The current story is set on a fantasy island (see fig. 2). The terrain is fairly large and includes a variety of spaces that invite exploration. The children can climb a dormant volcano, build a bridge crossing the waters to reach another smaller island, and explore a labyrinth of caverns winding under the island.
The main constructive activity is to build and develop small local ecosystems on the bare parts of the island. The terrain serves as an open land which the child explores to decide where to plant and populate. Various seeds for planting garden vegetables and trees are stored in crates and serve as starting points for building micro-ecosystems on the island. Additionally, the child can elicit the assistance of several genies, such as a cloud genie to provide rain, or the fire-flies to illuminate the vast underground expanse. Our immediate plans are to have the genies make their actions explicit, in the case where the child cannot perceive cause and effect right away.
When the user drops a seed on the ground, the corresponding plant, flower or tree will start to grow. The pace in which this happens can be predetermined; we may choose to see the system grow very quickly, or, in the case of a school project, extend it over the period of a semester.
The tomatoes, carrots, pumpkins and other plant objects contain a set of characteristics that contribute to their growth. They all have values for their age, the amount of water they hold, the amount of light they need, their proximity to other plants of their kind. These values determine the health of the plant and its size. Visual cues aid the child in determining the state of a plant or flower. When the cloud has been pouring rain over it for too long, the plant opens an umbrella; when the sunlight is too bright, it wears sunglasses. Sound in the environment also enriches the surroundings in a variety of ways. Different environmental sounds are experienced depending on where each participant is standing (e.g. the children by the shoreline will hear the water, in the rainforest will hear the birds, etc.)
The narrative, as implemented in this stage, revolves around the construction of the ecological microworlds and the decisions being made through the interaction with the genies. Every action is being recorded and even when there is no interaction, the world continues to evolve through time. Each child may choose to join or leave the story set at any time, while the stories on the island continue and others can explore its mysteries. Parts of the island, as well as all of the objects and agents, are Inventor models. This allows the current scene to be easily saved and converted into VRML (Virtual Reality Modeling Language), creating a 3D snapshot of the world at any given time as an artifact that can be browsed on the Internet.
Figure 2: A scene from the GULLIVR Story Set: at the bridge.
The above activities can be performed collaboratively between children at two or more remote CAVEs or Immersadesks. Each tracked participant's presence in the virtual space is established using an avatar. These avatars consist of a separate head, body, and hand. As each person's head and hand are tracked, this allows the environment to transmit gestures between the participants such as the nodding of the user's head, or the waving of user's hand to the other participants. As these avatars have sufficiently detailed representations, the participants can communicate notions of relative position to one another with phrases such as ``it is behind you'' or ``turn to your left.'' (fig. 3). We have found this to be effective in establishing co-presence over our previous experiences in transcontinental collaboration[Leigh et al, 1995].
The children communicate with other remote participants using their voice. A high-speed ATM voice conference system is used to allow the various human participants to talk to each other. In a similar way, the children interact with the computer-controlled agents using a speaker independent voice recognition system. In this case, the child wears a wireless microphone and communicates actions using simple phrases (e.g. ``fire-flies, get brighter!'' or ``signpost, where is the bridge?'') In trials of this technique, the system was able to reliably respond to the user's utterances even in relatively noisy conditions. Live video for the communication of the remote participants has also been tested. However, we would like to investigate whether this feature effectively improves communication or instead destroys their suspension of disbelief.
Given that some of the genies possess a wide range of capabilities and personalities that have to be believably represented, we have chosen to take an approach commonly used in human-factors circles. Rather than investing enormous amounts of time architecting a complex artificial intelligence engine to realistically simulate artificial agent behavior, we use the Wizard-of-Oz prototyping technique to quickly simulate the behavior of pre-built software systems. In GULLIVR, Wizard-of-Oz prototyping is used for controlling the `helper' genies that possess a more anthropomorphic character. A remote human being takes on the role of each genie- controlling their behavior in response to changing conditions. The child is unaware of which characters in the environment are controlled by the computer and which by a human operator.
While the behaviors of some of the genies (e.g. the fire-flies or the sign post) will be realizable in the short-term, the side-kick may always require a human operator to maintain the illusion of believability.