Evaluation of the Uncanny Valley in CG Characters
Total Page:16
File Type:pdf, Size:1020Kb
SBC - Proceedings of SBGames 2012 Computing Track – Full Papers Evaluation of the Uncanny Valley in CG Characters Laura M. Flach Vanderson Dill Christian Lykawka Rafael H. de Moura M´arcio S. Pinho FACIN/PUCRS Soraia R. Musse Virtual Reality Group Virtual Humans Simulation Lab FACIN/PUCRS FACIN/PUCRS Abstract To answer these questions, we proposed an evalu- ation methodology based on a questionnaire with This paper revisits the Uncanny Valley subject in subjects. First, we selected representative char- order to evaluate its effects on the perception of acters from recent and well known movies and characters that currently use computer graphics, games, and other characters that are unknown. animation and computer simulation. It is based In order to evaluate the public's empathy with on surveys that generated hundreds of samples the chosen characters, we conducted a two-stages and showed preliminary results about new criteria questionnaire. In the first stage we presented still and correlations regarding the familiarity of the images and, in the second stage, videos with the characters with the public. The analysis of famil- performance of the same characters in the same iarity in those characters showed great agreement scenes. with the original curve of the uncanny valley. Fu- ture enhancements are then suggested to add new As the results, more than two hundred of analysed dimensions to the original graph. samples leaded to conclusions that indicated the uncanny valley hypothesis is valid for CG char- acters. The following sections cover the related Keywords:: Uncanny Valley, Computer works on the subject, the model for evaluation Graphics, Modern Media, Movies, Games (including the details of both surveys), obtained results and final considerations. Author's Contact: 2 Related Work flaura.flach,[email protected] In 1970, the roboticist Masahiro Mori [Mori 1970] 1 Introduction published a theory on how humans react emo- tionally to artificial beings. According to Mori, This paper studies the effects of the uncanny val- robots should not be made too similar to real hu- ley caused by CG characters, first presented by mans because such robots can fall into the "un- Mori in 1970 [Mori 1970] for robots. According canny valley", where too high degree of human to his work, robots made too similar to real hu- realism evokes an unpleasant impression in the mans can fall into the "uncanny valley", where too viewer. Mori plotted the industrial robots on a high degree of human realism evokes an unpleas- hypothetical graph of familiarity versus appear- ant impression in the viewer. The increasing ex- ance (Figure 1). posure of virtual characters to the general public motivated our group to test the original hypoth- For creatures, including robots, movement is gen- esis using more up-to-dated samples from movies erally a sign of life [Mori 1970]. As show in Fig- and games1. ure 1, adding movement changes the shape of the uncanny valley by exaggerating the peaks and Our study is an evaluation on how people perceive valley. To reproduce human like movements, sim- CG characters that are presented in modern dig- ilarity of velocity and acceleration are required. ital media. We try to answer: Does the uncanny For example, a robot has 29 artificial muscles in valley exists in CG characters? Does previously the face to make human like facial expressions, know the character affects the empathy with it? and laughing is a kind of sequence of face dis- Moreover, does adding movement to this char- tortions, where the distortion speed is an impor- acters changes the shape of the uncanny valley tant factor. If we cut the speed in half, laughing curve, like Mori [Mori 1970] suggested? looks unnatural. This illustrates how slight vari- ations in movement can cause a robot, puppet, or 1Images from movies and games where used here prosthetic hand to tumble down into the uncanny for scientific purpose. valley. XI SBGames – Brasiliá – DF – Brazil, November 2nd - 4th, 2012 108 SBC - Proceedings of SBGames 2012 Computing Track – Full Papers elicit a model of a human other but do not measure up to it" [MacDorman and Ishig- uro 2006]. If an entity looks sufficiently non- human, its human characteristics will be no- ticeable, generating empathy. However, if the entity looks almost human, it will elicit our model of a human other and its detailed normative expectations. The non-human characteristics will be noticeable, giving the human viewer a sense of strangeness. Several studies have started empirical testing of the uncanny valley theory. Both Hanson [Hanson 2006] and MacDorman [MacDorman 2006] cre- ated a series of pictures by morphing a robot to a human being. This methods appears useful, since it is difficult to gather enough stimuli of highly Figure 1: This graph represents the hypothe- human-like robots. However, it can be very dif- sized emotional response of human subjects plot- ficult, if not impossible, for the morphing algo- ted