Design Agents in 3D Virtual Worlds

Design agents are rational agents that monitor and modify elements of a designed environment. Special characteristics of design agents include the ability to reason about patterns and concepts, and the ability to act autonomously in modifying or changing the design to achieve their own goals. 3D Virtual Worlds are multi-user distributed systems that provide a designed environment and a closed world environment for studying design agents in a multiagent system. We present a model for a design agent reasoning process and a model for constructing a memory of the agent’s knowledge and interaction with a virtual world. The reasoning process includes sensation, perception, conception, hypothesizing, and planning a sequence of actions. Each agent has a constructed memory: a dynamic and changing view of the designed world that is determined by the agents sense data and reasoning. This model of design agents is compared to the BDI model of rational agents. We have implemented these models by extending the Active Worlds platform so that each object in the 3D world can have agency. We illustrate the models with a door agent and a multi-agent office that operate within a 3D world. 1 Agent-Based Virtual Worlds 3D Virtual Worlds are multi-user distributed systems that provide a closed world environment for studying design agents in a multiagent system. The 3D virtual worlds that we have been working with allow multiple users to connect to a server through a client, where the processing is distributed between the server and the client. Our early investigations used a virtual campus implemented in LambdaMOO [Maher and Skow, 1997] as the virtual world environment, and we now use a virtual learning environment implemented in Active Worlds 1 (AW) as the software environment for developing our agent societies. In Active Worlds, the users appear as avatars and are able to interact with and change the objects in the world. This environment provides an unbounded closed environment http://www.activeworlds.com Figure 1: A 3D model of a virtual conference room. for studying agent reasoning, interaction, and communication. Figure 1 shows a virtual conference room that was designed and used for meetings and seminars. The objects in this room exhibit interactive behaviors preprogrammed in the Active Worlds platform, but they do not have agency. Our research starts with this platform as a basis for developing agent-based virtual worlds. With the exception of human controlled avatars, each object in a virtual world has both a 3D model that supports the visualisation of the object’s function in the world, and is a software object that can have agency to support autonomous behaviour. When each 3D object has agency, the virtual world is composed of design agents in the sense that the world is designed and each object in the world can sense the world and respond by adapting the designed world. Research in the field of design computation and cognition is increasingly agent based. This ranges from computational models of design processes to agent-based generative systems, see, for example, [Gero and Brazier, 2002]. The combination of design computation research and of agentoriented virtual worlds provides a theoretical basis for associating agency with the designed physical world [Cohen, 1999] and can lead to agents designing virtual worlds [Maher and Gu, 2002]. 3D virtual worlds are networked, multi-user environments that support communication and collaboration in a place-like context. The virtual worlds that we are considering as the baFigure 2: Inserting an object into an AW world. sis for our agent-based virtual world are object-oriented systems that associate a 3D model and a behaviour with each element of the world. Examples of such worlds include: Active Worlds and VirTools2. For the remainder of this paper, we will describe our agent-oriented world using an adaptation of the Active Worlds platform. New 3D objects are added to an AW world by copying an existing object, moving it as required, and editing a dialog box to configure it, as shown in Figure 2. The dialog box allows the world builder to specify a 3D model and a script that describes the behaviour of the object. The 3D models can be taken from a standard library provided by Active Worlds, or can be generated in a 3D modelling package and added to the library. The behaviour of an object is limited to the preprogrammed behaviours allowed by the scripting language. By making the world agent-based, a person designing the world is able to select or generate an agent model in a way that is similar to selecting or generating a 3D model for the object. Although the focus of this paper is on agent reasoning and agent memory, we intend that these agents communicate and collaborate. In this application, we define a multi-agent system to be an aggregation of agents that share some ontological connection, such as a room agent plus a set of wall agents that collectively comprise a virtual conference room. The multiagent system facilitates inter-agent communication and manages resources on behalf of the agents. This includes computational resources, such as a thread pool and a connection to a virtual world. This paper presents a model for a design agent reasoning process and a model for constructing a memory of the agent’s knowledge and interaction with the world and other agents. By establishing a common model for reasoning and agent memory, the world can be designed using a consistent representation and therefore expectation of what it means for the objects in the world to have agency. 2 Reasoning Process Model for Design Agents We have developed a reasoning process model for design agents that is independent of the type of design or envihttp://www.virtools.com sense−data, biases biases percepts, concepts,