Elderly or physically challenged people whose motor function have declined often have deteriorated environment cognitive abilities. Therefore, power-assist robots with perception-assist function have been proposed to assist such persons. In perception-assist, it is necessary to make a proper virtual tunnel to perform the intended upper-limb task properly. When there are several target objects such as dishes and bowls on the table during the eating task, it is necessary to determine which target is selected by the robot user. In this study, experiments are conducted to measure the daily living behaviors of healthy people in three dimensions, and parameters necessary to automatically generate the virtual tunnel are obtained experimentally. In this paper, an algorithm which automatically generates the virtual tunnel is proposed. The effectiveness of the proposed algorithm was evaluated by conducting experiments.