against anthropomorphism of a robot. The un- rithm to create meaningful blends. canny valley is the region of negative emotional response towards robots that seem "almost hu- With the same idea, Seyama and Na- man". Also, the movement amplifies the emo- gayama [Seyama 2007] investigated the uncanny tional response [Mori 1970]. valley by measuring observers' ratings of pleas- antness of static facial images, whose degree of realism was manipulated by morphing faces of dolls, masks, and CG characters into real human There are several theories that were proposed to faces. Their results showed that the uncanny explain the cognitive mechanism underlying the valley emerged only when the face images uncanny valley phenomenon. Some of the possi- involved abnormal features such as bizarre eyes. ble answers found in the literature ( [MacDorman and Ishiguro 2006], [Karl F. MacDorman 2009]) MacDorman et. al. [MacDorman et al. 2009], are: presents four empirical studies that explores these issues related to the uncanny valley in still im- • Mate selection: Automatic, stimulus-driven ages. In each one, they tested different features appraisals of uncanny stimuli elicit aversion in the face of a CG character, like skin tex- by activating and evolved cognitive mech- ture, level of detail, proportions, eye positions. anism for the avoidance of selecting mates The main findings of the studies were that the with low fertility, poor hormonal health, or more human the photorealisticaly-textured CG ineffective immune systems based on visible face looked, the easier it was for people to agree features of the face and body that are pre- on its degree of human likeness. Also, as human dictive of those traits [Green et al. 2008], likeness increased, the best-looking facial propor- [Donovan 2002]. tions were generally closer to the original propor- • Mortality salience: Viewing an "uncanny" tions of the human model. Another interesting robot elicits an innate fear of death. Plays finding is that facial proportions that are far from on our subconscious fears of reduction, re- ideal look eerier at higher levels of details that placement, and annihilation [MacDorman facial proportions that are nearly ideal. Besides, and Ishiguro 2006]. the CG face looked less eerie when the texture of the eyes and skin were at a similar level of pho- • Pathogen avoidance: Uncanny stimuli may torealism that when their level of photorealism activate a cognitive mechanism that origi- differed greatly. (For more details, see [MacDor- nally evolved to motivate the avoidance of man et al. 2009]). potential sources of pathogens by eliciting a disgust response. "The more human an or- Wallraven et. al. [Wallraven et al. 2008], provide ganism looks, the stronger the aversion to its an experimental framework for evaluating facial defects" [Donovan 2002], [Karl F. MacDor- expressions, and presents two detailed evaluation man 2009]. results, using their framework, of facial animation techniques. Also does a investigation of impor- • Violation of human norms: The uncanny tant information channels in the visual process- valley may "be symptomatic of entities that ing of facial expressions, and demonstrate how XI SBGames – Brasiliá – DF – Brazil, November 2nd - 4th, 2012 109 SBC - Proceedings of SBGames 2012 Computing Track – Full Papers the experimental results can be used to improve lead different perceptions? Additionally, in case the perceptual quality of animation techniques. a person felt discomfort by looking to a charac- The most important insights which can be gained ter, which parts of the face most influenced in from their research are that accurate simulation that discomfort? The following topics presents of facial motion (rigid head motion and non-rigid detailed descriptions of how the characters and deformations) is important for recognizing ani- the questions were chosen, and how the survey mated faces fast and accurately, and for perceiv- itself was conducted. ing intensity, sincerity and typicality of animated faces. The perceived intensity of animated facial 3.1 The Characters expressions can be increased by enhancing shape and motion information and including eye mo- A central issue in the construction of an appropri- tion. As long as motion information is available, ate question form is the choice of the characters degrading shapes and texture does not have a no- being presented. Considering the questions that ticeable effect on recognition (at least not for the must be covered, we tried to choose a not very experimental settings used). large number of characters to avoid a too long form. Criteria for choosing the characters were Chaminade et. al. [Chaminade et al. 2007], used set in order to obtain a group that allowed us to a biological motion classification task to investi- cover all the questions above. The first criterion gate the influence of discrete variations of ani- was the human likeness of each character. This mated characters' appearance on the perception feature is necessary because it is present in the of its actions.