Abstract
As the visual appearance of an artificial agent becomes more humanlike it has been proposed that observers find the agent less acceptable and this phenomenon has been described as the “uncanny valley” (Mori, 1970). However, there is little evidence to support the existence of the uncanny valley and the further hypothesis that character motion would deepen the valley. To explore this phenomenon we first constructed a stimulus set that provided an uncanny valley pattern of responses to static form and then explored the effect of motion in modulating this pattern. Specifically, we used computer animation software (Poser, 3D Studio Max) to obtain 7 body models (robot A, robot B, mannequin, skeleton, zombie, low and high resolution human) and obtained independent ratings of human likeness and acceptability to images of these models. Plotting acceptability ratings versus human likeness ratings revealed the predicted “uncanny valley” in acceptability just before the highest levels of human likeness. Next we animated these forms using 3D motion capture data that had been processed to depict different levels of realistic motion. These different levels of realism included the original natural movement and movements with increasing levels of degradation. The motion degradation process involved periodically freezing and unfreezing degrees of freedom of single joints along the moving arm so that the movement appeared, to varying degrees, jerky and mechanical in nature. Results indicated that the natural motion generally improved the acceptability ratings of all but the highly realistic figure which became less acceptable when animated. All levels of degraded motion generally brought acceptability down to values below that obtained from static form alone. These results clearly show the impact of motion on the acceptability of computer characters, but failed to produce the previously predicted deepening of the uncanny valley.