Human maneuver model learning with prediction based filtering

Lots of researches on Human-Machine Systems (HMS) have been investigated to improve the task performance with human-machine cooperation. Most of the works have focused on a single operator as enhancing individual skill. Thus, few studies on teamwork assist for cooperative tasks by multiple human-beings. In our previous works, to realize a teamwork assist system, quantification technique of "Concern For Others: CFO," which was a key factor of cooperative performance of such cooperative tasks, was proposed. The CFO is defined as difference between the command input in a cooperative target task and predicted input by the human maneuver model in the solo task. The human maneuver model in the solo task is learned by prior solo experiment as a calibration. Since the maneuver model is very important to calculate the CFO, precise and efficient learning method would be required. Therefore, in this paper, a efficient learning technique, with recurrent neural networks and prediction based filtering, for the solo task model is proposed. Finally, the prediction performance are experimented, and validity of proposed method is discussed.