Optimal human-in-the-loop interfaces based on Maxwell's Demon

Interactions with complex systems require safe and reliable interfaces. However, such interfaces must be able to account for substantial uncertainty resulting from the unpredictable nature of human behavior. In this study we demonstrate that a controller/filter unit operating as a Maxwell's Demon can be used to synthesize human-machine interfaces that effectively filter user input in real time. Our Maxwell's Demon Algorithm (MDA) was applied in mechanical and software filtering of user actions for the cart-pendulum inversion task. Software filtering was implemented and tested using a custom Android application. Additionally, a haptic device was employed to create a mechanical filter. Results from nine healthy subjects show that both software and mechanical filters increased the success rate of subjects in the swing-up task. This result suggests that the MDA may be applied to design reliable human-machine interfaces for rehabilitation, training, teleoperation and other shared control tasks.