Co-adaptation improves performance in a dynamic human-machine interface

Abstract

Despite the growing prevalence of adaptive systems in daily life, methods for analysis and synthesis of these systems are limited. Here we find theoretical obstacles to creating optimization-based algorithms that co-adapt with people in the presence of dynamic machines. These theoretical limitations motivate us to conduct human subjects experiments with adaptive interfaces, where we find an interface that decreases human effort while improving closed-loop system performance during interaction with a machine that has complex dynamics. Finally, we conduct computational simulations and find a parsimonious model for the human's adaptation strategy in our experiments, providing a hypothesis that can be tested in future studies. Our results highlight major gaps in understanding of co-adaptation in dynamic human-machine interfaces that warrant further investigation. New theory and algorithms are needed to ensure interfaces are safe, accessible, and useful.