Control science tops list of USAF science and technology priorities

Washington DC
Source:
This story is sourced from Flight International
Subscribe today »

If the chief scientist of the US Air Force is correct, the key technology challenge for airpower over the next two decades is not directed energy, cruise missile defence or even satellite-killing weapons.

In a sweeping new 153-page report Technology Horizons, USAF chief scientist Werner Dahm instead identifies advances in "control science", an obscure niche of the software industry, as potentially the most important breakthrough for airpower between now and 2030.

Control science develops verification and validation tools to allow humans to trust decisions made by autonomous systems, which, Dahm writes, must make huge leaps in capability over the next decade for the USAF's budgets to remain affordable.

Although too primitive to unleash the inherent power of modern autonomous systems, Dahm's report could make control science a major funding priority for at least the next 10 years.

Dahm's staff delivered the report - the seventh in a series of USAF science and technology roadmaps dating to 1945 - to chief of staff Gen Norton Schwartz in May, although a public version of the document was not released until 21 July.

Beyond autonomous tools, Dahm forecasts the Air Force Research Laboratory will make major advances in the efficiency of conventional turbofan engines, as well as create a new class of hypersonic aircraft and missiles.

Dahm also recommends a new emphasis on cyber threats, especially developing solutions to the risks posed to the USAF's reliance on the satellite-based global positioning system data for routine operations.

MACHINE RULES

But the "most essential insight" in the report, according to Dahm, is the need to reduce costs associated with manpower. Dahm's report recommends launching a major effort over the next decade to develop the tools that transfer more decision-making capability from humans to software-driven machines.

Despite billions spent on developing unmanned aircraft systems over the previous decade, Dahm believes the USAF has only scratched the surface of autonomous capability. Indeed, the USAF prefers to use the term "remotely piloted aircraft" (RPA) rather than UAS, as the MQ-1 Predator and MQ-9 Reaper are actively controlled by pilots on the ground.

According to Dahm's report, the major barrier preventing the USAF from gaining more capability from autonomous systems is the lack of validation and verification (V&V) tools. The USAF has no way to trust that autonomous systems make decisions better than humans in the same situation.

"The lack of suitable V&V methods today prevents all but relatively low levels of autonomy from being certified for use," the Dahm report says.

Increased autonomy is not only sought for UAS operations. The USAF also could increase autonomous decision-making within the combined air operations centre, which directs all aspects of an air war. The centre is now staffed by hundreds of people.

Meanwhile, USAF officials have sought to educate the public that RPA fleets also require a perhaps surprising level of manpower to operate.

Last November, Schwartz told an audience of international air chiefs in Dubai that a single combat air patrol - a term meaning a capability to maintain a continuous orbit over a given area - comprised of four MQ-9s requires 185 people.

This staff includes seven pilots, seven sensor operators and seven maintainers based in Nevada, 59 people assigned to a launch and recovery element in theatre and 83 people assigned to either analyse or manage the data collected by the aircraft in flight.

With US defence spending expected to be squeezed by budgetary pressures, Dahm's report urges the USAF to focus science and technology priorities on reducing manpower costs with automation. That goal is impossible without advances in validation and verification capability, the report says.

"Key attributes of such autonomy include the ability for complex decision making, including autonomous mission planning, and the ability to self-adapt as the environment in which the system is operating changes," the report says.

Lee Pike, who co-authored an academic paper on validation and verification for autonomous UAS as a researcher for Oregon-based information assurance specialist Galois, says building trust in such software-driven systems is impossible using normal aerospace testing methods, where each potential fault is exhaustively tested.

Since the conditions for autonomous decision-making are infinite, "the time it would take to exhaustively test would be longer than the universe has existed", Pike says.

Unlike aerodynamic modelling, software modelling is discrete rather than continuous, he adds.

"If you change the value of a single bit or condition then the whole behaviour of the software can completely change," he says.

However, in an ironic twist, Pike adds, control science has produced validation and verification tools that could in theory provide confidence in more autonomous systems, but the tools themselves are not yet trusted by the aerospace community as reliable.

"While they are not foolproof," Pike says, "they are orders of magnitude more advanced than we do today."