To help others, we need to infer one’s goal and intention and
make an action which complements one’s action yet to meet
the underlying goal. In this study, we consider the
computational mechanism how a person can infer the other’s
intention and goal from his or her action, which is not
completed or fails to meet the goal. As a minimal motor control
task toward a goal, we analyzed single-link pendulum control
tasks and its variation. By analyzing two types of pendulum
control tasks, we show that a sort of fractal dimension of
movements is characteristic of the difference in the underlying
motor controllers. Further, using the fractal dimension as a
criterion of similarity between movements, we show that the
simulated pendulum controller can make an action toward the
goal, toward which other’s incomplete action was made, but
was not observable in behavior due to its failure.