K. Mauldin, M. Rickard, J. Park, C. Jones
California Baptist University,
United States
Keywords: human performance, machine learning, wearables, flow state, gaze monitoring, smart frames, focus, performance improvement, training programs
Summary:
Flow state is a state of deep absorption in a task that is positively related to increased performance and intrinsic motivation. In order to enter this highly sought after state, appropriate focus and arousal levels must be achieved and maintained. We have proposed and performed early investigation into the dynamic understanding of flow state through measurement of variables that represent attention allocation and span that can be collected in a non-obstructive manner using a wearable sensing device. The lightweight, wearable system supports continuous gaze monitoring of eye gaze, pupil diameter, blink state and head pose during natural, real-world activities. The ocular imagery and environmental data are collected through conventional-looking spectacle frames in traditional form factors (Blink FramesⓇ) that house microcameras, various sensors, supporting microelectronics, and a rechargeable battery. Arousal levels are then gauged by heart rate data collected using a commercially available pulse sensor. By coupling these real-time measurements, longitudinal data is collected during routine arbitrary or scripted activities. Participants wearing a set of instrumented frames and a heart rate monitor perform a set of activities using a computer or tablet. The combined data set includes measurements of heart rate and heart rate variability, eye gaze direction, pupil diameter, blink rate and blink speed, head pose (roll and yaw) and head motion vector. To infer and understand statistical properties related to flow state, Artificial Intelligence methods - specifically learning systems using machine learning models – are applied to the data. Such methods allow the detection of multivariate and highly complex relationships in the data, as compared to classical statistical analysis. Time-aware analysis models such as Recurrent Neural Networks are used to label periods of time with their inferred flow state. From this data, estimated flow state and associated performance level is computed during engagement in various tasks. We have conceptualized a training regimen in which attributes associated with increased flow state can be determined for various types of tasks. Patterns of behavior, especially pose and gaze, that are connected to greater flow are identified on an individual basis and feedback to the user is provided to encourage these behaviors. The resulting training regimen can help wearers to understand the patterns of behavior that will aid them in achieving flow state and maximizing attention and performance.