Dynamic difficulty using brain metrics of workload

Dynamic difficulty adjustments can be used in human-computer systems in order to improve user engagement and performance. In this paper, we use functional near-infrared spectroscopy (fNIRS) to obtain passive brain sensing data and detect extended periods of boredom or overload. From these physiological signals, we can adapt a simulation in order to optimize workload in real-time, which allows the system to better fit the task to the user from moment to moment. To demonstrate this idea, we ran a laboratory study in which participants performed path planning for multiple unmanned aerial vehicles (UAVs) in a simulation. Based on their state, we varied the difficulty of the task by adding or removing UAVs and found that we were able to decrease error by 35% over a baseline condition. Our results show that we can use fNIRS brain sensing to detect task difficulty in real-time and construct an interface that improves user performance through dynamic difficulty adjustment.