Mission planning for coupled human-robot teams using adaptive human performance models

Abstract

The increasing prevalence of autonomous robots in a variety of domains has motivated the development of task allocation and scheduling algorithms which enable cooperative multi-robot teams to execute missions in a coordinated fashion. These missions often entail completing discrete tasks in distinct locations to maximize reward (e.g. information) while minimizing cost (e.g. fuel and time) and adhering to spatial and temporal constraints (e.g. vehicle dynamics and task time windows). Traditional problem formulations typically assume the team to be composed entirely of independent, autonomous vehicles. However, many current and future applications require tight coordination between humans and autonomous systems. Existing algorithmic approaches to multi-agent planning do not extend well to operations in which humans cooperate closely with robotic teammates due to the dynamic and stochastic nature of human performance. Reliable prediction of human task execution is challenging, and even proven models are subject to time-varying characteristics (e.g. fatigue and distraction) as well as differences between individuals (e.g. experience and skill). The difficulty of accurately modeling human performance demands a planning architecture that is highly responsive to heterogeneous, hard-to-predict agents. This thesis presents fast algorithms that integrate humans into the planning problem using well-established models from the human factors community to produce task allocations and schedules for tightly coupled human-robot teams. Humans are treated as dynamic agents, and feedback from their realized performance is leveraged to adapt agent models in real-time. The efficacy of this approach is investigated through multiple experiments involving human interaction with simulations of unmanned aerial vehicles (UAVs). Results indicate that adaptive human performance modeling provides distinct advantages in the context of mission planning for human-robot teams.