Predicting Perceived Naturalness of Human Animations Based on Generative Movement Primitive Models

Abstract

We compared the perceptual validity of human avatar walking animations driven by six different representations of human movement using a graphics Turing test. All six representations are based on movement primitives (MPs), which are predictive models of full-body movement that differ in their complexity and prediction mechanism. Assuming that humans are experts at perceiving biological movement from noisy sensory signals, it follows that these percepts should be describable by a suitably constructed Bayesian ideal observer model. We build such models from MPs and investigate if the perceived naturalness of human animations are predictable from approximate Bayesian model scores of the MPs. We found that certain MP-based representations are capable of producing movements that are perceptually indistinguishable from natural movements. Furthermore, approximate Bayesian model scores of these representations can be used to predict perceived naturalness. In particular, we could show that movement dynamics are more important for perceived naturalness of human animations than single frame poses. This indicates that perception of human animations is highly sensitive to their temporal coherence. More generally, our results add evidence for a shared MP-representation of action and perception. Even though the motivation of our work is primarily drawn from neuroscience, we expect that our results will be applicable in virtual and augmented reality settings, when perceptually plausible human avatar movements are required.

Evaluating Perceptual Predictions based on Movement Primitive Models in VR- and Online-Experiments

“…First, we compare the results of the VR-and the online experiments and contrast these with previous findings in a naturalness perception experiment [Knopp et al 2019]. We demonstrate that our paradigm works as intended by presenting the catch trial results.…”

“…Besides the involvement of the MNS in RM, Graf et al [2007] also show that visual movement prediction is a real-time process that includes effect estimations of motor commands before the motor action is performed. Visual Movement Prediction also requires prior information [Schröger et al 2015], such as visual identifications of the percepts, therefore making tasks of visual prediction more difficult compared to sheer tasks of identifying or distinguishing movements, such as in Knopp et al [2019]. This is consistent with the predictive coding framework, which follows from a Bayesian view of the MNS and also explains how we can infer movement intentions from movement observations [Kilner et al 2004].…”

“…We then show the predictions of logistic regression for different MP types and finally present results demonstrating the effect of contact events in our experiments. Figure 3 compares the mean confusion rates over complexity parameters of MP-Types of a naturalness perception experiment [Knopp et al 2019] with the two experiments described in this study. The previous experiment measured confusion rate in a task where participants had to choose the more natural one of two walking movements.…”

“…The representation of the movement was different as well: 3D avatars in the VR-, and stick figures in the online experiment. We also compared these results to published data in [Knopp et al 2019] , which used different movements, and a non-predictive paradigm. Our results indicate that measured confusion rates generalize across movements, paradigm, and rendering specifics.…”

“…This study is inspired by [Knopp et al 2019], and shares the same assumptions about MPs as possible common representation for action and perception. While this previous work focused on the perception of naturalness of movements, the current study addresses the predictive mechanism inherent in MP models.…”