[week 10 summaries]

The Uncanny Valley: Effect of Realism on the Impression of Artificial Human Faces

This paper investigated the observer’s impression on artificial human faces with different realistic levels. Particularly, the uncanny valley hypothesis is being investigated and discussed.

By Mori, the “uncanny valley” refers to the period with decreasing pleasantness when the artificial human faces are approaching to higher degree of realism. Several experiments are conducted to observe the merge of uncanny valley. To quantize the percentage of ‘realistic’, the ratio of source image (pure artificial face) and destination image (pure real human face) during morphing is utilized. All those experimental images start with pure artificial face and smoothly change to a real human face. Raters are asked to score their feeling of pleasantness over those randomly shown images. After the survey, scores are organized to form a statistic chart.

One thing I learned from this paper is the way to rule out factors one by one to trace the root cause of some experiment result. This step is needed no matter experiment result meets the expectation or not. If the result is not the way you expect, it’s may not because your theory is wrong. It is possible that other unexpected confounding factors have more influence over the sample. Similarly, if the result turns out to be following your hypothesis, there is chance that other confounding factors can better explain it than your theory.

The first trial in the paper gives a data set that shows no uncanny valley. Though this experiment is carefully designed, several aspects are analyzed as confounding factors that may cause a wrong result. Then experiment 2 is conducted with modification based on the analysis, in which a first decrease then increase curve is shown. To better understand this curve and diagnose whether that is uncanny valley, two more detailed experiments are conducted with different ways to generate artificial models. By subtracting similar data without control factor, some confounding factors can be canceled out or at least eliminated. The remaining shows clearly a decrease of pleasantness over more realistic artificial face. So conclusion that uncanny valley does merge, but only when the face images involved abnormal features is drawn. The experiment process and data analysis are very convincing. But to make it complete, influences from other factors like hairstyle, skin smoothness are included in future study.

Physiological Measures of Presence in Stressful Virtual Environment

This paper talks about finding a way to measure the level of presence in Virtual Environment. The requirement of this measurement should be reliable, valid, sensitive and objective.

To compare different ways of measurement, a virtual environment is created for experiment. This VE contains three parts, one training room, one pit room and a 20ft hole beneath the pit room. The training room is quite ordinary, and is adjacent to the pit room. The pit room is borded with a narrow walkway. A 20ft virtual space can be seen right beneath the floor though the walkways.

Three physiological metrics are chosen as the measurement of presence in VE. They are

-change in heart rate

-change in skin conductance

-change in skin temperature

Three experiments are conducted to evaluate those three metrics. Multiple Exposures ask subjects to carry a book from training room to pit room and then return to training room. Passive Haptics has the same instruction as Multiple Exposures, but was conducted with control group. Each subject will experience walking on the laboratory floor and walking on a 1.5-inch wooden ledge. In the second case, subject’s edge-probing foot will feel the the edge of the wooden ledge. Frame Rate asks subjects to pick up a boxes in training room, move to pit room and drop it to a specific target in the hole . This procedure implicitly force subject to look down into the pit. By changing the frame rate, its effect on presence sense is analyzed.

Reliability: By hypothesis, an environment like the pit room will consistently evoke physiological reactions. And the acquired data shows the reactions. The heart rate is higher, skin conductance is higher and skin temperature is lower when the subjects expose in pit room. Besides, though decreased due to habituation, these metrics never go to zero or unusable range.

Validity: Based on questionnaire, the change of the heart rate appears as a better metric since it has the highest correlation for both case.

Sensitivity: The term multi-level sensitivity is utilized to distinguish sensitivity in different dimensions. Like both as ‘high presence’ situation, the subjective has a more strong reaction in magnitude when entering from training room to pit room than the comparison between with/without the ledge. In Passive Haptics case, changes are more significant with ledges. In Frame Rate study, besides the anomalous at 10FPS, the reaction is more sensitive with higher frame rate.

Objectivity: The objectivity of these measurements are guaranteed from both subject bias and experimenter bias.

In all, this paper introduces how measurement is evaluated regarding to the sense of presence. It has the conclusion as change of heart rate has the best performance by analyzing from reliability, validity, sensitivity and objectivity. I am wondering the extension of this conclusion. The measurement is based on the physiological study that people will get stressed out in environment like that. But if the VE is relaxing and comforting, there might not be changes in heart rate or other two metrics. In cases like that, designers may need to find new measurements.

Variations in Physiological Responses of Participants During Different Stages of an Immersive Virtual Environment Experiment

This paper also investigates the change of physiological parameters influenced by the sense of presence. Different from Meehan’s research, which refers to the physiological change caused by stress, this paper described an experiment environment which is less stressful. Also, this research cares more about the change from before getting into the VE to the end of the experiment. Besides, texture quality and virtual characters’ level of realistic are included for discussion as factors that influent the sense of present.

The virtual experience for candidates is divided into 4 segments, each lasts 90 seconds. The baseline segment is when the participants are asked to stand still in dark before entering the VE. This period is considered relaxing and activeless, the data of which can be served as the baseline for other segments. After this segment, the VE appears as a training room. The purpose of training room is let the participants get familiar with VE tools. In this segment, participants are asked to move frequently until they feel confident with those tools. The last 90 sec is recorded for analysis. Then the experiment begins with a virtual street and virtual characters. This session is divided into two segments: first 90 seconds of the experiment and second 90 seconds of the experiment.

Change of heart beat (HR) is recorded for every participant. Also, HRV, as refer to Heart Rate Variability is recorded as a way to distinguish the change of HR is caused by mental stress or physical stress. Different combinations of virtual street texture and virtual character are used to measure the presence.

Based on the change of HR and HRV, analysis and hypothesis are given to explain their behaviors. The first rise of HRV and HR from the end of baseline segment to training segment may be caused by the curiosity of entering the VE. The last rise of them by the end of experiment is possible from the newly noticed technical problems or “social interaction” disturbs. The fall of them during the first half of experiment may come from less move and activities in the virtual street. Though those explanations are all hypotheses, but the changes that have been detected prove the value to take a deep investigation. Also, by comparing results between different texture and virtual characters, the one with non-repetitive textures has the biggest influence over HR and HRV. So, if higher or more intense response is expected, more fidelity models need to be considered.

Comments are closed.