Abstract
The human-centered automation principle, saying that the human should have the final authority over the automation, has been regarded as the essential design requirement of automated systems. However, the reliability of human performance can be decreased by the effects of time pressure, high workload, and so on. Therefore, adaptive automation systems, which are characterized as the dynamic function allocation between the human and the automation, are expected. In order to realize such systems, the estimation of operators’ workload are necessary. The present research, therefore, has developed a workload estimation method using the physiological data of an operator. A wearable sensing device called JINS MEME was introduced to obtain operators’ electrooculography (EOG), acceleration, and gyro sensor data while they handled a complex simulation task provided by Smart Grid Simulator. A machine learning method, Support Vector Machine, has successfully identified two types of categories of operators’ workload conditions, “High” and “Acceptable”, over 90% accuracy using 10 parameters based on JINS MEME outputs. In addition, based on the detailed analysis of individual differences including each parameter, the effective utilization method of machine learning in workload estimation for adaptive automation has been discussed.
