Abstract
For even a simple point-to-point movement of the hand there are an infinite number of possible ways the arm can move to obtain the final hand position. One solution used to obtain a unique movement plan is to select the movement which is in some sense optimal. Many optimization criteria have been proposed by computational theorists in human motor control, but the particular form of a human criterion remains controversial. We examined this issue from the standpoint of visual perception by producing simulations of 14 different optimization criteria (e.g. minimum torque change, minimum jerk, minimum angular jerk) and presenting these simulations to observers. As an additional control condition we obtained 3D human movement data of the same motions depicted by the simulations. The set of movements were chosen to demonstrate the variety of kinematic features found in natural human movement such as bell-shaped velocity profiles, gently curved paths and corrective submovements. Both the biomimetic simulations and the human movement recordings were displayed as a computer animation on a human form as well as a movie clip of a humanoid robot. In two separate experiments 10 participants either rated the naturalness of the movement or the similarity of a pair of movements. The results of the naturalness judgements revealed complex interactions between the type of movement and the mode of presentation (humanoid robot or human animation). The similarity data was examined with MDS which revealed clusters of different optimization criteria, however natural human movement didn't appear to belong to any of these clusters. These results will be discussed in terms of the kinematic features of natural and synthetic movement that best explain this complex pattern of results.