Computational Modeling of Monitoring within Human-Autonomy Teams with Novel Delegations of Authority and Responsibility
Open Access
- Author:
- Villa, Amber
- Graduate Program:
- Aerospace Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- June 24, 2022
- Committee Members:
- Amy Pritchett, Program Head/Chair
Amy Ruth Pritchett, Thesis Advisor/Co-Advisor
Jacob Willem Langelaan, Committee Member
Russell Richard Barton, Committee Member - Keywords:
- human-autonomy teams
computational modelling
Work Models that Compute
Advanced Air Mobility
Authority/Responsibility - Abstract:
- Even when a vehicle can ‘fly itself,’ it needs to interact with others. These interactions should be designed to not only support the individual ‘taskwork’ needed for nominal operations, but also to support teamwork that helps detect hazards. The overall concept of operations, the structure of the human-autonomy team (HAT) that will execute the operations, and the allocation of who does what within the team and how they interact must all be designed. With radically new vehicle technology, autonomous capabilities, and proposed operations, new design principles and methodologies are needed. The collective behaviors of HAT’s emerge from the combination of the underlying work dynamics, the function allocation and team dynamics, as well as the individual agent dynamics. All of these must be modeled to completely capture the dynamics of a human-autonomy team. Methods are needed to predict, analyze, and measure the combination of all these underlying dynamics of a HAT and its work. The current reliance on human-in-the-loop testing, which can only occur later in design once the functional prototypes of technology are established and human personnel trained, leaves testing until the end of the design process where it is difficult to explore more options or make substantial changes. Computational work models have been proposed for early in the design process. To date, they have been used to analyze small human-autonomy teams including flight crew automation interactions in one aircraft or human-robots in space flight extra vehicular activities. Applying a computational modeling approach requires expansion in the models of HAT collective work that they can analyze. This thesis establishes a method of analyzing Concepts of Operation (CONOPS) of human-autonomy teams (HAT) and their collective task work and teamwork. A computational model of a proposed CONOPS for Advanced Air Mobility (AAM) was developed to demonstrate the use of Work Models that Compute (WMC) in identifying the work’s dynamics, timing constraints, and information requirements. Further work dynamics were identified via the framing of a moderately complex package delivery case study and a defined function allocation divvying up the work among a set of agents. Together with the function allocation, limitations of certain agents led to increased time to completion, while task-limits of other agents demonstrated that the inability to delay time-critical actions will lead to chronic over-saturation of that agent. A degraded battery was modeled in the case study to explore the safety-producing behaviors exhibited by human-autonomy teams. The mismatch of authority to execute an action and the responsibility to verify its outcome is modeled as triggering monitoring. Multiple points where monitoring could occur were described for this exemplar case including how the information requirements may differ for detection of degraded battery performance compared to detection of insufficient energy for an upcoming flight. The clear time constraints of detection of a degraded battery would ideally be caught before the vehicle is in flight, but risks being missed where agent task load delays their monitoring. This thesis demonstrates that concepts of operations can be translated into a computational model and further developed through simulation utilizing the Work Models that Compute (WMC) framework. The work model can be analyzed for work dynamics and to identify timing constraints and information requirements of the actions comprising the work. Additionally, the impact of function allocation on the team and task work dynamics can be measured by the task load and information requirements on each agent as well as the overall completion time. Third, the ability to generate monitoring actions during runtime based on mismatches between authority and responsibility in the function allocation allows analysis of the task load and information requirements of basic, full and extended forms of monitoring. The computational model and simulation allows analysis of the emergent system dynamics of a wide-range of HAT structures as represented by the number of agents, their assumed capability and task limits and the allocation of authority and responsibility across the team.