There are many tasks that requires us to interact with physical environment, such as opening a door, turning a steering wheel, rotating a coffee mill, et al. In these tasks, the arm is usually constrained to the environmental geometry. Although there are infinite possibilities for human subject to select his/her arm trajectories as well as interacting forces when performing the tasks, experiments of human constrained motion however show that there clearly exist some characteristics inherent in all measurement data. Specifically, in this research, it is shown that, when human rotating a crank, he/she optimizes the criterion that minimizes the change of both the end-effector force as well as the muscle forces. This numerical result is strongly supported by human experiments data. Since this criterion is different from the minimum torque change criterion proposed to evaluate human reaching movement in free motion space, it is then suggested that human may use different optimal strategies with respect to different task requirements as well as environmental conditions.