Virtual Human Interaction Lab – Stanford University
{ Projects }
Avatars and Behavioral Modeling
Virtual reality enables us to create a powerful and persuasive stimulus: the virtual self. Using digital photographs, we can create avatars that have a striking resemblance to the self. We can then manipulate the virtual self in myriad ways that would be difficult or even impossible in the real world. The virtual self can modify its appearance or perform a behavior that the real self cannot, thus serving as a novel type of model. According to social cognitive theory, models can be valuable stimuli for encouraging the imitation of particular behaviors. Thus, we are investigating how using self-models and virtually manipulating social cognitive constructs such as identification, self-efficacy, and vicarious reinforcement can influence imitation, particularly in the context of health and consumer behaviors. Is seeing the virtual self engage in a healthful activity more or less effective than a virtual other? When an avatar shows positive benefits of using a product in the third person, does the consumer then go out and buy that product? Can behaviors be encouraged by seeing the virtual self model health-related rewards and punishments such as weight loss, weight gain? [publications: 1 2 3]
The Proteus Effect
Cyberspace grants us great control over our self-representations. At the click of a button, we can alter our gender, age, attractiveness, and skin tone. But as we choose our avatars online, do our avatars change us in turn? In a series of studies, we've explored how putting people in avatars of different attractiveness or height change how they behave in a virtual environment. [publications: 1 2 3]
Transformed Social Interaction
In collaboration with the Research Center for Virtual Environments and Behavior, we are interested in the experience of social presence as well as task performance within collaborative virtual environments. We are utilizing virtual reality simulations in which people interact in real-time within a collaborative virtual environment. Specifically, we seek to: 1) learn more about the behaviors that occur during collaboration, and 2) explore the idea of transforming social interaction by selectively augmenting and decrementing these behaviors in order to provide the interactants with novel tools during interaction. In other words, by selectively rendering behaviors that were not actually performed, or alternatively by not rendering behaviors that were in fact performed, immersive virtual environments allow for conversational strategies that are not possible in face-to-face interactions or videoconferencing. We are examining the effect of implementing these novel strategies, and testing their influence on conversation in terms of task performance, learning, persuasion. See our wikipedia entry on TSI. [publications]
Avatar Identity
What are the implications of having an avatar, that is, a digital model that represents you in virtual reality? We are studying the ties that individuals have to an avatar. Specifically, how much does an avatar need to resemble (both visually and behaviorally) its respective owner in order for person-specific influences to take effect? When does this digital representation stop being George Bush and start being John Kerry?
Using a variety of affective, behavioral, and cognitive measures we are exploring the phenomenon of virtual self, and examining the implications of avatar representation. [publications: 1 2 3]
Learn how Kerry could have won the election through facial identity capture.
Learning in Immersive VR
In collaboration with Berkeley's CITRUS lab we are exploring how immersive virtual reality extends the benefits of video learning, allowing the user to enter the same world as the teacher. First, immersive settings allow users to see in full three dimensions, greatly increasing detail, presence (i.e., learners feel psychologically as if they are in the digital learning environment, as opposed to the physical space) and social presence (i.e., they feel as if the digital reconstruction of the instructor is a real person). Second, as opposed to stationary video, immersive virtual settings allow users to control how they view the environment by allowing them to change aspects such as camera position and orientation, even allowing a disconnect between their own representation and their point of view in real-time. Third, video settings only allow users to watch the instructor; immersive virtual reality allows the user to interact with the instructor and the environment, as well as to perform novel functions such as sharing body space with the instructor during learning. In the first experiment completed using this paradigm, we demonstrated, via subjective self-report of the learners as well as more objective measures involving expert coder ratings of learners performing the tai chi moves later on in physical space, that people learned more in the immersive virtual reality system than in the 2D video system. [publications: 1 2 3 4]
Avatars in Second Life
Despite the incredible popularity of the online virtual world Second Life, there is really no empirical data measuring a) what exactly to people do inside of SL, and b) what are the effects of interacting via avatars over time. Our longitudinal (8 week), large sample size (80 participants) study examined the influence of avatar appearance on virtual and offline behavior. At the start of the study, experimenters gave each participant an assigned avatar shape (tall, short, overweight, or opposite gender), L$1000, and a scripted object which would track online behavior such as chat content, animation use, and locations visited. Factors such as major, programming experience, and gender were split as evenly as possible between the four conditions. For six weeks, participants spent a minimum of six hours actively participating in Second Life activities. At the conclusion of each week, participants completed a web-based questionnaire which gathered information about real world activities and attitudes as well as reactions to the past week's Second Life experiences. Our data shed light not only on the behaviors which occur in SL, but also the effect those behaviors have on the users' "first life". Moreover, by tracking changes to their avatars' appearance, we can begin to answer the question of how avatar choice effects psychology and behavior. Researchers interested in accessing our large data set should
click here. [publications: 1]
Haptic Communication in Social Interaction
