Proposal of a Modeling Method for Human Operations in Complex Tasks

This study proposes a high-accuracy model for predicting human operations in complex tasks using a neural network (NN). By recursively incorporating task information, the NN predicts torque commands and simulates a plate control task (PCT) without relying on real-time human input. The PCT involves controlling a two-degree-of-freedom virtual plate with a rolling ball, driven by gravity and air resistance, while following various target trajectories. Experimental data were collected from ten participants performing three different target trajectories. Training sets were used to build models, which were then evaluated on separate datasets. Comparative analyses with a proportional-derivative (PD) model and McRuer’s crossover model showed that the proposed NN-based method achieved significantly lower prediction errors in torque, plate angle, and ball position. These findings indicate that this recursive NN model is well-suited for capturing complex human-in-the-loop behaviors and can enhance assistive technologies in advanced human-machine interaction systems. It operates quickly for real-time use.