Perception, decision-making needs for unmanned resupply on demand push autonomy research
Unmanned cargo helicopters have been proven in Afghanistan, resupplying remote forward bases and taking convoys off dangerous roads. But they fly carefully preplanned missions, or require a trained operator to take control at the drop zone.
The vision is for a machine that can be called up on demand, without specialist training, to negotiate obstacles and threats, select a safe landing zone and unload supplies or evacuate casualties—all autonomously.
This is the goal of the Office of Naval Research's (ONR) Autonomous Aerial Cargo Utility System (Aacus) program, which will stage its first flight demonstrations in February and March next year. Aacus is developing a sensing and computing capability that will be portable between platforms, enabling any fly-by-wire helicopter to become an autonomous logistics machine.
Aurora Flight Sciences andare working on Phase 1 of Aacus under 18-month contracts awarded in October 2012, totaling $28 million. Aurora plans to demonstrate its system on 's H-6U Unmanned Little Bird. Lockheed will use the Kaman K-Max, now flying unmanned cargo missions in Afghanistan as well as research flights for the U.S. Army.
Critical design reviews were completed last week. “I am very positive about the program, which you cannot say about all science and technology programs,” says Roger McGinnis, ONR's Aacus program manager. “We are seeing them progressing quickly, with a lot of success. In simulations, the algorithms are working quite well. And the different hardware is coming along well.”
Aacus has its origins in Afghanistan, where improvised explosive devices generated the need to get convoys off roads and deliver supplies by air, while reducing the workload and danger for helicopter crews. The result is a program pushing the frontiers of airborne autonomy while potentially spinning off capabilities to the existing fleet.
For the Phase 1 fly-off planned for early next year at the' Quantico base in Virginia, ONR wants an untrained field operator—“not an aviator,” says McGinnis—to be able to request resupply using a tablet or mobile device. The “simple formatted request” will go to a main operating base some distance away, which will launch an aircraft in response.
The unmanned helicopter “will work its way through cleared airspace to the field operator,” he says, “then it has to determine if the landing zone requested by the operator is a safe place to land.” This means assessing the slope and surface and detecting any obstacles, such as power lines. “This is easy for a pilot, but it is hard to develop perception technology for a computer to decide.” Perception is the hardest part, he says. “What does a computer do with a 2-D image? That is challenging.”
The requirement is to make an angled approach to a landing at significant speed, not to hover while searching for a safe place to touch down, and to use onboard sensors. “We don't want it to slow down and loiter, we want to be on the ground within five minutes,” says McGinnis. “We want it to get on the ground, avoiding things it detects or where people on the ground say 'Don't go there'.”
Accomplishing the demo will require a tightly integrated suite of sensors, processors and algorithms. “It's a very challenging first demo,” he says. Each contractor's demonstration will involve a series of flights of increasing complexity. “We'll start simple, with no obstacles, then work up to obstacles the aircraft should detect and things like no-fly zones that we tell it to avoid.”
Key goals of Aacus are cost and making the system portable. “We want it to be aircraft-agnostic. We want to take it out of the first helicopter and in a short time put it into another helicopter and have it work as well,” McGinnis says. This is planned for Phase 2; the system is to be installed on theJUH-60A Rascal, a Black Hawk fly-by-wire testbed, for flights in February 2015.
The system must weigh less than 30 lb. “The beauty of this is it has to fit in a helicopter. It can't be too bulky to fit easily in an aircraft, with a lot of margin left over to deliver goods,” he says.
Two further phases are planned under Aacus, which will extend to 2017. Yet to be fully defined, these will involve moving obstacles plus weather and winds, possibly launching from a ship and other capabilities of interest to potential customers, such as casualty evacuation. “There is a lot of sensitivity around this mission, and robotic reliability will have to go way up,” says McGinnis.