Adaptive task planning based on physiological state estimation and state uncertainty improves performance in human–robot collaboration

Human–robot collaboration combines the strengths of humans and robots to enhance productivity, flexibility, and safety. A task planner can use knowledge of relevant human and robot states to optimize overall task performance. Digital twins use models and real-time data to estimate these human and robot states. In our study, we implement a system of digital twins in which state information from a human–robot collaborative system is used to inform task reallocation in response to disruptions in team performance, with the goal of maximizing overall team throughput. The case study involves an assembly task in which three humans (one real, two simulated) work with a dual-arm robot. Humans are responsible for performing assembly operations, while the robot delivers parts. In aggregate, the digital twins predict the throughput of the human–robot collaborative system and detect disruptions, such as when the predicted throughput drops below a threshold. When a disruption is detected, the task planner reallocates tasks among agents based on their estimated capacity to maintain performance. The study examines how overall throughput is affected by varying levels of uncertainty in the predicted states of the human–robot collaborative system. In addition, the study explores how the task planner considers varying levels of human stress to allocate tasks and improve performance. The results of a study with 40 participants show that incorporating human stress estimates into task planning improves team performance, and that lower uncertainty values also improve team performance, but with a smaller effect. Together, these two factors improve throughput on average by 5.3% compared to a planner that uses high uncertainty and omits human stress estimates. These findings demonstrate that incorporating even simple physiological indicators and principled uncertainty filtering can significantly enhance task allocation decisions, offering a scalable approach to embedding human-aware intelligence into collaborative robotic systems.