Science of Autonomy

Basic Research

Science of Autonomy focuses on multidisciplinary research topics in autonomy that explore interconnections between fields such as artificial intelligence, machine learning, control theory, human factors, biology, cognitive science, psychology, economics, operations research, applied mathematics, oceanography, physics and neuroscience. The focus of the program is on new methods that are mathematically rigorous and/or grounded in general scientific principles.

Research Concentration Areas

Research in Science of Autonomy should be of broad implications and not specifically focused on unique needs of one type of system or platform. However, it should also have strong naval relevance to problems involving factors such as:

• Long deployments in spatially and temporally variable and uncertain environments with limiting staffing, communications and other resources
• Robustness to and prediction and exploitation of unique aspects of ocean and littoral environments
• Users with a wide range of skills and experience
• Diverse environments and systems
• Domains with low bandwidth, noisy, short-range and intermittent communications
• Complex missions with heterogeneous systems, including significant differences in on-board autonomy, mobility, sensing and communications capabilities
• Rapid and dynamic responses to user needs and changes in the operating space
• The need for automation to explain its capabilities to the user and reliably execute the required tasks in the required time

The focus of the program is on autonomy methods and not on the development of advanced platforms, sensors or communications hardware.

Research Challenges and Opportunities

• Scalable, self-organizing, survivable, organizational structure/hierarchy of heterogeneous unmanned vehicles appropriate to naval mission domains. Of particular interest are mathematically rigorous and biologically inspired methods that are robust to harsh and uncertain environments, and can be tailored to perform complex tasks under realistic assumptions on perception and communications. Also of interest are methods for achieving more sophisticated decentralized forms of learning, perception and intelligence.
• Autonomous learning, reasoning and decision-making in unstructured, dynamic and uncertain environments. Of particular interest are architectures that bring together methods seeking higher levels of autonomy and involving broad sources of knowledge with flexible and robust perceptual and behavioral/control processes. Also of interest are methods suitable for systems deployed on time scales for which the mission time is much more than the validity of a priori assumptions and information, including perishability of a priori knowledge; mission times on scales that make it likely that problems will be encountered, which were not foreseen in typical failure management-and-effects analyses, changes in goals, mission constraints and priorities; and where it may be difficult to judge mission progress.
• Human interaction/collaboration, including new human-autonomy function divisions and relationships/roles and advancing bi-directional human/autonomy communication and comprehension. Of particular interest are methods that enable new divisions of roles and responsibilities between humans and autonomy, such as teams; the development of new scientific models of human/autonomy interaction; and human collaboration with advanced autonomy methods such as machine learning, autonomous-teaming skills, complex reasoning and decentralized, scalable systems.
• Perception and intelligent control/decision-making. Of particular interest are methods that can autonomously adjudicate between wide-area exploration and exploitation, are robust to naval environments, can make fast, satisficing decisions with limited information, and are appropriate to long-duration deployments with continuous learning and adaptation.
• Methods supporting verification and safe autonomy with advanced methods and in complex unstructured/uncertain naval environments.