The role that the internal model of the others plays in cooperative behavior

Abstract

Internal model of the others is claimed to be essential for the prediction of the others' behavior to realize a cooperative behavior. "Theory of mind" is known as a leading framework of the internal modeling. The present research consists of two parts. In the former part, experimental results are shown on the cooperation between a human subject and a software model of a mobile robot. The cooperative task given to the human subject is to carry a stick from a start point to a goal point by holding the one edge of the stick, in a simulation world displayed on a computer terminal. A virtual robot holds the other edge, whose behavior is controlled by a simple deterministic rule. The human subject is asked to perform the task under the following two different conditions A and B. In condition A, the subject is not told about the existence of the robot which holds the other edge, and only the stick is displayed. In condition B, the existence of the robot is told, but still not shown on the display. The performance of the carrying task is much better in the condition A, under which the subject can separate the movements of the stick and of the robot for their prediction. This result suggests that a construction of a behavior model of the other robot is helpful to carry out the cooperative task. In the latter part, a neural network replaces the human subject to perform the same cooperative task with the robot. The network outputs a motion to move the stick, and has a simple layered structure with an additional part to learn to predict the next motion of the robot. The experiments show the improvement in the task performance through learning the prediction. This also suggests the explicit modeling of the others is effective for the cooperation.