Problem

Virtual reality avatars currently do little more than add visual interest to what formerly were text-based shared spaces. Navigating your avatar around a virtual world is about as satisfying as moving a chess piece from one square to another on a a chess board. Perhaps it's even less satisfying because we expect our avatar to do more for us than act as an announcement that "we have arrived." One may as well have typed into the chat line: Francesca arrives in the Emerald room.

One of the things we expect from avatars is that they will provide a means for visual expression. We want our avatars to be able to gesture, at times wildly to draw attention to ourselves, at times more subtly to give emphasis to our speech acts.

To get what we want, avatars that we can control, we need to take a look at the design of avatars. In particular we need to design articulated limbs for avatars, and the means to control these in real time. Then we can begin to address the question of the efficacy of communicating via avatars.

Analysis

A tool can be built to explore a small space of avatar designs, including the means of controlling the gestural movements of the avatar. One type of avatar that people will want to use is a humanoid-type figurine with articulated limbs. These limbs should be controllable by the human via some standard computer input device, most likely a mouse. Because we are exploring the possible expressive power of this type of avatar, we will want to vary the look of the avatar (including its gestural movements), and the control law mapping between the input device and the output gestures.

Suggested Design

Rigid body simulation and control

The avatars are modeled as multi-linked rigid bodies. Simulation and rendering of these avatars will be handled by a virtual world manager, most likely Impulse. (Impulse is a rigid body dynamics simulator developed by Brian Mirtich for his PhD degree at Berkeley.)

Users would be able to control the movements of the avatars via a mouse interface. Currently such controls exist for single-link rigid bodies. The two working examples are a blimp flight simulator, and a cart driving simulator.

The output is a real-time interactive 3d graphics rendering of the avatars in their environment.

Shared world

The tool would allow at least two people to use the virtual environment at the same time via client applications. The virtual world manager will run on a server machine. Each user would be controlling their own avatar, and seeing the virtual world on their own machine.

Experiment

1. Independent variables
   • The type of avatar.
2. Dependent variables
   • The type of expression which is being generated.
   • The number of matches between intended meaning for a gesture and its understood meaning.
   • The length of time to invent a gesture.
3. Participants
   • Expert computer users such as graduate students in the department.
4. Method

   I. Single user task

   A single participant is asked to invent and perform a number of gestures that express possible non-verbal actions one might do in a meeting. for example,

   1. I want to speak next.
   2. I have to interrupt now.
   3. I agree with the current speaker, but don't need to talk next.

   II. Dual user task

   One participant is controlling the avatar, and the other user is observing the virtual world. The first participant is asked to perform the same types of tasks as in the single user experiment. The second participant will be asked to identify the meaning of gestures which are being performed by another person.

   Both tasks are repeated twice, once with each type of avatar, with the order of tasks systematically varied.

5. Results and Discussion

   In the first task we study how a user might express their intentions using the avatar. The types of actions that the user is asked to perform have been chosen because they are a commonplace uses of non-verbal communication. We are interested in the types of gestures the user invents, how long it takes the user to come up with a personally satisfactory gesture and how different the gestures are from one another. Difference can be measured in terms of temporal duration, distance of movement and repetition of movement.

   We expect that the controls will be difficult to use; hence we are mainly interested in observing how the participant goes about learning to control the avatar and then inventing gestures.

   In the second task, assuming we get that far, we wish to observe how a human relates to the avatar and interprets its gestures. We hope the the context of a "meeting" will help to disambiguate the different gestures.

Grading Contract

Tasks

1. Implement a controllable avatar with a single-jointed limb. (This task includes developing a control interface for multi-link bodies in the simulator. Currently only single-link bodies are controllable.)
2. Implement an avatar with a multi-jointed limb.
3. Develop a multi-user virtual environment. (Currently the system is single user.)
4. Conduct experiments.

Contract

• B for partial implementation of (1).
• B+ for full implementation of (1) and experiment with task I.
• A for implementation of (1) and (2) and experiment with task II.
• A for implementation of (1) and (3) and experiment with task I.
• A+ for implementation of (1), (2) and (3) and experiment with task II.

last updated 28 October, 1997