This is real-time trajectory generation, not just waypoint to waypoint,” says McMillan. “This is real-time sense-and-avoid, and reacting to pop-ups.”
At least one team is expected to go ahead to Phase 2 of Aacus. The next step is to expand capability and test in more stressing environmental conditions. The goal is to improve the capability to detect and avoid small obstacles, including power lines, to operate in different types of obscurants and precipitation, and demonstrate the system's ability to function in a GPS-denied environment. The next flight demo is expected in 2015. A third phase would involve porting Aacus to a different helicopter to prove that it is platform-agnostic, as intended.
Sikorsky, meanwhile, is taking a similar platform-agnostic, open-architecture approach to its company-funded Matrix Technology autonomy development program. Both the S-76 SARA and optionally piloted Black Hawk have similar FBW systems, and the next phase of the Mural program will see the autonomy high-performance computer used in SARA installed in the optionally piloted Black Hawk.
SARA first flew in July and initially focused on testing the full-authority FBW system installed in the S-76 testbed. “At the end of 2013 we entered Phase 2, involving autonomy, artificial intelligence and perception,” says Cherepinsky. “Several [EO and lidar] sensors were installed in the aircraft to provide all-round obstacle avoidance and situational awareness.”
The S-76 then flew “open loop” to collect sensor data and test the discrimination algorithms. “Now we have our FAA OPV ticket and are about to start flying the algorithms closed-loop,” he says, meaning the autonomy system will fly the aircraft. Flights are planned over Connecticut and New York, where there is “interesting terrain” ranging from forests and mountains to lakes and ocean.
The plan is to demonstrate autonomous cargo, medevac and other missions. These will involve SARA taking off from an airport, avoiding traffic and obstacles and behaving like a piloted aircraft, flying into an area with a variety of terrain and hazards, and arriving in a constrained area where a user is waiting. The aircraft will fly toward the requested location, scout for landing zones, communicate any constraints to the user and be redirected or allowed to land.
“We make the machine aware of what the mission is,” says Miller. “If it is medevac, they have to bring the casualty to the aircraft, so a viable landing zone has to allow access. The system takes that into account when evaluating landing sites.”
Piloted flights have shown the system can pick suitable sites. “SARA has flown over, selecting landing zones open-loop in an obstacle-rich environment. Then the pilots go over, select sites and would agree with the system,” says Van Buiten. “In some cases, flying open-loop at 1,000 feet, the pilots would say no way that's a landing zone, then they would drop down and see they really could land there,” says Cherepinsky.
Sikorsky is working closely with the FAA with the goal of developing autonomy that is certifiable. Autonomy produces behavior that is emergent, not deterministic, and cannot be predicted with the certainty required by certification authorities. “Most of the algorithms have emergent behavior, and there is no way to certify them by conventional methods,” he says.
The company's approach bounds the emergent behavior of the high-level intelligence algorithms with low-level, deterministic intelligence that can be certified conventionally. “The boundaries are not just spatial, they are also temporal, and involve an understanding of airspace and traffic,” says Cherepinsky.
Sikorsky is not waiting for an opportunity to develop an unmanned rotorcraft and has already received FAA certification for a rig-approach “autonomy app” for the S-92. “We are looking for the next set of apps. We are thinking of one for the Black Hawk to do the cargo mission. We do not have to have fly-by-wire to bring on FBW-like apps,” Miller says. “We think there are fairly near-term opportunities to field these kinds of apps, commercial and military.”
Watch videos of the two Aacus demonstrations and see Sikorsky's S-76 SARA testbed fly at: ow.ly/vUduc