Air Force seeks autonomous systems that learn and adapt
Connecting state and local government leaders
As adversaries continue to bolster their technology-based capabilities, the Air Force seeks resilient systems to thwart disruptions and sustain operations.
As the military has begun assessing operations in contested environments against technically sophisticated adversaries, resilient capabilities have become a larger priority.
To address the need for systems that can operate in contested environments, the Air Force is looking to advance machine intelligence by developing and testing resilient autonomous systems that can assess, recover, learn and adapt from adverse events, both benign and hostile. A new broad agency announcement for a four-year, $7 million Resilient Autonomous Systems program calls for research proposals that will demonstrate the benefits of improved resiliency in a simulated but realistic Air Force domain.
The project will focus on activities related to machine intelligence without a human in the loop, with resiliency being demonstrated by a solution’s performance under increasingly adverse conditions and through multiple iterations of missions. Technologies must provide effective command and control of at least 100 fully autonomous assets within intermittent and degraded communications environments as well as prove capable of learning from prior experiences.
In order to assess the capabilities of technologies presented, solutions will participate in simulations in which opposing forces, known as red teams, will attempt to disrupt the specific intelligence, surveillance and reconnaissance missions within a 72 hour window. Proposers must develop and implement a fully autonomous solution capable of planning, coordination and control of the friendly, or blue team, forces.
Possible approaches that could enable resilient autonomous solutions include:
- Distributed planning and constraint optimization
- Multiagent coordination
- Distributed information management
- Representation and feature learning
- Reinforcement learning
- Transfer learning
- Case-based reasoning
- Game theory and opponent modeling
White papers will be accepted for 2016 funding until March 30, 2016.
NEXT STORY: Fingerprinting from a distance