We are exploring the use of networked digital touch in collaborative virtual environments. Specifically, we are examining how often people touch one another using standard 6DOF and force-feedback devices, how haptic patterns of interactants correlate with other behaviors, attitudes, and personality attributes, and the effect that virtual person-to-person touch has on copresence, trust, and relationship formation. [publications: 1 2]
Facial Tracking and Emotion Abstraction in Communication
Much research has demonstrated that facial expressions not only correlate with emotions but can also be used to accurately predict more specific mental states. In our current Facial Tracking and Behavior work, extend this work to use facial features to predict applied behaviors. For example, in our 'Driving Project,' we demonstrated that facial geometry features, especially features involving the eyes and mouth, can be used effectively as predictors of poor driving behavior, identifying accidents two seconds before they occurred with an average accuracy of 75.8%. These classifiers could be incorporated in advanced driver warning systems for improved vehicle safety. In our 'Online Shopping Project,' we demonstrated that the face can predict buyer intent with an average accuracy of 72.61%. This opens up possibilities for practical applications and further research using facial expressions as predictors of online buying intention. In our most recent study, 'Monitoring Operator Fatigue,' we demonstrated that the face can accurately predict such variables as errors, fatigue level, and learning factors while performing a repetitive motor task. The results were outstanding, showing an average accuracy of 77.7% accuracy at predicting errors, 89.2% accuracy at predicting fatigue levels, and an average accuracy of 74.2% at predicting learning rates. These results indicate that monitoring facial features alone can yield answers to optimizing performance at the work place.
[publications: 1 2 3 4 ]
Presence
The construct of presence has often been used as a metric to evaluate the utility of a virtual environment. While there is no consensus on an exact definition, the general notion concerns the degree to which the user actually feels as if they are present in the virtual environment (as opposed to present in the physical world). Moreover, related concepts are social presence, the degree to which people feel connected to other people in the virtual world, and self presence, the degree to which people believe their own avatar is actually them. Despite broad research on the topic of presence, reliable measures are still lacking, and much debate as to how to quantify the construct exists. Our research in these areas focuses on developing behavioral measures of these constructs (as opposed to self report measures), and on determining the relationship between how real a virtual world or avatar looks and/or behaves and the subjective experience of presence. Overall, findings indicate that behavioral measures are more reliable than self report ones, and that increasing realism can sometimes be counterproductive and result in less subjective presence. [publications: 1 2 3 4 5 6 7 8 9]
Eyewitness Testimony and Virtual Police Lineups
Investigations in collaboration with the Research Center for Virtual Environments and Behavior, the National Science Foundation, and the Federal Judicial Center examine the possibility that virtual environments will aid in our understanding of how witnesses of crimes identify suspects. Current police lineup procedures rely predominantly on mug photographs when conducting lineups. We believe that conducting lineups using three-dimensional, digital busts of heads and bodies offers a number of advantages: witnesses can view the suspects and distracters from any angle (not just front and profile), witnesses can move as close or as far from the suspects as they desire, test administrators can use mathematical models to distance the distracters from the actual suspects by simply morphing the underlying mesh model of the suspect, the lineup can occur within the virtual crime scene (i.e. the suspects and foils can lineup in the context of the actual crime, and finally, the virtual suspects and distracters can be animated, not just static. More information about this project can be found here. (see video) [publications]
Homuncular Flexibility
In this line of studies, we are examining a concept first developed by Jaron Lanier called "homuncular flexibility" – learning to remap physical degrees of freedom onto digital representations in interactive tasks. The crucial question we are addressing with homuncular flexibility is: Can people learn to remap degrees of freedom that are not essential to a task in order to control novel digital actions which are releveant to a task? For example, if the task were using a hand to paint a wall, could a person learn use one physical hand to control multiple virtual hands – splitting degrees of freedom from the arm to control the XYZ Position of five hands at once? In collaboration with Jaron Lanier and Stanford's LIFE lab, we are examining the conditions in which people learn to remap degrees of freedom onto digital space, including the amount of time it takes to form a mental representation of the remapping, the ability to dual task both hands at the same time, and the nature of learning tasks that facilitate being able to use remapped information.
Massively-Multiplayer Online Role-Playing Games
Every day, millions of users interact with each other in online virtual environments through avatars. Nowhere else but in these environments do we see such pervasive social interaction, relationship formation, and collaboration between avatars. If we are interested in avatars, collaborative virtual environments, and virtual reality, these Massively-Multiplayer Online Role-Playing Games have a lot to tell us. Recent findings are presented online. [publications]
Diversity Simulation
Using immersive virtual reality, it is possible for someone to literally experience the world as another person. In other words, someone can become a passenger to someone else in an immersive simulation that is designed to demonstrate what it is like to walk a mile in the shoes of another. We are using these simulations to explore relationships of gender, status, and race, and testing the "extended contact" hypothesis, namely that wearing the face of another in a simulation designed to highlight diversity issues can increase awareness. (see video) [publications: 1 2]
