https://www.onr.navy.mil/-/media/Files/Funding-Announcements/BAA/2021/N00014-21-S-F003.ashx
Topic 25: (ONR) Systems-Level Foundations for Agile, Dynamic, and Ad Hoc Human Autonomy Teams
Background: There is currently little agreement on what a Human Autonomy Team (HAT) is, how it differs from other types of organizations, and when it would be the most appropriate, robust, and effective framework to solve particular classes of problems. Much research on HAT has focused narrowly on the real-time task performance of relatively small teams performing structured and short time duration tasks. Importantly, it has not examined longer duration longitudinal effects and the dynamics of how HAT members learn to work together, nor considered that effective teams require the ability to jointly train, rehearse, plan, and make agreements while working together in order to create and maintain common ground, assess performance, and improve together afterwards. This is also critical for ad hoc collaborations in which minimally trained humans and minimally tailored machines may be encountering each other for the first time and must adapt and learn as they go. Further, the true value of HAT may lie in exploiting the extreme level of heterogeneity possible between humans and future intelligent systems to create wholly new types of organizations rather than trying to mimic fully human ones or force humans into the rigid frameworks of multi-agent machine systems. Thus, one framework for HAT is (1) Teams are set up to achieve common goals believed achievable in a bounded period of time, without requiring that every member have the same understanding of the goal; (2) Teams exploit role specialization and have strong bi-directional interdependencies between teammates; and (3) Individual identities of teammates matter, allowing for unique relationships between teammates in flexible control hierarchies with dynamically changing roles, responsibilities, and functions. Other frameworks for HAT exist in the literature and similarly emphasize common goals, interdependency, boundedness of the team, and that autonomous agents should be capable of holding specialized roles recognized by their human teammates. Note that it is not assumed that all human roles would have a high degree of interaction with intelligent systems on joint tasks, but effective teammates should at minimum be able to avoid interference with each other, and ideally recognize unplanned opportunities to assist and collaborate in unexpected situations.
Objective: To develop theoretical foundations, models, and principles for the effective design of Human Autonomy Teams, including interactions with dynamic environments, extreme heterogeneity among team members, time pressure and task uncertainty. An important focus is studies of longitudinal effects that may occur over diverse time scales from ad hoc teams that learn to work together as they go to long-term stable teams that jointly train and work together.
Research Concentration Areas: The topic requires a synergistic approach across scientific fields (biology, neuroscience, psychology, sociology), complex socio-technical systems (human factors, management theory, economics/game theory) and systems methods for engineering HAT (computer science, robotics, engineering): A general theory for when team members should communicate with each other, how to recover from breaks in communication, and how human/autonomy communications differ from fully human or machine teams; Computational models of how humans process information and update their mental models in interactions with heterogeneous teammates and resulting principles for interaction design including transparency for rapid situation understanding, reasoning, and projection of future outcomes and uncertainties; Principles of when teams are the most appropriate organization, and when general (e.g., Theory of Mind, Neural or Behavioral Synchrony) or specialized task capabilities will make the most effective autonomous teammates; Dynamic models of how HAT interactions change over different time periods of team operations including joint training effects and trust modulation with high heterogeneity teams.
Anticipated Resources: It is anticipated that awards under this topic will be no more than an average of $1.5M per year for 5 years, supporting no more than 8 funded faculty researchers. Exceptions warranted by specific proposal approaches should be discussed with the topic chief during the white paper phase of the solicitation. Additional year(s) of funding may be added at the discretion of the topic chief and will be based on availability of additional funds and team performance.
Research Topic Chiefs: Marc Steinberg, ONR, 703-696-5115, marc.steinberg@navy.mil; Tom McKenna, ONR, 703-696-4503, tom.mckenna@navy.mil